ALLEGATO 1.

BOOK DEGLI ABSTRACT DEL 48� MEETINGSCIENTIFICO DELLA SOCIETA’ ITALIANA DI

STATISITCA(SEGUE IL BOOK)

8 – 10 GIUGNO 2016 – UNIVERSITA’ DI SALERNO

SIS 201648TH SCIENTIFIC MEETING OF THE ITALIAN STATISTICAL SOCIETY

Università degli Studi di SalernoJune 8th – June 10th, 2016

Scientific Programme Committee

Monica Pratesi (chair), Università di PisaAntonio Canale, Università di TorinoTonio Di Battista, Università G. d’Annunzio di Chieti-PescaraStefano Falorsi, IstatAlessio Farcomeni Sapienza, Università di RomaMichele Gallo, L’Orientale Università di NapoliPaolo Giudici, Università di PaviaAntonio Irpino, Seconda Università di NapoliGiovanna Jona Lasinio, Sapienza Università di RomaGianfranco Lovison, Università di PalermoAlessandra Luati, Università di BolognaMarzia Marcheselli, Università di SienaPaolo Mariani, Università di Milano-BicoccaCira Perna, Università di SalernoMariano Porcu, Università di CagliariFilomena Racioppi, Sapienza Università di RomaLaura Ventura, Università di PadovaDaniele Vignoli, Università di Firenze

Local Organizing Committee

Cira Perna (chair), Università di SalernoGiuseppina Albano, Università di SalernoPietro Coretto, Università di SalernoGiuseppe Giordano, Università di SalernoMarcella Niglio, Università di SalernoMaria Lucia Parrella, Università di SalernoLuigi Pieri, Tesoriere Società Italiana di StatisticaMarialuisa Restaino, Università di SalernoGiuseppe Storti, Università di SalernoMaria Prosperina Vitale, Università di Salerno

Dear Friends and Colleagues,

On behalf of the Local Organizing Committee and of the Scientific Programme Committee, wewelcome you to the 48th Scientific Meeting of the Italian Statistical Society (SIS2016). As manyof you know, this biennial conference has become a traditional national and international meetingfor connecting researchers in statistics, demography and applied statistics in Italy. The confer-ence aims at bringing together researchers and practitioners to discuss recent developments instatistical methods for economics, social sciences, and all fields of application of statistics. Thisbook of abstracts gives us the opportunity to perceive that many well-known Italian statisticiansof the national and international community and many young researchers belonging to academia,statistical agencies and other institutions are emerging to discuss current issues and innovationsin the field of statistics and related disciplines.They are presenting, at the 48th Scientific Meeting of the Italian Statistical Society, their notablecontributions, both from the methodological and empirical point of views. In this respect, theabstracts included in this volume provide a comprehensive overview of the current Italian sci-entific researches in theoretical and applied statistics. The Scientific Programme Committee hastried to provide a balanced and stimulating program that will appeal to the diverse interests of theparticipants. The Local Organizing Committee hopes that the conference venue will provide anappropriate environment to enhance your contacts and to establish new ones. The Scientific Pro-gramme Committee, the Session Organizers, the local hosting University and many volunteershave contributed substantially to the organization of the conference. We acknowledge their workand the support of our Society. Wishing you a productive, stimulating conference and a pleasantstay at the University of Salerno.

Monica Pratesi Fisciano, June 8th, 2016(Chair of the Scientific Programme Committee)

Cira Perna(Chair of the Local Organizing Committee)

i

Conference organization

Conference rooms location. All conference activities will take place at the University ofSalerno campus located in Fisciano (SA). Sessions are hosted in three buildings near “Piazza delSapere” which is the heart of the campus. Buildings are denoted by their campus code (appearingon their front). There will be a total of nine rooms labeled with the same color within the samebuilding as denoted below

Bulding Aula Magna Room: Aula Magna “V. Buonocore”

Bulding D1 (ground and 1st floor) Rooms: A, B, D and Sala Studio

Bulding D2 (1st floor) Rooms: “G. De Rosa”, “V. Foa”, SP/1 and SP/2

Registration Desk. The desk is located at the foyer of Aula Magna “V. Buonocore”. The regis-tration desk assists participants and it is the main source of information. Conference registrationwill take place every day starting from 8:30 a.m.

Organization of sessions

Plenary Sessions 5’ Chair, 30’ talk, 15’ discussant, 10’ floor discussionSpecialized Sessions [SPE] 5’ Chair, 20’ talk, 25’ discussant and floor discussionSolicited Sessions [SOL] 20’ talk, 15’ discussionContributed Paper Sessions [CON] 15’ talk, 15’ discussionSpeed Poster Session [POS] 3’ poster presentation

Instructions for presentations and posters. The lecture rooms are equipped with a PC, remotepointers, and a computer projector. Both Authors and Chairs should reach the room at least tenminutes before the scheduled time to set up files on the computers and to check the equipment.Possible file formats for presentation are PDF (Acrobat) and PPT (Powerpoint). Contributionsare presented in the order they are listed in the programme.The poster session will take place at the foyer of Aula Magna. The authors have to be presentduring the time assigned for the session. Authors are responsible for placing the posters on paneldisplays and for removing them. The maximum size of the poster is A0 portrait (841×1189mm).

SIS2016 App and network connections. Conference scheduling is aided by a dedicated Appfor both Android and iOS systems. The SIS2016 App can be downloaded from the conferencewebsite and serves as an interactive guide with real time features.The campus is served by the eduroam roaming access service. There is also a dedicatedcampus WiFi network for conference participants. Instructions to connect to the WiFi are givenat the main desk.

Coffee breaks and lunches. Coffee breaks and buffet lunches will take place at the Sala Buffetof the Aula Magna building. Participants must have their conference badge in order to attend thecoffee breaks. Lunches are not included in the conference fees, and tickets are available at theregistration desk.

Transportation services. There is a dedicated Bus service operating from Salerno to Fiscianoin the early morning, and from Fisciano to Salerno after the last session every day. There isalso a public transportation service. Schedule and relevant information are given through theconference website and the SIS2016 App.

ii

Campus map

Car

Par

kin

g

Te

rmin

al B

us

Au

la M

agn

aV

. Bu

on

oco

re

Pia

zza

Gab

rie

le D

e R

osa

Ch

iost

ro

de

lla P

ace

Me

nsa

G. D

e R

osa

V. F

oa

SP

/1, S

P/2

Co

nfe

ren

ce R

oo

ms

A, B

, D, S

ala

Stu

dio

Bu

ildin

g D

2

Bu

ildin

g D

1

Pia

zza

de

lS

ape

re

Co

nfe

ren

ce R

oo

ms

iii

Conference programme

 June 8th, 2016 08.30 Registration opening (Foyer Aula Magna)09.30 Conference opening (Room: Aula Magna)10.00 Plenary A

E. Baldacci. Assessing Economic Crises and their Impacts: Data Gaps and Innovation in Statistical Production (Room: Aula Magna)

11.00 Coffee break11.30 Parallel Specialized Sessions

SPE­01 Inference, sampling and survey design (Room: V. Foa) SPE­02 Multivariate models for risk assessment (Room: SP/2)SPE­03 Bayesian nonparametrics (Room: SP/1)

SPE­04 Statistical methods for the analysis of gene­environment interaction in thestudy of complex pathologies  (Room: G. De Rosa) 

13.00 LunchY­SIS Young­Statisticians lunch seminar (Room: V. Foa) 

14.15 Parallel Solicited Sessions

SOL­01 Subjective wellbeing and demographic events over the life course (Room: V. Foa)

SOL­02 Statistics for equitable and sustainable development (Room: SP/1)SOL­03 New approaches to treat undercoverage and nonresponse (Room: SP/2)SOL­04 Statistical models and methods for network data (Room: G. De Rosa) SOL­05 Recent developments in computational statistics (Room: A)

SOL­06Statisticians meet naturalists: issues on ecological and environmental statistics (Room: B) 

SPE­05 Nonlinear time series (Room: Aula Magna)15.45 Coffee break16.15 Parallel Contributed Sessions

CON­01 Bayesian statistics (1) (Room: A)CON­02 Statistical modeling (Room: B)CON­03 Demographics and social statistics (1) (Room: D) CON­04 Environmental statistics (Room: Sala Studio)CON­05 Health statistics (Room: G. De Rosa)CON­06 Labor market statistics (Room: V. Foa)CON­07 Robust statistics (Room: SP/1) CON­08 Sampling methods (Room :SP/2)

17.45 Parallel Solicited Sessions

SOL­07From survey data to new data sources and big data in official statistics (Room: G. De Rosa) 

SOL­08 Symbolic data analysis methods and applications (Room: V. Foa) SOL­09 Compositional analysis (Room: SP/1) 

SOL­10 Sustainable development: theory, measures and applications (Room: SP/2) 

19.15 Welcome Aperitif

June 9th, 201609.00 Parallel Solicited Sessions

SOL­11 Detecting heterogeneity in ordinal data surveys (Room: G. De Rosa) 

SOL­12 Active ageing: age management and lifelong learning strategies (Room: V. Foa) 

SOL­13 Statistical models for evaluating policy impact (Room: SP/1)SOL­14 Usage of geocoded micro data in the economic analysis (Room: SP/2)SOL­15 Statistical models in functional data analysis (Room: A) 

10.30 Plenary B D.B. Dunson. Probabilistic inference for big & complex data (Room: Aula Magna)

11.30 Coffee break

iv

12.00 POS Poster speed session (Room: Aula Magna)13.30 Lunch

POS Poster exhibition (Foyer Aula Magna)

14.30 SUS2 Stats Under the Stars 2 – Winners presentation (Room: Aula Magna) 14.45 Parallel Specialized Sessions 

SPE­06 Spatial analyses in demography (Room: V. Foa)  SPE­07 Recent developments in volatility modeling (Room: G. De Rosa) SPE­08 Advances in ordinal contingency table analysis (Room: SP/1)  SPE­09 Statistical models for directional and circular data (Room: SP/2) SPE­10 The interplay between frequentist and bayesian inference (Room: A) SPE­11 Société Française de Statistique (Room: B) 

16.15 Coffee break16.45 SIS SIS President’s lecture (Room: Aula Magna)17.45 SIS SIS members assembly (Room: Aula Magna) 20.00 Social dinner

June 10th, 201609.00 Parallel Solicited Sessions

SOL­16 Forecasting economic and financial time series (Room: G. De Rosa)SOL­17 Immigrations and integration in Italy (Room: V. Foa)

SOL­18Open data, linked data and big data in public administration and official statistics (Room: SP/1)

SOL­19 Evaluation of prognostic biomarkers (Room: SP/2)SOL­20 Models for studying the mobility of students (Room: A) 

10.30 Plenary CSalvatore Strozza. Foreign immigration in Italy: a forty year old history (Room: Aula Magna) 

11.30 Coffee break12.00 Parallel Specialized Sessions

SPE­12 National accounts (Room: SP/2)

SPE­13Statistical tools for monitoring the educational system and assessing students’ performances (Room: G. De Rosa)  

SPE­14 Robust inference by bounded estimating functions (Room: SP/1) 13.30 Lunch14.30 RT

Round table: Viste e .....riviste. Classificazione e ranking (Room: Aula Magna)

16.00 Coffee break16.15 Parallel Contributed Sessions

CON­09 Economic data analysis (Room: G. De Rosa)CON­10 Quantile methods (Room: V. Foa) CON­11 Statistical algorithms (Room: SP/1) CON­12 Statistics for medicine (Room: SP/2)  CON­13 Statistics for the education system (Room: D) CON­14 Testing procedures (Room: B)  CON­15 Time series analysis (Room: A)  CON­16 Forecasting methods (Room: Sala Studio)

17.30 Parallel Contributed SessionsCON­17 Bayesian statistics (2) (Room: G. De Rosa) CON­18 Business statistics (Room: V. Foa) CON­19 Clustering and classification (Room: SP/1)CON­20 Demographics and social statistics (2) (Room: D)CON­21 Statistical inference (Room: A)CON­22 Survey methods (Room: B) 

18.45 Closing ceremony (Room: Aula Magna) 

v

Social events

Welcome Aperitif, Wednesday, June 8th, 2016 at 19:15. The Welcome Aperitif will takeplace in “Piazza del Sapere” (University Campus) right after the end of the last session of thefirst conference day. A selection of sparkling wines from the DUBL collection of Feudi di SanGregorio (www.feudi.it) will be served. Conference registrants and accompanying persons mustbring their conference badge and ticket in order to attend the reception. Preregistration is requireddue to safety and organizational reasons.

Social Dinner, Thursday, June 9th, 2016 at 20:00. Social Dinner will take place at the Lloyd’sBaia Hotel (Vietri sul Mare). The social dinner is not included in the conference fees, and ticketscan be obtained from the registration desk. Conference registrants and accompanying personsshould bring their tickets in order to attend the event. Buses will leave from the UniversityCampus (Fisciano) and from Salerno, the timetable will be available on the conference websiteand through the SIS2016 App.

Satellite events

International Workshop on “Perspectives on Financial and Actuarial Modelling”, June 7th,Università di Salerno, Room: “G. De Rosa”. The workshop aims at promoting the scien-tific debate on the application of advanced mathematical and statistical models to solve rele-vant financial and actuarial problems. The approach is interdisciplinary since we encouragecontributions and attendance from researchers and practitioners involved in different fields in-cluding actuarial sciences, statistics, data science, finance, financial econometrics. The pro-gramme will include 8 keynote invited talks as well as a contributed poster session. Web-site: www.labeconomia.unisa.it/fam

GRASPA tutorial, June 7th, Università di Salerno, Room: “V. Foa”. The tutorial on Linear-Circular environmental data modelling provides an extensive overview of statistical methods tojointly model linear and circular variables under a unique umbrella. Circular data refer to datarecorded as points for which directions are measured, typically in the fields of biology, geogra-phy, medicine, and astronomy. Analyzing circular data is challenging because traditional statis-tics are not meaningful, and may even be misleading when the particular definition of the circulardomain is ignored. The tutorial faces theoretical and applied sessions aiming at introducing themost used distributions and models able to account for specific linear-circular data features.

H2DM’16: “International Workshop on High Dimensional Data Mining”, June 7th, Naples.In recent years, in a wide range of scientific and technological domains, huge amount of mul-tidimensional and heterogeneous data are generated. This represents a great challenge in manycross-disciplinary fields. Traditional statistical methods are unable to analyse data with highdimensionality and heterogeneity. New data mining methods, such as multivariate data analy-sis, graphical models and data visualization tools need. The workshop represents an excellentvenue for discussing recent development of algorithms, methods, applications and software formining high dimensional data. The Workshop is supported by the Italian Statistical Society, theLaboratory for Genomics, Transcriptomics and Proteomics of Italian National Reseach Council,and the "Data Mining et Apprentissage" group of the Société Française de Statistique. Web-site: http://h2dm2016.sciencesconf.org

vi

Stats Under the Stars2, June 7-8th, Lloyd’s Baia Hotel, Vietri sul Mare (SA). Stats under theStars2 is an all night-long event aimed at fostering the link between Statistics and young datascientists. It represents the second edition of a statistical competition in which teams of youngpeople are asked to analyze a dataset and to answer hot business questions. This clash representsnot only an opportunity to meet and exchange ideas, but also a way to highlight the usefulness ofStatistics as a key decision making tool. Website: www.labeconomia.unisa.it/sus2

Summer School of the Italian Statistical Society:“Statistical Methods for Social NetworkAnalysis and Community Detection”, June 11-15th, Conference Centre “Mario Cacace”,Anacapri (NA). The School is addressed to young researchers, graduate students and institu-tions involved in the study and analysis of social networks. The courses are divided into twomodules. The first one relates to statistical methods for description, representation and mod-eling of relational data. The second module relates to methods for the identification of highlyconnected subgraphs (community) in large networks. Website: www.scuolaretisis.unisa.it

vii

Table of contents

Welcome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iConference organization . . . . . . . . . . . . . . . . . . . . . . . . . . iiCampus map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiConference programme . . . . . . . . . . . . . . . . . . . . . . . . . . ivSocial events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viSatellite events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

Day 1: Wednesday, June 8th, 2016 1

10:00-11:00 – Plenary Session A: Emanuele Baldacci 1

11:30-13:00 – Parallel Specialized Sessions 1SPE-01: Inference, sampling and survey design . . . . . . . . . . . . . 1SPE-02: Multivariate models for risk assessment . . . . . . . . . . . . . 2SPE-03: Bayesian nonparametrics . . . . . . . . . . . . . . . . . . . . . 3SPE-04: Statistical models for the analysis of gene-environment in-

teraction in the study of complex pathologies . . . . . . . . . 4

14:15-15:45 – Parallel Solicited and Specialized Sessions 6SOL-01: Subjective wellbeing and demographic events over the life

course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6SOL-02: Statistics for equitable and sustainable development . . . . 7SOL-03: New approaches to treat undercoverage and nonresponse 8SOL-04: Statistical models and methods for network data . . . . . . . 10SOL-05: Recent developments in computational statistics . . . . . . 12SOL-06: Statisticians meet naturalists: issues on ecological and en-

viromental statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 13SPE-05: Nonlinear time series . . . . . . . . . . . . . . . . . . . . . . . . 15

16:15-17:45 – Parallel Contributed Sessions 16CON-01: Bayesian statistics (1) . . . . . . . . . . . . . . . . . . . . . . . 16CON-02: Statistical modeling . . . . . . . . . . . . . . . . . . . . . . . . 17CON-03: Demographics and social statistics (1) . . . . . . . . . . . . 18CON-04: Environmental statistics . . . . . . . . . . . . . . . . . . . . . . 20CON-05: Health statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 21CON-06: Labor market statistics . . . . . . . . . . . . . . . . . . . . . . 23CON-07: Robust statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 24CON-08: Sampling methods . . . . . . . . . . . . . . . . . . . . . . . . 25

17:45-19:15 – Parallel Solicited Sessions 27

viii

SOL-07: From survey data to new data sources and big data in offi-cial statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

SOL-08: Symbolic data analysis methods and applications . . . . . . 29SOL-09: Compositional analysis . . . . . . . . . . . . . . . . . . . . . . . 31SOL-10: Sustainable development: theory, measures and applica-

tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Day 2: Thursday, June 9th, 2016 34

09:00-10:30 – Parallel Solicited Sessions 34SOL-11: Detecting heterogeneity in ordinal data surveys . . . . . . . 34SOL-12: Active ageing: age management and lifelong learning

strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35SOL-13: Statistical models for evaluating policy impact . . . . . . . . 36SOL-14: Usage of geocoded micro data in the economic analysis . 38SOL-15: Statistical models in functional data analysis . . . . . . . . . 39

10:30-11:30 – Plenary Session B: David Dunson 41

12:00-13:00 – Poster Speed Session 41

14:45-16:15 – Parallel Specialized Sessions 54SPE-06: Spatial analyses in demography . . . . . . . . . . . . . . . . . 54SPE-07: Recent developments in volatility modeling . . . . . . . . . . 55SPE-08: Advances in ordinal contingency table analysis . . . . . . . . 56SPE-09: Statistical models for directional and circular data . . . . . . 57SPE-10: The interplay between frequentist and Bayesian inference . 58SPE-11: Société Française de Statistique . . . . . . . . . . . . . . . . . 60

16:45-17:30 – SIS President’s Lecture 62

Day 3: Friday, June 10th, 2016 62

09:00-10:30 – Parallel Solicited Sessions 62SOL-16: Forecasting economic and financial time series . . . . . . . 62SOL-17: Immigrations and integration in Italy . . . . . . . . . . . . . . 63SOL-18: Open data, linked data and big data in public administra-

tion and official statistics . . . . . . . . . . . . . . . . . . . . . . 64SOL-19: Evaluation of prognostic biomarkers . . . . . . . . . . . . . . 66SOL-20: Models for studying the mobility of students . . . . . . . . . . 67

10:30-11:30 – Plenary Session C: Salvatore Strozza 69

12:00-13:30 – Parallel Specialized Sessions 69SPE-12: National accounts . . . . . . . . . . . . . . . . . . . . . . . . . 69SPE-13: Statistical tools for monitoring the educational system and

assessing students’ performances . . . . . . . . . . . . . . . . . 70SPE-14: Robust inference by bounded estimating functions . . . . . . 71

16:15-17:30 – Parallel Contributed Sessions 71CON-09: Economic data analysis . . . . . . . . . . . . . . . . . . . . . 71

ix

CON-10: Quantile methods . . . . . . . . . . . . . . . . . . . . . . . . . 73CON-11: Statistical algorithms . . . . . . . . . . . . . . . . . . . . . . . 74CON-12: Statistics for medicine . . . . . . . . . . . . . . . . . . . . . . . 75CON-13: Statistics for the education system . . . . . . . . . . . . . . . 76CON-14: Testing procedures . . . . . . . . . . . . . . . . . . . . . . . . 77CON-15: Time series analysis . . . . . . . . . . . . . . . . . . . . . . . . 79CON-16: Forecasting methods . . . . . . . . . . . . . . . . . . . . . . . 80

17:30-18:45 – Parallel Contributed Sessions 81CON-17: Bayesian statistics (2) . . . . . . . . . . . . . . . . . . . . . . . 81CON-18: Business statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 82CON-19: Clustering and classification . . . . . . . . . . . . . . . . . . . 83CON-20: Demographics and social statistics (2) . . . . . . . . . . . . 84CON-21: Statistical inference . . . . . . . . . . . . . . . . . . . . . . . . 85CON-22: Survey methods . . . . . . . . . . . . . . . . . . . . . . . . . . 86

Authors Index 89

x

Plenary Session A • Wednesday 8, 10:00-11:00 • Room: Aula Magna

FINANCIAL CRISES AND THEIR IMPACTS: DATA GAPS AND INNOVATION INSTATISTICAL PRODUCTION

by Emanuele BaldacciEurostat

Chair: Paolo Giudici (Università di Pavia)Discussant: Alessandro Viviani (Università di Firenze)

Abstract: Financial crises damage output and social cohesion. Lack of timely and accuratedata makes it more difficult to assess risks’ build up. Information gaps can alsolimit the ability to respond to crises. This calls for better data to monitor economicand financial risks. Several measures taken by the international official statisticscommunity address information needs. These include efforts to fill the data gaps,ensure policy relevance of key indicators, and measure the "unmeasured" complexdimensions of economy and society. Harnessing new data sources and promotinginnovation in statistical production processes are key to improving timeliness andadequacy of statistical information services. Nowcasting and predictive analyticscan enhance the provision of early warnings about crises.

Specialized Session SPE-01 • Wednesday 8, 11:30-13:00 • Room: V. Foa

INFERENCE, SAMPLING AND SURVEY DESIGNChair: Maria Giovanna Ranalli; Discussant: Elisabetta Carfagna

Title: Resampling from finite populations under complex designs: the pseudo-populationapproach

Presenter: Pier Luigi Conti, Sapienza, Università di RomaCo-author(s): Federico Andreis, Daniela Marella, Fulvia MecattiAbstract: In this paper, resampling techniques based on pseudo-populations in the pres-

ence of a general pps sampling design are studied. Different forms of calibratedpseudo-populations are introduced and discussed. The influence of calibration onthe performance of resampling is studied via simulation.

Title: A joint use of model based and design based frameworks for defining optimalsampling designs

Presenter: Paolo Righi, IstatCo-author(s): Piero Demetrio FalorsiAbstract: Official statistics commonly produce a large number of estimates relating to both

different parameters of interest and highly detailed estimation domains or sub-populations. The parameter estimation is primarily performed defining a samplingstrategy that is the couple sampling design and estimator. The sampling design isa fundamental step for defining an efficient strategy. Assume the domain indicatorvariables are available for each sampling unit in the sampling frame, the surveysampling designer could attempt to select a sample in such a way that sampling

1

errors, at the domain level, would be controlled. To do that the researcher has tomake assumptions respectively on the values of the characteristic to be studied andthe level of uncertainty of these values since they are unobserved before the sur-vey. Statistical superpopulation models are the tool for building the assumptions.Next, methods defining the optimal sample design or optimal sample allocationshould take into account these assumptions in their objective functions. The firstaim of the paper is to highlight statistical models are implicitly considered in themethods such us Neyman allocation. If the sampling designer is awakened thatimplicitly or explicitly he will use a model for planning the sample, then he cantry to choice the best model, i.e. the model reducing the model uncertainty asmuch as possible. Falorsi and Righi (2015) propose a method, where a superpop-ulation model is explicitly formalized. The second aim of the paper is to show thechoice of a superpopulation model influence the choice of the sampling design(selection scheme and inclusion probabilities). The allocation method proposedby the authors is suitable to different superpopulation models, selection schemes,domains of estimate and it offers, de facto, a generalized framework for dealingwith the sample allocation.

Title: A unified approach for robustness in survey samplingPresenter: Anne Ruiz-Gazen, Toulouse School of Economics, University Toulouse CapitoleCo-author(s): Jean-François Beaumont, David HazizaAbstract: The classical tools of robust statistics have to be adapted to the finite population

context and several robust estimation methods exist. Recently, a unified approachfor robust estimation in surveys has been introduced. It is based on an influencemeasure called the conditional bias that allows to take into account the partic-ular finite population framework and the sampling design In the present paper,we focus on the design-based approach and we recall the main properties of theconditional bias and how it can be used for robust estimation of a total. The imple-mentation in R is also presented with some functions that etimate the conditionalbias and calculate the proposed robust estimators for particular designs. Somerecent advances using the conditional bias approach are summarized at the end.

Specialized Session SPE-02 • Wednesday 8, 11:30-13:00 • Room: SP/2

MULTIVARIATE MODELS FOR RISK ASSESSMENTChair: Marina Borgi; Discussant: Giampiero Gallo

Title: A Bayesian nonparametric approach to macroeconomic riskPresenter: Monica Billio, Università Ca’ Foscari di VeneziaCo-author(s): Roberto Casarin, Michele Costola, Michele GuindaniAbstract: We apply a Bayesian non parametric test for distributional changes to panel data

from a macroeconomics and finance for detecting structural changes in the se-quence of cross-sectional conditional distributions.

Title: Bank risk contagion:an analysis through big dataPresenter: Paola Cerchiello, Università di PaviaCo-author(s): Paolo Giudici, Giancarlo NicolaAbstract: A very important and timely area of research in finance is systemic risk modelling,

2

which concerns the estimation of the relationships between different financial in-stitutions, with the aim of establishing which of them are more contagious/subjectto contagion. The aim of this paper is to develop a systemic risk model which,differently from existing ones, employs not only the information contained infinancial market prices, but also big data coming from financial tweets. Froma methodological viewpoint, we propose a new framework, based on graphicalGaussian models, that can estimate systemic risks with stochastic network mod-els based on two different sources: financial markets and financial tweets, andsuggest a way to combine them, using a Bayesian approach. From an appliedviewpoint, we present a systemic risk model based on big data, and show thatsuch a model can help predicting the default probability of a bank, conditionallyon the others.

Title: A Markov-switching regression model with non-Gaussian innovations for sys-temic risk measurement

Presenter: Luca De Angelis, Università di BolognaCo-author(s): Cinzia ViroliAbstract: In this paper we propose an approach to obtain more reliable measures ofsystemic

risk. In particular, we propose a multivariate Markov-switching regression (MSR)model considering the Normal Inverse Gaussian (NIG) distribution as condition-alform of financial returns and model innovations. It is indeed well-known that theGaussian distribution is not able to capture many stylized facts of the return seriessuch as skewness, excess of kurtosis and heavy tails. Through a large simulation-study, we show that a NIG-based MSR model allows to adequately account forboth skewness and fat tails in the data and, according to model selection criteria,is even preferred over other popular distributional assumptions such as Student-tand GED. We develop an EM algorithm which allows the estimation of the modelparameters in closed form and the derivation of the scores of the model estimatorsthat allows dynamic specification tests for autocorrelation and for the first-orderMarkov assumption.

Specialized Session SPE-03 • Wednesday 8, 11:30-13:00 • Room: SP/1

BAYESIAN NONPARAMETRICSChair: Bernando Nipoli; Discussant: TBA

Title: Bayesian Nonparametric Modeling of Dynamic International RelationsPresenter: Daniele Durante, Università di PadovaCo-author(s): David B. DunsonAbstract: Real world networks are often associated with a dynamic component and the de-

velopment of flexible statistical methodologies to learn how the connectivity pat-terns are wired across time is a fundamental goal in many fields of application.News media reports currently provide a rich variety of complex dynamic networksalong with novel applied questions. Our focus is on modeling dynamic networksof international relationships among countries during recent conflicts and crises.We accomplish this goal by leveraging a recent Bayesian nonparametric modelwhich merges latent space representations of network data with Gaussian pro-cesses

3

Title: Bayesian autoregressive semiparametric models for gap times of recurrent eventsPresenter: Alessandra Guglielmi, Politecnico di MilanoCo-author(s): Giorgio Paulon, Maria De IorioAbstract: We propose an autoregressive Bayesian semi-parametric model for the waiting

times between recurrent events with covariates. Time-dependency is taken intoaccount through an autoregressive model, but the parameters are a sample from aDirichlet process, thus inducing clustering of the individuals in the sample. Co-variates may be included in this framework. The model proposed is within theclass of Dirichlet process mixtures. We build a Gibbs sampler to perform fullposterior inference from this model; the code has been implemented in the Julialanguage. We illustrate the model through an application to recurrent hospitaliza-tions of cancer patients.

Title: Restricted Nonparametric Mixtures models for Disease ClusteringPresenter: Abel Rodriguez, University of California, Santa CruzCo-author(s): Tatiana XifaraAbstract: Identifying disease clusters (areas with an unusually high incidence of a particular

disease) is a common problem in epidemiology and public health. We describea Bayesian nonparametric mixture model for disease clustering that con- strainsclusters to be made of contiguous areal units. This is achieved by modifying theexchangeable partition probability function associated with the Ewen’s samplingdistribution. The model is illustrated using data on US lung cancer rates.

Specialized Session SPE-04 • Wednesday 8, 11:30-13:00 • Room: G. De Rosa

STATISTICAL MODELS FOR THE ANALYSIS OF GENE-ENVIRONMENT INTERACTION INTHE STUDY OF COMPLEX PATHOLOGIES

Chair: Clelia Di Serio; Discussant: Adriano Decarli

Title: An introduction to next generation sequencing for studying omic-environmentinteractions

Presenter: Claudia Angelini, Istituto per le Applicazioni del Calcolo “Mauro Picone”Abstract: Next generation sequencing (NGS) is a modern technology that allows collect-

ing high throughput multi-omic data at a genome-wide resolution in an easy andcheap way [1-3]. Such data are now offering the possibility to investigate genomicalterations, transcriptional signatures, epigenetic profiles of thousands of individ-uals, tissues and cell types, thus providing life scientists with novel approachesfor better understanding complex disease etiologies [4]. On the other hand, inthe last couple of decades genetic epidemiology studies have moved from in-vestigating single variants in few candidate genes to simultaneously consideringmillions of single nucleotide polymorphisms (SNPs) in genome-wide associationstudies (GWAS). Despite the fact that such studies allowed identifying hundredsof genetic markers associated with common diseases, they accounted only for arelatively small proportion of diseases heterogeneity. Such a limit is due to themultifactorial nature of complex diseases, where several variables contribute andinteract to determine the observed phenotype. To overcome this problem, it isnecessary to consider and better modeling both gene-gene and gene-environmentinteractions [5,6]. Statistical methods for estimating both main and interaction ef-

4

fects depend on the experimental design, the type and amount of variables one isconsidering [5,6,7]. Gene interaction terms allowed understanding how differentindividual might have different prognosis, or might respond in different mannerto a certain treatment, and how the environment can determine or modulate theindividual response. Nevertheless, to date a significant part of individual diseasesusceptibility is still left unexplained due to both the huge number of variablesinvolved and their complex relationships. To fill this gap, researchers have todevelop novel statistical methods to better study gene-environment interactionsand, also, to integrate -in a unified framework- different layers of biological in-formation available from multi omic data. In fact, most of the results obtained sofar were based on the identification of variants or SNPs that are observed at thegenomic layer of an individual. However, in the last few years, other omics aredemonstrating their role in determining and modulating the phenotype but are stillnot used on large population based studies. Among them epigenomics [8,9] is as-suming an increasingly important role in evaluating the response to environmentalstress, and in the development and life span of an individual and his progenies.In this talk we provide an overview of the methods available to process NGS dataand discuss how multi-omic data can be used to improve our way of studyingcomplex diseases from a novel perspective.

Title: Statistical approaches for the evaluation of genetic associations in complex dis-eases: the heterogeneity of asthma phenotypes

Presenter: Lucia Calciano, Università degli Studi di VeronaCo-author(s): Laura Portas, Simone AccordiniAbstract: Asthma is a complex disease, which is influenced by the interaction among the

genetic background, environmental exposures and life-style factors. The associ-ation between genetic factors and health outcomes is usually evaluated by usingclassic statistical methods that test the effect of a single genetic variant at a time,by controlling for the multiple hypotheses testing. For almost all complex traits,this strategy is suboptimal because the identified genetic effects could only ac-count for a minor portion of the estimated trait heritability. Moreover, continuousoutcomes should be used to catch the heterogeneity of asthma phenotypes thatmay result from different risk factors.

Title: Improved case-only approach to study genome-wide gene-environment interac-tion

Presenter: Yadav Pankaj, University of Kiel, GermanyAbstract: Case-only (CO) design has been proposed as a valid approach with increased sta-

tistical efficiency over case-control and cohort studies in detecting gene-environmentinteractions (G×E). In the past, CO studies of G×E usually followed a candidate(or single-) gene approach, but their utility on genome-wide level remained unex-plored. There are two important issues that might potentially affect the genome-wide CO studies of G×E. First, the role of linkage disequilibrium in CO studyof G×E was not addressed in the past. Second, hidden stratification in the studypopulation can severely compromise a CO study. The effect of population stratifi-cation on a CO study is partially explained, but a systematic analysis was lackingon consequences of this phenomenon. This work therefore systematically as-sessed through simulations the role of linkage disequilibrium and the impact ofpopulation stratification in CO studies of G×E. Simulations show that, single nu-cleotide polymorphisms (SNPs) in linkage disequilibrium with a truly interacting

5

SNP can be used as proxies to indirectly infer the respective G×E signal. Further,simulations reveal that when study sample is divided by both genetic and exposurefactors, a CO study provides an inflated type I error rate. Moreover, it is shownthat genomic control-based methods can be successfully employed to correct forpopulation stratification in CO studies.

Solicited Session SOL-01 • Wednesday 8, 14:15-15:45 • Room: V. Foa

SUBJECTIVE WELLBEING AND DEMOGRAPHIC EVENTS OVER THE LIFE COURSEChair: Bruno Arpino

Title: Cultural and institutional drivers of basic psychological needs satisfactionPresenter: Giulia Fuochi, Università di PadovaCo-author(s): Pierluigi Conzo, Arnstein Aassve, Letizia MencariniAbstract: This paper provides a framework for why institutions and cultural values affect the

happiness of nations. In doing so, we lean on what is known in the psychologyliterature as self-determination theory (SDT) which predicts that satisfaction ofthree basic psychological needs (autonomy, relatedness and competence) drivesubjective well-being. By exploiting European country level data, we assess if andto what extent institutional quality and cultural values influence the Europeans’satisfaction of these needs, controlling for GDP and education. While we findmixed results for institutions we find a robust and positive impact of the culturaltrait “generalized morality” (i.e. high trust and respect and low obedience). Thisfeature has a strong impact on all three needs satisfaction indicators.

Title: Five reasons to be happy about childbearingPresenter: Letizia Mencarini, Università di Milano BocconiCo-author(s): Arnstein Aassve, Francesca LuppiAbstract: Differently to previous studies analysing the relationship between overall subjec-

tive wellbeing and childbearing, we consider here explicitly the different domainsof subjective wellbeing. Using data from the Household Income and Labour Dy-namics in Australia panel survey and fixed effect estimations, we show that indeedthe domains of subjective wellbeing have various patterns when held up againstchildbearing events. We find indeed an anticipation effect, where life satisfactionpeaks around the childbearing event, for then to decline. From our analysis, twodomains stand out: satisfaction with the partner and satisfaction with the finan-cial situation. Other domains show a rather brokered pattern, which show littleconsistency with the standard shape observed for the overall life satisfaction.

Title: Migration motivations and migrants’ satisfaction in the life course: A sequenceanalysis of geographical mobility trajectories in the United Kingdom

Presenter: Beata Nowok, Administrative Data Research Centre - Scotland, University ofEdinburgh

Abstract: An individual’s residence history interrelates closely with other lifetime trajecto-ries such as family or employment careers. These dynamic and interacting pro-cesses produce migration patterns that differ between individuals. In this paperwe apply sequence analysis methods to investigate existence of distinctive long-term patterns of migration drivers among individuals changing place of residence

6

within the United Kingdom. Based upon migrants’ reasons for moving we linkgeographical mobility to various life domains, like job, housing or personal life.We use 18 waves of the British Household Panel Survey and treat sequences ofall annual observations for an individual as a conceptual unit. We cluster thosetrajectories using pairwise dissimilarities derived applying optimal matching tech-nique. We indentify distinctive groups of migrants and compare their satisfactionwith life overall and with housing. Since young adults are most mobile and canhave very fluid residence histories we perform a separate analysis of them.

Title: Does becoming a parent change the meaning of happiness and life satisfaction?Evidence from the European Social Survey

Presenter: Andre Pirralha, Research and Expertise Centre for Survey Methodology - Uni-versitat Pompeu Fabra

Co-author(s): Henrik DobewallAbstract: This study aims to determine whether subjective well-being has a different psy-

chological meaning for parents and childless respondents. We analyzed compar-ative data of the third round of the European Social Survey (ESS; N=10,913).After case-control matching, in line with previous work, we found that in someEastern European countries and Ireland non-mothers report higher SWB levelsthan mothers. Mothers from the other countries and fathers reported higher SWBlevels than non-mothers from these countries and non-fathers. However, resort-ing to structural equation modelling, we tested for measurement invariance acrossgroups and no indication of differences between parents and childless respondentswas found. Having established full scalar measurement invariance, it was deter-mined that measurement non-invariance is not responsible for the low SWB levelsof mothers in some European countries.

Solicited Session SOL-02 • Wednesday 8, 14:15-15:45 • Room: SP/1

STATISTICS FOR EQUITABLE AND SUSTAINABLE DEVELOPMENTChair: Matteo Mazziotta

Title: Wellbeing and sustainable development: a multi-indicator approach to evaluateurban waste management systems

Presenter: Enrico di Bella, Università di GenovaCo-author(s): Cavalletti Barbara, Matteo CorsiAbstract: This work refers to a vast field of scientific literature discussing the evaluation of

environmental sustainability of urban waste management systems (WMS). Exist-ing literature describes performance of WMS as a multi-indicator problem eventhough actual evaluation is frequently unidimensional. In our work, we first pro-vide further evidence that this is the correct approach with the help of data from32 Municipalities in the Liguria Region. Furthermore, we argue that the staple ap-proach to multi-indicator systems (composite indicators) is not going to performwell in the absence of some (positive) correlation between different performanceindicators. Looking at our case study, we illustrate the merits of an alternativeapproach to performance evaluation based on partial order theory.

7

Title: Small Area Estimation for Local Welfare Indicators in ItalyPresenter: Caterina Giusti, Università di PisaCo-author(s): Stefano Marchetti, Luca Faustini, Linda PorcianiAbstract: Considering the local areas where citizens live is fundamental to investi- gate

the impact of the increasing financial difficulties and of the reduction of publicfunding on several crucial needs that may give riser to social protection, such ashealth care, old-age and unemployment. In this work we analyze some key indica-tors of social protection services from the perspective of the beneficiaries, theItalian households. We used the EU-SILC 2013 data together with data comingfrom ad- ministrative archives to produce small area estimates of social protectionindicators at provincial level in Italy (LAU 1), such as the households’ Head CountRatio before and after the social transfers, and the proportion and the mean ofspecific categories of benefits, such as family/children and old-age benefits.

Title: Does socio-economic variables influence the Italians’ adherence towards a sus-tainable diet?

Presenter: Tiziana Laureti, Università di TusciaCo-author(s): Luca SecondiAbstract: At first sight, the Mediterranean diet appears to be the best and most well-balanced

diet to follow as it links environmental and human health. In this paper, by usingrepeated cross-sections of the ISTAT “Aspects of daily-life” survey over the period1997-2012, we assess the adherence to the Mediterranean dietary pattern in theItalian population and explore socio-demographic and lifestyle factors that mightinfluence adherence to this sustainable diet by means of a dynamic pseudo-panelmodel. We found that not smoking is related with healthier tendency towards foodas well as people who regularly practice sport have a higher level of adherence toMD. Education and wealth proved to be interrelated in influencing Italians’ foodconsumption habits.

Title: Sustainability of wellbeing: an analysis of resilience and vulnerability throughsubjective indicators

Presenter: Fabiola Riccardini, IstatCo-author(s): Maria Bachelet, Filomena MagginoAbstract: Many scholars have focused on how the concepts of vulnerability and resilience

may be employed in the analysis of sustainability. Different approaches have beenproposed, concerning different fields of application (from environmental to finan-cial settings). While much of the existing literature on vulnerability and resilienceis sector-specific, we propose a more holistic approach that allows the sustainabil-ity of human well-being to be analysed as a whole. In particular we apply thoseconcepts to BES (Benessere, Equo e Sostenibile) framework, where 12 domainsare included. Moreover, while the majority of studies consider the vulnerabilityand resilience as aspects of the sustainability of a “system” (a society, a country,or even the whole planet), we focus on an individual dimension of well-being,considering the exposure to risk and the ability to recover of the single persons.We show how the proposed scheme lends itself to the use of both objective andsubjective indicators of well-being. Finally, using Italian data provided by tha Na-tional Institute of Statistics (ISTAT), we propose an example of how to analysesthe subjective aspects of wellbeing in terms of sustainability.

8

Solicited Session SOL-03 • Wednesday 8, 14:15-15:45 • Room: SP/2

NEW APPROACHES TO TREAT UNDERCOVERAGE AND NONRESPONSEChair: Fulvia Mecatti

Title: Methodological perspectives for surveying rare and clustered population: to-wards a sequentially adaptive approach

Presenter: Federico Andreis, Università di Milano BocconiAbstract: Sampling a rare and clustered trait in a finite population is challenging: traditional

sampling designs usually require a large sample size in order to obtain reasonablyaccurate estimates, resulting in a considerable investment of resources infront ofthe detection of a small number of cases. A notable example is the case of WHO’stubercoulosis (TB) prevalence surveys, crucial for countries that bear a high TBburden, the prevalence of cases being still less than 1%. In the latest WHO guide-lines, spatial patterns are not explicitly accounted for, with the risk of missing alarge number of cases; moreover, cost and logistic constraints can pose furtherproblems. After reviewing the methodology in use by WHO, the use of adaptiveand sequential approaches is discussed as natural alternatives to overcome thelimitations of the current practice. A small simulation study is presented to high-light possible advantages and limitations of these alternatives, and an integratedapproach, combining both adaptive and sequential features in a single samplingstrategy is discussed as a promising methodological perspective.

Title: Dealing with under-coverage bias via Dual/Multiple Frame designs: a simula-tion study for telephone surveys

Presenter: Emanuela Furfaro, Università degli studi di Milano-BicoccaAbstract: Multiple-frame surveys are commonly used to deal with under-coverage.The use

of more than one frame introduces the possibility that frames overlap leadingto in-creased inclusion probabilities of units that appear in multiple lists. Followingtheguide example of a dual frame set-up (DF) in telephone surveys, this contribu-tionpresents an extensive simulation study where different types of screeningstodeal with the overlap issue are compared with the proper DF approach. Weempiricallyshow that the efforts of screening do not guarantee estimators moreefficientthan the DF estimators that do not need screening. Moreover simulationresults showthat screening at sample level does not correct for the increased in-clusion probabilityof units in both frames produced by the DF set-up nor improveefficiency. Thedifferent estimation options will be compared with regards to sur-vey costs, amountof information required and statistical properties of the finalestimates.

Title: Weight adjustment procedures for the treatment of unit nonresponse in surveysPresenter: David Haziza, Université de MontérealAbstract: Weighting procedures are commonly applied in surveys to compensate for non-

sampling errors such as nonresponse errors and coverage errors. Two types ofweight adjustment procedures are commonly used in the context of unit nonre-sponse: (i) nonresponse propensity weighting followed by calibration, also knownas the two-step approach and (ii) nonresponse calibration weighting, also knownas the one-step approach. We discuss both approaches and warn against the poten-tial pitfalls of the one-step procedure. Results from a simulation study, evaluating

9

the properties of several point estimators, are presented.

Title: Empirical likelihood multiplicity adjusted estimator for multiple frame surveysPresenter: Ewa Kabzinska, University of SouthamptonCo-author(s): Yves G. BergerAbstract: Multiple frame surveys are commonly used for a variety of reasons, including cor-

recting for frame undercoverage, increasing precision of estimation of populationparameters for groups of interest, targeting rare populations and reducing surveycosts. Several approximately design unbiased estimators have been proposed forinference from multiple frame surveys. It has been shown that most of the exist-ing estimators can be generalized as a class of Generalized Multiplicity-AdjustedHorvitz-Thompson Estimators. We develop an Empirical Likelihood approachto the Multiplicity-adjusted estimator. The proposed estimator allows for severalmultiplicity adjustments. It can handle auxiliary information and can be appliedto a variety of parameters of interest expressed as unique solutions to estimatingequations. Under certain sampling designs, Wilks-type confidence intervals canbe calculated without variance estimates.

Solicited Session SOL-04 • Wednesday 8, 14:15-15:45 • Room: G. De Rosa

STATISTICAL MODELS AND METHODS FOR NETWORK DATAChair: Maria Rosaria D’Esposito

Title: Measuring stability of co-authorship structures in timePresenter: Marjan Cugmas, University of LjubljanaCo-author(s): Anuška FerligojAbstract: Kronegger et al (2011) studied the co-authorship networks of four scientific dis-

ciplines in Slovenia and identified the most typical collaboration structure as hav-ing three basic positions: multi—core, semi—periphery and periphery. Cugmaset al (2015) confirmed the assumed structure being present in all Slovenian sci-entific disciplines and furthermore addressed the question of the stability of ob-tained cores. The presentation addresses the measurement of the stability of ob-tained cores assuming different operationalization of the stability of cores. Severaladopted Rand and Wallace indices are presented and compared based on empir-ical co-authorship networks of some selected Slovenian scientific disciplines intwo time periods.

Title: A dynamic discrete-choice model for movement flowsPresenter: Johan Koskinen, University of ManchesterCo-author(s): Tim Mueller, Thomas GrundAbstract: Consider the case when individual units, call them actors, are affiliated with

at most one, higher organisational unit but these affiliations change over time.This could be the case of people working for organisations or people living inneighbourhoods. We draw on dynamic models for social networks to propose anactor-oriented model for how these affiliations change over time. These modelsspecifically take into account constraints of the system and allow for the systemto be observed at discrete time-points. Constraints stem from the fact that noteverybody can get the same job or live in the same neighbourhood, something

10

which induces dependencies among the decisions. The model encompasses twomodelling components: a model for determining the termination of an affiliation;and a discrete-choice model for determining the new affiliation. For estimationwe employ a Bayesian data-augmentation algorithm, that augments the observedstates with unobserved sequences of transitions. We apply the proposed methodsto a data set of house-moves in Stockholm and illustrate how we may infer themechanisms that sustain and perpetuate segregation on the housing market.

Title: Prototyping and Comparing Networks through Archetypal AnalysisPresenter: Giancarlo Ragozini, Università di Napoli Federico IICo-author(s): Domenico De Stefano, Maria Rosaria D’EspositoAbstract: In this paper we propose a method to prototype, synthesize and compare a set of

N networks that refer to a common scenario and that are comparable among eachother. Examples of this type of data are:a set of collaboration networks, each de-fined for a different scientific field; or a set of ego networks,where egos belong toa same category; a set of governance networks, etc. For these kind of sets of net-worksit can be of interest to find a small number of representative networks thatcan serve as a condensed view of the data set. In a statistical perspective this goalamount to find a small number of networks that are able to typify the networkstructures starting from the observed ones. In addition, these networks shouldhave a clear and interpretable profile in terms of their most relevant features andtheir specificity in contrast to the others.Given the set of N networks, we proposeto find these representative networks by using the archetypal analysis, yieldingwhat we call Archetypal Networks.The Archetypal Networks can serve to under-stand the data structure, as benchmarks for the othernetworks, and are useful alsoto compare networks among each other.We perform a simulation study to showhow the proposed methodology works.

Title: Modeling network dynamics: evidence from policy-driven innovation networksPresenter: Susanna Zaccarin, Università di Milano BocconiCo-author(s): Annalisa Caloffi, Domenico De Stefano, Federica Rossi, Margherita RussoAbstract: Longitudinal networks are special relational structures where the time is an ex-

plicit element of the representation. In previous network research, the time di-mension has been projected out by aggregating the edges, even in cases of theavailability of detailed information on the temporal sequences of contacts or in-teractions. Nowadays there is a growing interest in explicitly account for timedimension in network analysis. In particular, the leading questions are related tothe underlying mechanisms that induce the network evolution. An important mod-eling approach – stochastic actor oriented models, SAOM– to study longitudinalnetworks focuses on the analysis of the underlying mechanisms that induce thenetwork dynamics (network structures) from the individual actor choices (microdynamics). These models are a family of statistical models aiming at describingnetwork dynamics according to some typical network effects, the so-called localstructures representing a large variety of network statistics. The stochastic processmodeled by these effects provides a representation of the changes of the networkover time. In the present contribution we use SAOM as a suitable methodolog-ical framework to investigate the evolution of policy-driven innovation networksamong heterogeneous agents. The empirical application focuses on a set of poli-cies that support networking among SMEs and other partners implemented bythe Tuscany region (Italy) in the programming period 2000-2006. We consider

11

a set of about 350 agents involved in 168 funded projects of variable durationranging from 4 up to 18 months. Model specification accounts for several struc-tural hypotheses on strategic partner choice in undirected networks (transitivity,popularity, homophily), controlling for agents characteristics (e.g. agent type, lo-calisation) and project duration. In order to realistically frame how the decisionto create or dissolve collaboration ties is taken, we consider either one-sided ortwo-sided initiative models.

Solicited Session SOL-05 • Wednesday 8, 14:15-15:45 • Room: A

RECENT DEVELOPMENTS IN COMPUTATIONAL STATISTICSChair: Brunero Liseo

Title: A conditional algorithm for Bayesian finite mixture models via normalized pointprocess

Presenter: Raffaele Argiento, CNR-IMATI and School of Mathematics, Statistics and Ac-tuarial Science (SMSAS) at the University of Kent, Canterbury

Abstract: Modelling via finite mixtures is one of the most fruitful Bayesian approach, partic-ularly useful when there is unobserved heterogeneity in the data. The most pop-ular algorithm under this model is the reversible jump MCMC, that can benontrivial to design, especially in high-dimensional spaces. In this work, we firstintroduce a class of finite discrete random probability measures obtained by nor-malization of finite point processes. Then, we use the new class as the mixingmeasure of a mixture model and derive its posterior characterization. The re-sulting new class encompasses the popular finite Dirichlet mixture model; here,in order to compute posterior, we propose an alternative to the reversible jump.In particular, borrowing notation from the nonparametric Bayesian literature, weset up a conditional MCMC algorithm based on the posterior characterization ofthe unnormalized point process. In order to show the performance of our algo-rithm and the flexibility of the model, we illustrate some examples on the popularGalaxy dataset.

Title: Thompson sampling for species discoveryPresenter: Stefano Favaro, Università di Torino and Collegio Carlo AlbertoCo-author(s): Marco Battiston, Yee Whye TehAbstract: This work proposes a new methodology for discovering new species, when obser-

vations are sampled from different populations. Using a metaphor, we imagine Jpopulations of animals to be available and we can sequentially choose from whichof these populations to collect further samples. Both labels and fre- quencies ofthese species are unknown a priori. At each time step, the proposed strategy sug-gests where to collect the next observation in order to maximize the number oftotal species observed. This strategy is based on a joint use of the Hier- archi-cal Pitman-Yor process, to estimate the unknown distributions of animals, andof Thompson Sampling for the sequential allocation problem. Performances ofthe algorithm are compared to those of other three strategies through simulations.

12

Title: An application of Reinforced Urn Process to advice network dataPresenter: Antonietta Mira, Università della Svizzera italiana e Università dell’InsubriaCo-author(s): Stefano Peluso, Pietro Muliere, Francesca Pallotti, Alessandro LoniAbstract: We propose a model for network data built as a system of Polya urns. The

urns are located on the nodes of the network and their composition is updatedthrough the walk of the Reinforced Urn Process of Muliere, Secchi and Walker(Stochastic Processes and their Applications, 2000). A local preferential attach-ment scheme is implied, where node popularity positively depends on its strengthand on the present position of the urn process. We derive the likelihood of in-finitely reinforced random network both for directed and undirected, weightedand unweighted network data. The model is applied to the advice network ofstudents at the Universita’ della Svizzera italiana.

Title: Bootstrap prepivoting in the presence of many nuisance parametersPresenter: Nicola Sartori, Università degli Studi di PadovaCo-author(s): Ruggero Bellio, Ioannis Kosmidis, Alessandra SalvanAbstract: We consider a scalar parameter of interest in the presence of nuisance parame-

ters. Standard likelihood inference is based on the asymptotic normality of thesigned root likelihood ratio statistic, also called directed deviance. Such asymp-totic result could be quite inaccurate with many nuisance parameters. An alterna-tive, which guarantees higher accuracy, is based on the asymptoic normality of ananalytical modification of the directed deviance, called the modified directed de-viance. In a two-index asymptotic setting, in which both information and numberof the nuisance parameters grow, the modified directed deviance has been provedto be very accurate even in extreme scenarios with many parameters and verylimited information.Here, we consider a different solution based on the paramet-ric bootstrap distribution of the directed deviance, called bootstrap prepivoting.In standard setting, this has the same accuracy as the modified directed deviance,although being more computationally demanding. On the other hand, one ad-vantage of bootstrap prepivoting is that it is available even in some non regularcases in which the modified directed deviance is not computable. We investigatethe properties of the bootstrap prepivoting in the two-index asymptotic setting.Preliminary simulation results indicate that the method is at least as accurate asthe modified directed deviance, and even more accurate in some very extremescenarios.

Solicited Session SOL-06 • Wednesday 8, 14:15-15:45 • Room: B

STATISTICIANS MEET NATURALISTS: ISSUES ON ECOLOGICAL AND ENVIROMENTALSTATISTICS

Chair: Lorenzo Fattorini

Title: Estimating the abundance of wildlife ungulate populations in Mediterraneanareas: methods, problems and findings

Presenter: Francesco Ferretti, Università di Siena and Maremma Regional Park AgencyCo-author(s): Andrea SforziAbstract: Wild ungulates have a great importance from ecological, conservation and man-

agement points of view. Monitoring their population trends is fundamental to

13

evaluate appropriate management/conservation actions. We used a two-stage strat-ified sampling to estimate densities of Italian roe deer Capreolus capreolus italicusand fallow deer Dama dama, and indices of relative abundance of wild boar Susscrofa, through pellet group counts, in a Mediterranean protected area. Accordingto an adaptive management framework, field tests, simulations and revisions ofsampling procedures lead to substantially reduce the width of confidence inter-vals and to define a final scheme (2000-2007). Large strata (defined according tohabitat/local features) were partitioned into spatial units; a sample of units was se-lected through a probabilistic scheme (1st stage). Plots were placed onto selectedunits through tessellation stratified sampling; pellet groups were counted withinthem (2nd stage). In small strata, only the 2nd stage was performed. Unbiasedestimators of abundance and conservative estimators of their variances were de-rived. This sampling design has been used yearly since 2007 to monitor ungulatedensities in the area and to evaluate management actions to improve the conser-vation of native Italian roe deer and to limit impacts of fallow deer and wild boaron ecosystems/agriculture.

Title: The monitoring of forests in Europe: methods, problems and proposalsPresenter: Marco Ferretti, TerraData environmetricsAbstract: In Europe, forest condition monitoring (FCM) is based on two networks, termed

Level I and Level II. With the former, annual sample surveys are carried out on ca.6000 plots selected according to a systematic design. At each grid intersection,a number of trees is selected and assessed in terms of defoliation (an estimate ofthe lack of foliage on trees’ crown). Data are then used to obtain information onstatus and trend of forest health in Europe. With the latter, several investigationsare carried out in purposively selected sites to study relationships between forestsand environmental drivers. In the past scarce attention has been paid to define (i)requirements for status and change assessment, as well as (ii) a sampling strategyable to provide unbiased and consistent estimators of FCM parameters and theirchanges over time. Here, problems related to FCM in Europe and proposals forpossible solutions are presented. If implemented, they will permit proper estimateof FCM status and changes at country and European levels. Further examples ofsampling-related issues typical of forest monitoring are presented and discussed.

Title: The power of generalized entropy for biodiversity assessment by remote sensing:an open source approach

Presenter: Duccio Rocchini, Fondazione Edmund MachCo-author(s): Luca Delucchi, Giovanni BacaroAbstract: The assessment of species diversity in relatively large areas has always been a

challenging task for ecologists, mainly because of the intrinsic difficulty to judgethe completeness of species lists and to undertake sufficient and appropriate sam-pling. Since the variability of remotely sensed signal is expected to be relatedto landscape diversity, it could be used as a good proxy of diversity at specieslevel. It has been demonstrated that the relation between species and landscapediversity measured from remotely sensed data or land use maps varies with scale.While traditional metrics supply point descriptions of diversity, generalized en-tropy’s framework offers a continuum of possible diversity measures, which differin their sensitivity to rare and abundant reflectance values In this paper, we aimat: i) discussing the ecological background beyond the importance of measuringdiversity based on generalized entropy and ii) providing a test on an Open Source

14

tool with its source code for calculating it.We expect that the subject of this paperwill stimulate discussions on the opportunities offered by Free and Open SourceSoftware to calculate landscape diversity indices.

Specialized Session SPE-05 • Wednesday 8, 14:15-15:45 • Room: Aula Magna

NONLINEAR TIME SERIESChair: Simone Giannerini; Discussant: Giampiero Gallo

Title: Probabilistic properties of Self Exciting Threshold Autoregressive processesPresenter: Marcella Niglio, Università di SalernoCo-author(s): Giordano Francesco, Cosimo Damiano VitaleAbstract: In the present paper we focus the attention on the ergodicity (and stationarity) of

the Self Exciting Autoregressive (SETAR) process. In more detail we review andcompare the results given in the literature highlighting the main theoretical issues.Starting from these contributions, we debate, taking advantage of case studies,on the opportunity to further investigate on the statistical properties of the SETARmodels and in particular on the possibility to weaken its stationarity conditions.

Title: Optimal prediction of stochastic trendsPresenter: Tommaso Proietti, Università di Roma Tor Vergata, CREATES, AarhusCo-author(s): Alessandro GiovannelliAbstract: A recent strand of the time series literature has considered the problem of estimat-

ing highdimensional autocovariance matrices, for the purpose of out of sampleprediction. For an integrated time series, the Beveridge-Nelson trend is dened asthe current value of the series plus the sum of all forecastable future changes. Forthe optimal linear projection of all future changes into the space spanned by thepast of the series, we need to solve a high-dimensional Toeplitz system involvingn autocovariances, where n is the sample size. The paper proposes a non paramet-ric estimator of the trend that relies on banding, or tapering, the autocovariancesequence to achieve consistency. We derive the properties of the estimator andcompare it with alternative parametric estimators based on the direct and indirectnite order autoregressive predictors. We then consider the estimation of trendswithin a multivariate system composed of a target series (e.g. gross domesticproduct) and a set of observable dynamic factors.

Title: On model selection from a finite family of possibly misspecified modelsPresenter: Howell Tong, University of Electronic Science and Technology, Chengdu, China,

and London School of Economics, UKCo-author(s): Hsiang-Ling Hsu, Ching-Kang IngAbstract: Model selection problems are usually classified into two categories according to

whether the data generating process (DGP) is included among the family of can-didate models. The first category assumes that the DGP belongs to the candidatefamily, and the objective of model selection is simply to choose this DGP. Thesecond category assumes that the DGP is not one of the candidate models. In thiscase, one of the top concerns is to choose the model having the best predictioncapability. However, most existing model selection criteria can only perform wellin at most one category, and hence when the underlying category is unknown, the

15

choice of selection criteria becomes a key point of contention. In this article, wepropose a misspecification-resistant information criterion (MRIC) to rectify thisdifficulty under the fixed-dimensional framework, which requires that the set ofcandidate models is fixed with the sample size. We prove the asymptotic effi-ciency of MRIC regardless of whether the true model belongs to the candidatefamily or not. We also illustrate MRIC’s finite-sample performance using MonteCarlo simulation.

Contributed Session CON-01 • Wednesday 8, 16:15-17:45 • Room: A

BAYESIAN STATISTICS (1)Chair: Bernardo Nipoti

Title: Reference priors based on composite likelihoodsPresenter: Federica Giummolè, Università Ca’ Foscari di VeneziaCo-author(s): Valentina Mameli, Laura VenturaAbstract: In this paper we propose reference priors obtained by maximizing theaverage

alpha-divergence from the posterior distribution, when the latter is computedusinga composite likelihood. Composite posterior distributions have already beencon-sidered in Pauli et al. (2011) and Ribatet et al. (2012), when a full likelihood forthe data is too complex or even not available. The use of a curvature correctedcomposite posterior distribution, as in Ribatet et al. (2012), allows to apply themethod in Liu et al. (2014) for maximizing the asymptotic Bayes risk associatedto an alpha-divergence. The result is a Jeffreys type prior that is proportional tothe square root of the determinant of the Godambe information matrix.

Title: On Bayesian nonparametric inference for discovery probabilitesPresenter: Bernardo Nipoti, Collegio Carlo AlbertoCo-author(s): Julyan Arbel, Stefano Favaro, Yee Whye TehAbstract: Given a sample of size n from a population of species with unknown proportions,

a common problem of practical interest consists in making inference on the prob-ability Dn(l) that the (n+ 1)-th draw coincides with a species with frequency lin the sample, for any l ≥ 0. Under the general framework of Gibbs-type priorswe show how to derive credible intervals for a Bayesian nonparametric estimatorof Dn(l).

Title: Relabelling in Bayesian mixture models by pivotal unitsPresenter: Roberta Pappada’, Università degli Studi di TriesteCo-author(s): Leonardo Egidi, Francesco Pauli, Nicola TorelliAbstract: A simple procedure based on relabelling to deal with label switching whenexplor-

ing complex posterior distributions by MCMC algorithms is proposed. Althoughitcannot be generalized to any situation, it may be handy in many applicationsbe-cause of its simplicity and low computational burden. A possible area whereitproves to be useful is when deriving a sample for the posterior distribution aris-ingfrom finite mixture models when no simple or rational ordering between the-components is available.

16

Title: On Deconvolution of Dirichlet-Laplace MixturesPresenter: Catia Scricciolo, Università di VeronaAbstract: The problem of estimating the mixing distribution of location mixtures is consid-

ered. Some empirical Bayes procedures are reviewed focussing on a recent pro-posal based on a two-step "bin-and-smooth" procedure applied to the histogramcounts. Understanding of the theoretical properties of this estimator is enrichedwith some contributions on the accuracy of the recovery for the mixing density.

Contributed Session CON-02 • Wednesday 8, 16:15-17:45 • Room: B

STATISTICAL MODELINGChair: Pieter Kroonenberg

Title: A New Bivariate Regression Model for Count Data with Excess ZerosPresenter: Pouya Faroughi, Islamic Azad University, Sanandaj, IranCo-author(s): Noriszura IsmailAbstract: Count data often display excessive number of zero outcomes compared to what is

expected in Poisson regression. Zero-inflated Poisson (ZIP) regression has beensuggested to handle purely zero-inflated data, whereas zero-inflated generalizedPoisson regression model has been fitted for zero-inflated data with additionaloverdispersion. For bivariate and zero-inflated data, several regression modelssuch as bivariate zero-inflated generalized Poisson (BZIGP) have been consid-ered. This paper introduces a new form of BZIGP. The main advantages of suchnew form of BZIGP regression are that it has flexible form of mean-variance rela-tionship, it can be fitted to bivariate zero-inflated count data with positive, zero ornegative correlations, and it allows additional overdispersion of the two dependentvariables.

Title: Dynamic latent class profiles in cross-sectional surveys: some preliminary re-sults

Presenter: Brian Francis, Lancaster University, UKCo-author(s): Valmira HotiAbstract: This paper takes a latent class approach to the problem of assessing changing hu-

man values over time in cross-sectional surveys. Traditional methods of analysiswould assume that human value class progfiles are time constant but with the pro-portion of each class varying obver time. However, this assumption may not betrue and the nature of human values may be changing over time We investigatetyhis possibility by investigating the European Social Survey gfor the UK. The21 human values in the ESS are analysed and various models of profile changeare considered. The conclusion is that human values are indeed chaningover timewith change focused on one or two specific items.

Title: The use of deviance plots for non-nested model selection in loglinear models,structural equations, three-mode analysis

Presenter: Pieter M Kroonenberg, Department of Education and Child Studies- LeidenUniversity - The Three-Mode Company

Abstract: Mallows defined the Cp statistic with an associated Cp-plot to be used in modelselection in regression analysis. The deviance plot is a generalisation of this

17

idea, where the loss, expressed in Residual sum of squares, or the Chi-squaredstatistic is graphed against the degrees-of-freedom, thus allowing for comparingDeviance/df ratios between models. It is shown that the RMSEA (Root MeanSquared Error of Approximation), AIC (Akaike’s information criterion) and theBIC (Bayesian information criterion) are lines in the deviance plot so that thesecriteria can be used for additional support in model selection. Various authorshave developed special procedures for automatic selection by finding the pointwhere the convex hull of the optimal solutions shows the greatest inflexion. Suchprocedures have made simulation studies possible to evaluate the quality of theselection procedures.

Title: Variable selection through Multinomial LASSO for PCMRPresenter: Antonio Lucadamo, Università del SannioCo-author(s): Luca GrecoAbstract: The use of multinomial regression, with particularly attention to classification,

may be made complicated by the occurrence of multicollinearity issues. Thisis true, for instance, in chemometrics when soil texture classification is accom-plished by reflectance spectrometry, since the set of measurements at differentwavelengths constitutes a high dimensional set of highly correlated predictors.Asolution to deal with multicollinearity is the use of the Principal ComponentMultinomial Regression. Its actual employ depends on the ability to select a verylow dimensional set of components.Here, a procedure is described to select thecomponents to be used that is based on the grouped LASSO.

Title: Integrating CUB Models and Vignette ApproachesPresenter: Omar Paccagnella, Università di PadovaCo-author(s): Serena Pavan, Maria IannarioAbstract: CUB models and the Anchoring Vignettes approach are two solutions, recently

introduced in the literature, for analysing ordinal data. This work aims at inte-grating the main features and advantages of these two approaches, in order toimprove the quality of the analysis of ordinal data (for instance in cross-countrycomparisons) and overcoming the limits that characterise each model solution.

Contributed Session CON-03 • Wednesday 8, 16:15-17:45 • Room: D

DEMOGRAPHICS AND SOCIAL STATISTICS (1)Chair: Antonino Di Pino

Title: Gender egalitarianism, education and life-long singlehood: A multilevel anal-ysis

Presenter: Daniela Bellani, Universitat Pompeu FabraCo-author(s): Gosta Esping-Andersen, Lesia NedoluzhkoAbstract: In this study we use multilevel modeling to examine how gender relations in-

fluence life–long singlehood across educational strata in Europe. We focus onwomen and men aged 40-55 who have never experienced a co-residential part-nership. Our results reveal reverse U-shaped relationship between levels of genderequity and singlehood: in countries where traditional gender norms still prevail aswell as in countries where gender symmetric norms are broadly diffused through-

18

out the population, we observe low levels of non-partnering. At the same timeincidence of singlehood is comparatively high in the countries which undergotransitional stage from traditional to egalitarian gender regime. The highest risksof singlehood correspond to the intermediate levels of gender equity. Our resultsalso show relevant differences in the association of gender egalitarianism on sin-glehood across levels of education for each gender.

Title: Fear of Crime and Victimization among Sexual Harassed Women: Evidencefrom Italy

Presenter: Luigi Colangelo, Università del SannioCo-author(s): Paola ManciniAbstract: This paper is about the relationship between criminal victimization and people’s

perception of safety. The study refers simultaneously to three indicators of fear ofcrime: feeling unsafe alone in the streets after dark, fear of sexual violence (heror her relatives), fear of crime affecting own habits - against sexual victimizationand characteristics of the place of residence, controlling demographic and socio-economic attributes of Italian women. The main results show that a) the prior vic-timization and crime exposure shape perceived future risk; b) fear of crime partlyreflects worries about one’s degraded context; c) married women seem more af-fected by the fear of crime, especially the fear of suffering directly or indirectlyany sexual violence.

Title: A survival approach for the analysis of cruise passengers’ behavior at the des-tination

Presenter: Stefano De Cantis, Università di PalermoCo-author(s): Mauro Ferrante, Anna Maria Parroco, Noam ShovalAbstract: The present work aims at proposing an analysis of cruise passengers’ behavior at

the destination through a survival analysis approach. Data collected through GPSdevices on cruise passengers’ behavior in the port of Palermo and Dubrovnikare analyzed in order to show similarities and differences among behaviors atthe destination, according to socio-demographic characteristics. Results are ofinterest from both a methodological perspective, related with the processing andthe analysis of GPS data, and from the destination management perspective.

Title: Retirement of the Male Partner and the Housework Division in the Italian Cou-ples: Estimation of the Causal Effects

Presenter: Antonino Di Pino, Università di MessinaCo-author(s): Maria Gabriella CampoloAbstract: The estimation of the effect of retirement of one of the partners on the couple’s

labour division may suffers of lack of robustness as a consequence of misspec-ification of the latent bargaining process between partners. In this study, usingItalian Time Use data, we provide a correction of the effect of the latent influ-ence of bargaining on estimates. In particular, the woman’s perception of equityin labour division, assumed as a relevant proxy of bargaining, is here adopted tocorrect the estimates of the causal effects of retirement of the male partner on thetime devoted to housework by both partners

Title: Many women start, but few continue: determinants of breastfeeding in ItalyPresenter: Francesca Lariccia, IstatCo-author(s): Antonella PinnelliAbstract: The aim of this study is to analyse factors of onset and duration of breastfeed-

19

ing in Italy, examining the characteristics not only of mother and child but alsothe nature of the healthcare provided, also concerning the implementation of theBaby Friendly Hospital Initiative (BFHI) and the formal and informal supportreceived by the woman. The data originates from the Multipurpose HouseholdSurvey "State of health and utilization of health service" conducted by ItalianNational Institute of Statistics in 2004-2005. After a preliminary descriptive anal-ysis, logistic regression models were used to study the determinants of the onsetof breastfeeding and the duration of exclusive/predominant breastfeeding. Resultsshow that support of primiparous women is the measure to take in order to im-prove frequency and duration of breastfeeding, as regards both the current childand future offspring.

Contributed Session CON-04 • Wednesday 8, 16:15-17:45 • Room: Sala Studio

ENVIRONMENTAL STATISTICSChair: Fabrizio Durante

Title: Measuring sustainable economic development through a multidimensional Giniindex

Presenter: Filippa Bono, Università di PalermoCo-author(s): Marcella Giacomarra, Rosa GiaimoAbstract: This paper analyse the evolution of the sustainable economic development in-

equality in Italy as regard the efforts made by each administrative Regions as aresponse to the main EU policies issued in the environmental and energy sector.To this scope, a multidimensional generalization of the Gini index has been per-formed, taking into account two different dimensions (energy and environment),in a time period six years long (2008-2013). The multidimensional Gini resultsconfirm the positive effect recorded by such kind of EU policies in determininga reduction in the inequality levels among the Italian Regions. A counterfactualanalysis further underlined the relevant role played by the energetic dimensionagainst the environmental one in strengthening Regions performances.

Title: Modeling multi-site individual corals growthPresenter: Crescenza Calculli, Università di FoggiaCo-author(s): Barbara Cafarelli, Daniela Cocchi, Elettra PignottiAbstract: Monitoring corals growth trends via individual demographic variablesrepresents,

for marine biologists, a crucial issue for assessing corals population status.Thevon Bertalanffy function (VBGF) is widely used by non-statisticians for predict-ing the growth of several marine organisms. Nevertheless, parameter estimationof this model requires linearization techniques which neglect environmental fac-tors influence and potential variability in corals colonies. In order to overcomethese limitations, a more thorough approach based on a hierarchical non-linearmixed effects model is proposed. Two model specifications, respectively basedonthe standard and a new VBGF parameterizations, are evaluated using data ofthe solitary Mediterranean coral, L. pruvoti.

20

Title: GAMs and functional kriging for air quality dataPresenter: Francesca Di Salvo, Università di PalermoCo-author(s): Antonella Plaia, Mariantonietta RuggieriAbstract: In environmental sciences data having spatio-temporal structure are often ob-

served. They can be considered as discrete observations from curves along timeand/or space and treated as functional. Generalized Additive Models (GAMs)representa useful tool for modelling, for example, pollutant concentrations and describingtheir spatial and/or temporal trends.Very often the prediction of a curve at an un-monitored site is necessary. At this aim we extend kriging for functional data to amultivariate context. Actually, even if we are interested only in predicting a singlepollutant, for example PM10, exploiting its correlation with the other pollutantscan improve the estimation. Cross validation is used to test the performance ofthe proposed procedure.

Title: The Kendall distribution and multivariate risksPresenter: Fabrizio Durante, Libera Università di BolzanoAbstract: We present the multivariate probability integral transform and its distributionfunc-

tion, called Kendall distribution, and show some of its applications inenvironmen-tal risk. On one hand, we illustrate how the Kendall distribution maybe used incopula-based models to provide an estimation of engineering structuralrisk. Onthe other hand, we exploit it to determine a suitable notion of quantile forrandomvectors, which can quantify the compound risk of extreme events.

Contributed Session CON-05 • Wednesday 8, 16:15-17:45 • Room: G. De Rosa

HEALTH STATISTICSChair: Giorgio Eduardo Montanari

Title: Dental care systems across Europe: the case of SwitzerlandPresenter: Enrico di Bella, Università di GenovaCo-author(s): Lucia Leporatti, Ivo Krejci, Stefano ArduAbstract: During the last two years a strong debate emerged on a possible reform ofthe

Swiss dental healthcare system as the cantons of Vaud and Neuchâtel startedareferendum process to extend the mandatory healthcare insurance also to den-tal care. The aim of this work is to detect the impact of an insurance on dentalcare use and tocompare the systems applied in Europe in terms of level of unmetneeds for dentalcare services; different socio-demographic categories of individ-uals will be compared.Results highlight that the level of unmet needs for dentalcare in Switzerland is inline with the one recorded in other European countriesbut it is significantly higher than the level of unmet needs for general healthcare.In addition, internal divergences emerged within the Swiss cantons.

Title: Multi-state models for hospitalizations of heart failure patients in Trieste.Presenter: Francesca Gasperoni, Politecnico di MilanoCo-author(s): Francesca Ieva, Giulia BarbatiAbstract: In Italy, heart failure (HF) is the most common clinical diagnosis among people

over 65 and about 80,000 new cases per year are recorded. This is a chroniccondition whose incidence is strictly connected with age. A very flexible tool

21

for modelling the chronic nature of this disease is multi-state model. Multi-statemodels are based on continuous time stochastic process, that can have Markov orsemi-Markov property. In this presentation we show a multi-state model in whichthe states are the admission from and discharge to the first five hospitalizations,and death, an absorbing state. This representation aims at describing the maincharacteristics of heart failure patients’ hospitalization progression consideringvarious covariates recorded.

Title: Multi-state Approach to Administrative Data on Patients affected by ChronicHeart Failure

Presenter: Francesco Grossetti, Politecnico di MilanoCo-author(s): Francesca Ieva, Simonetta Scalvini, Anna Maria PaganoniAbstract: In the present study, we make use of data extracted from the administrative data

warehouse of Regione Lombardia within a Chronic Heart Failure (CHF) project.We exploit multi-state framework on data regarding hospital admissions, drugprescriptions, and outpatient cares to estimate the probability of transition from(re)admission to discharge and death adjusting for state dependent covariates. Tothe best of our knowledge, this is the first Italian attempt of investigating the ef-fects of pharmacological and outpatient cares covariates on patient’s admissionpath. This allows to better characterise disease progression and possibly iden-tify what are the main determinants of hospital admission and death in patientsaffected by CHF.

Title: Evaluation of health care services through a latent Markov model with covari-ates

Presenter: Giorgio Eduardo Montanari, Università di PerugiaCo-author(s): Silvia PandolfiAbstract: In this work, we focus on the evaluation of the health care services provided to

elderly patients by nursing homes of four different health districts in the Umbriaregion (Italy). For this purpose, we analyze data coming from a longitudinal sur-vey aimed at assessing several aspects of patient health conditions. In the analysis,we employ an extended version of the latent Markov model with covariates thatallows us to deal with dropout and non-monotone missing data patterns, which arecommon in longitudinal studies. Maximum likelihood estimates are obtained bya two-steps approach that allows for fast estimation of the model parameters andprevents some drawbacks of the standard maximum likelihood approach encoun-tered in the presence of many response variables and covariates. In the applicationto the observed data, we show how to obtain indicators of the effectiveness of thehealth care services delivered by each health district, by means of a resamplingprocedure.

22

Contributed Session CON-06 • Wednesday 8, 16:15-17:45 • Room: V. Foa

LABOR MARKET STATISTICSChair: Giovanni Busetta

Title: Multifactor Partitioning: an analysis of employment and firm sizePresenter: Annamaria Bianchi, Università di BergamoCo-author(s): Silvia BiffignandiAbstract: This paper discusses the effects of size on employment in Italy during the crisis

started in 2008. The multifactor partitioning technique is proposed for the anal-ysis.The approach is new in this application field and proves to be useful. Theempirical investigation shows a heterogeneous behaviour among classes, espe-cially for micro-units.

Title: Ugly Betty looks for a job. Will she ever find it in Italy?Presenter: Giovanni Busetta, Università degli Studi di MessinaCo-author(s): F. FiorilloAbstract: This paper deals with the impact of attractiveness on individual’s employa-

bility, stressing the first stage of the hiring process. During a field experi-ment from August 2011 to September 2012 we sent 8256 fictitious CVs to1542 advertised on-line job openings, with attractiveness of the photo includedin the CVs (or lack thereof) being the only significant difference. We ran a logitmodel to examine the impact of gender, attractiveness, job characteristics andeducation on callbacks and hence on job opportunities. Results of the analy-sis show massive discrimination based on gender and attractiveness. Surpris-ingly, such a discrimination holds especially for unattractive women who ap-ply for highly qualified positions.

Title: No country for foreigners: an analysis of hiring process in Italian labor marketPresenter: Giovanni Busetta, Università degli Studi di MessinaCo-author(s): Maria Gabriella Campolo, Demetrio PanarelloAbstract: This paper deals with the impact of ethnic origin on individual’s employability,

stressing the first stage of the hiring process. We studied whether there is a pref-erence for Italian candidates over foreigners and, if so, whether it depends ongender, type of work, education and/or level of customer contact required. To doso, we collected data through a field experiment carried out in Italy in the periodbetween July 2013 and October 2014. During that period we sent thousands offictitious résumés, answering to real online job postings. Using a probit analy-sis we studied both first and second generation foreign candidates with differentcountries of origin to estimate whether these two components make any differencein candidate’s treatment.

Title: Know your audience. Towards a partnership between employers and university.Presenter: Franca Crippa, Università Milano-BicoccaCo-author(s): Mariangela ZengaAbstract: Employability is a major concern for Higher Education Institutions (HEI), in

terms of the success in finding a job somehow in line with qualifications attainedthrough university programmes. Some universities have embodied functions andactions, typically performed by their Job Placement Offices, into an organic plan-

23

ning of overall counselling service. This is the case of the University of Milano-Bicocca, where the Counselling University Committee, favouring more effectivesynergies between the academic body and the employers, monitors students’ dy-namics pathways. The issues of knowing employers who refer directly to aca-demic job service, on the one side, and embodying them into the stochastic anal-ysis of students’ paths, on the other side, are explored relating to the previouslymentioned academic reality.

Title: Online Job Vacancies: a big data analysisPresenter: Ilaria Vannini, IstatCo-author(s): Daniela Rotolone, Cristian Di Stefano, Achille Pierre Paliotta, Domenica Fiore-

distella IezziAbstract: In the framework of Big Data analysis, the online job vacancies stands out as a rel-

evant research area. With the aim of setting up an Italian vacancy monitor, about70,000 job posts have been collected across a three months timeframe. By fol-lowing the typical phases of a Big Data processing, a procedure for the definitionof a model able to integrate, represent and analyse the acquired data, extractedfrom heterogeneous informative sources, will be described along with techniquesfor their recognition, standardization as well as cleaning, highlighting the mostchallenging features, obstacles encountered and the operative solutions adoptedfor the future replication of the activity to the whole web job market and for thesetting up an Italian vacancy monitor.

Contributed Session CON-07 • Wednesday 8, 16:15-17:45 • Room: SP/1

ROBUST STATISTICSChair: Giovanni Porzio

Title: Robust estimation of mixtures of skew-normal distributionsPresenter: Francesca Greselin, Università degli studi di Milano-BicoccaCo-author(s): L.A. García-Escudero, A. Mayo-Iscar, G. McLachlanAbstract: Recently, observed departures from the classical Gaussian mixture model in real

datasets motivated the introduction of mixtures of skew t, and remarkably widenedthe application of model based clustering and classification to great many realdatasets. Unfortunately, when data contamination occurs, classical inference forthese models is not reliable. In this paper we introduce robust estimation of mix-tures of skew normal, to resist sparse outliers and even pointwise contaminationthat could arise in data collection. Hence, in each component, the skewed natureof the data is explicitly modeled, while any departure from it is dealt by the robustapproach. Some applications on real data show the effectiveness of the proposal.

Title: Renyi’s Scoring RulesPresenter: Monica Musio, Università di CagliariCo-author(s): Alexander Philip DawidAbstract: We consider the Renyi entropy from the point of view of the theory of proper

scoring rules. We show that this leads to a new definition of the Renyi divergence.We also apply the associated scoring rule to construct parameter estimators, andconsider their robustness properties in the case of a location family.

24

Title: Robust classification of multivariate functional dataPresenter: Anna Maria Paganoni, Politecnico di MilanoCo-author(s): Francesca IevaAbstract: We propose the joint use of two different depth indexes to detect and discard

both shape and magnitude outliers in order to robustify a reference sample data,composed by G different known groups. We consider a real dataset composedby electrocardiografic signals: we classify a data minimizing a suitable distancefrom the centers of reference groups. We compare performance of supervisedclassification on test sets training the algorithm on original dataset and on therobustified one, respectively.

Title: A robust estimator for the mean direction of the von Mises-Fisher distributionPresenter: Giovanni C Porzio, Università di Cassino e del Lazio MeridionaleCo-author(s): Thomas Kirschstein, Steffen Liebscher, Giuseppe Pandolfo, Giancarlo RagoziniAbstract: The von Mises-Fisher distribution is probably the most widely used distribu-

tionto model data on the (hyper)sphere. Characterized by two parameters, the-mean direction and concentration parameter, it is symmetrical about the meandirection.The Maximum Likelihood estimator for this parameter is the direction-alsample mean. However, this estimator is not robust. Although the estimationresultcannot be corrupted arbitrarily (but only up to a certain amount) the needfor somealternative (robust) estimators has been recognized.Within this work, itis first discussed how the robustness of a directional locationestimator can be eval-uated. Then, a new robust estimator of the mean direction isintroduced.

Title: Robust Partial Possibilistic Regression Path ModelingPresenter: Rosaria Romano, Università della CalabriaCo-author(s): Francesco PalumboAbstract: This paper introduces a procedure to robustify the partial possibilistic regression

path modeling, which is a particular structural equation model that com- bines theprinciples of path modeling with those of possibilistic regression. Possi- bilisticregression allows to model relations between variables through fuzzy coef- fi-cients, but it is very sensitive to extreme values. Aim of the proposed procedure isthen the detection of extreme values to omit or lessen their effect on the modeling.A case study on the the motivational and emotional aspects of teaching is used toillustrate the procedure.

Contributed Session CON-08 • Wednesday 8, 16:15-17:45 • Room: SP/2

SAMPLING METHODSChair: Matteo Ruggiero

Title: Adaptive Randomly Reinforced Urn design and its asymptotic propertiesPresenter: Andrea Ghiglietti, Università degli Studi di MilanoAbstract: In response-adaptive designs for clinical trials the probability to assign patients

to treatments depends on previous patients’ allocations and responses.Some ofthese designs are based on urn models to construct randomized procedures for theallocation of the subjects to treatments andto incorporate information on treat-ment performances.In particular, we consider the class of randomly reinforced

25

urn (RRU) models in whichthe proportion of subjects allocated to the superiortreatment converges to one as the sample size increases to infinity.For this reason,RRU designs are not able to target any desired allocation proportion to satisfydifferent optimality criteria.To address this issue, we propose a new adaptive ran-domly reinforced urn (ARRU) designin which the allocation proportion convergeto desired values defined as functions of the response distribution.In addition, weestablish the convergence rate and a central limit theorem for the allocation pro-portion in both the cases ofequal and different responses means.

Title: PC algorithm from complex sample dataPresenter: Daniela Marella, Università Roma TreCo-author(s): Paola VicardAbstract: The association structure of a Bayesian network can be known in advance by sub-

ject matter knowledge or have to be learned from a database. In case of datadriven learning, one of the most known procedures is the PC algorithm where thestructure is inferred carrying out several independence tests under the assumptionof independent and identically distributed observations. In practice, sample selec-tion in urveys involves more complex sampling designs. In this paper, a modifiedversion of the PC algorithm is proposed for inferring casual structure from com-plex survey data.

Title: Optimal Adaptive Group Sequential Procedure for Finite Populations in thePresence of a Cost Function

Presenter: Silvia Missiroli, Università di Milano BocconiCo-author(s): Elisabetta CarfagnaAbstract: An adaptive group sequential procedure is proposed for sample allocation in strat-

ified sampling for finite populations. The combination of number of steps andsampling units to be allocated to the strata at each step that minimizes the vari-ance of the estimator is analyzed, given a linear cost function that includes thestep cost. The solution is investigated through a Monte Carlo simulation study.The results illustrate the improved efficiency with respect to other adaptive proce-dures presented in the literature and assess the impact of the cost function on theoptimal combination.

Title: The Rao regression-type estimator in ranked set samplingPresenter: Elvira Pelle, Università della CalabriaCo-author(s): Pier Francesco PerriAbstract: We revisit the Rao regression-type estimator in the context of the ranked set sam-

pling. The expression of the minimum mean squared error is obtained and a simu-lation study is carried out to evaluate the gain of efficiency upon some competitiveestimators.

Title: Modelling stationary varying-size populations via Polya samplingPresenter: Matteo Ruggiero, Università di Torino and Collegio Carlo AlbertoCo-author(s): Pierpaolo De Blasi, Stephen WalkerAbstract: We introduce the class of birth-and-death Polya urns, which allow for the re-

moval of particles, resulting in a sample of fluctuating size. We study the inducedpartition-valued processes in terms of group multiplicity counts by encoding thesampling process into a random walk on the lattice and by exploiting the prop-erties of some auxiliary embedded models. The asymptotic regimes of this newsampling scheme exhibit a phase transition from stationary partitions with a ran-

26

dom number of groups to partitions with almost surely infinitely many groups,structurally analogous to those induced by usual Polya urns. We identify the sta-tionary distribution of the vector of multiplicity counts in the first regime, andstudy the dynamics of the underlying population process, which also has an ex-plicit stationary distribution given by a mixture of Polya urns, and is related to aclass of Moran models well known in mathematical biology.

Solicited Session SOL-07 • Wednesday 8, 17:45-19:15 • Room: G. De Rosa

FROM SURVEY DATA TO NEW DATA SOURCES AND BIG DATA IN OFFICIAL STATISTICSChair: Paolo Righi

Title: Machine learning and statistical inference: the case of Istat survey on ICTPresenter: Giulio Barcaroli, Istituto Nazionale di StatisticaCo-author(s): Gianpiero Bianchi, Renato Bruni, Alessandra Nurra, Sergio Salamone, Marco

ScarnòAbstract: Istat is experimenting web scraping, text mining and machine learning techniques

in order to obtain a subset of the estimates currently produced by the samplingsurvey on “Survey on ICT usage and e-Commerce in Enterprises”, yearly carriedout by Istat and by the other member states in the EU. Target estimates of thissurvey include the characteristics of websites used by enterprises to present theirbusiness (for instance, if the website offers e-commerce facilities or job vacan-cies). The aim of the experiment is to evaluate the possibility to use the sample ofsurveyed data as a training set in order to fit models that will be applied to the gen-erality of websites. The usefulness of such an approach is twofold: (i) to enrichthe information available in the Business Register, (ii) to increase the quality ofthe estimates produced by the survey. These different objectives can be reachedby combining web scraping procedures together with text mining and machinelearning techniques, making optimal use of all available information. First re-sults of this experiment are presented and discussed, together with planned futurework.

Title: Forecasting Italian Youth Unemployment Rate Using Online Search DataPresenter: Stefano Falorsi, IstatCo-author(s): Silvia Loriga, Alessia Naccarato, Andrea PieriniAbstract: In recent years a stream of studies have been considered to exploit the predictive

capability of web search data; numerous papers suggest the use of big data toimprove the nowcasting and forecasting of the official economic indicators witha view to increasing the promptness of their circulation. In particular this workfocuses on the use of the time series of Google Trend query shares refferringthe keyword “Offerte di lavoro” (GT), and the Labour Force time series surveydata (LF) from 2004 in Italy. The weekly GT data were aggregated into monthlyaverages in accordance with the LF survey scheme of allocation of weeks of in-vestigation to months. The aim of this work is nowcasting and forecasting theyouth unemployment rate. Nowcasting is to be understood here as estimating themonthly unemployment rate for the month in which official survey is actually un-der way. The objective is to thus to provide a provisional estimate of the indicatorfor use until all the phases of the official survey have been completed. Given that

27

the GT information is subjected to no form of quality control, it is our view thatit can be used solely as a sort of “snapshot” providing indications about the phe-nomenon of interest and its evolution in time and space. In other words, the directinclusion of information on the phenomenon of interest from sources outside theofficial surveys can create major problems as regards the quality and accuracy ofestimates. On the other hand, however, taking advantage of information aboutthe spatial and temporal dynamics of a phenomenon closely related to the one ofinterest and readily available at little cost may well offer a good opportunity toproduce estimates that can be used as a “real time” estimates. The use of thekeyword “Offerte di lavoro” is justified on the grounds of its popularity amongcompeting job-search related keywords, as can be ascertained by comparing itsrelative incidence with respect to other search terms. Furthermore, its broad def-inition makes it ideally robust against irrelevant strong variations connected todemand or supply side of labor force of specific subgroups of job seekers. Whilethe fact that not all workers have access to the Internet must also be taken into ac-count, this is a minor issue in view of its ever-increasing use as a way of seekingemployment, in particular for young people. Furthermore, the two LF monthlyseries related to the young people - the overall sample and the sub-sample of In-ternet job seekers - show marked similarities. The first results of an empiricalstudy referring a comparative analysis between different time-series models usingthe official LF data and the auxiliary information from GT, are presented here. Inparticular we compare the results obtained by VAR model exploiting LF and GTseries cross-correlation, and two ARIMA models, the first one estimated withoutusing the GT auxiliary information and, the second one involving GT informa-tion as a covariate. The empirical study shows evidences that the GT seriesprovides a real and immediate image of the level of the youth unemployment rateand therefore constitutes a useful tool with a view to obtaining a reliable nowcastand forecast of this indicator if used together with the official time series. The re-sults obtained suggest that this approach can be used as an instrument to improvethe short term economic evaluation.

Title: Bayesian nonparametric methods for record linkagePresenter: Brunero Liseo, Università di Roma SapienzaCo-author(s): Andrea TancrediAbstract: Record linkage is a class of statistical and algorithmic methods which aim at iden-

tifying whether two or more observed records refer to the same statistical entityor not. Duplications of the same entity within one single source or across differ-ent files may be interpreted as “clusters of records”, showing strong similaritiesacross their fields. In this paper we frame the record linkage process into a for-mal Bayesian clustering model and we investigate the role of species samplingmodels as the natural prior specification for the clustering structure. In fact thedifferent latent entities which produce the records observed in one or more datasources can be effectively treated as the sampled species and the observed recordsas noisy measurements of their features. We also discuss an important issue inthe clustering approach to entity resolution, that is the necessity of bounding theclusters sizes even in the presence of large data sets.

28

Title: Exploring solutions for linking Big Data in Official StatisticsPresenter: Tiziana Tuoto, IstatCo-author(s): Loredana Di Consiglio, Daniela FuscoAbstract: In these years, Official Statistics has acknowledged the value of Big Data and

has started exploring the use of diverse sources in several domains. For some ofthe sources, the object can be, a part from privacy restrictions, easily related toa statistical unit. In those cases, if a unit identifier was available, the opportu-nity to link Big data to already existing statistical data at micro-level could allowto enlarge the content, the coverage, the accuracy and the timeliness of officialstatistics. This could be the case, for instance, of the Internet-scraped data. Inthis setting, new challenges arise for data integration experts in official statis-tics, due to the deep differences with respect to the familiar framework in whichadministrative data have been integrated for a long time in order to produce sta-tistical outputs, i.e. business and population registers. In this study, exploiting areal case as a starting point, we describe novelties and challenges in integratingInternet-scraped data with traditional statistical datasets, from the entity extrac-tion and recognition phases to the unit matching algorithms, as well. As casestudy, we propose the linkage of Internet-scraped information with data related toagritourisms, as reported in the Italian Farm Register, that is obtained by the inte-gration of seven administrative sources. In order to overcome limits and rigiditiesof the so far well-established linkage procedures, we explore new techniques notyet introduced in the official statistics production system. Finally, we devote thedue attention to the output quality evaluation, in order to entirely understand ben-efits and risks of the integration and to allow the analysts to take into accountpotential integration errors in subsequent analyses.

Solicited Session SOL-08 • Wednesday 8, 17:45-19:15 • Room: V. Foa

SYMBOLIC DATA ANALYSIS METHODS AND APPLICATIONSChair: Rosanna Verde

Title: Explanatory and discriminatory power of variables in Symbolic Data AnalysisPresenter: Edwin Diday, CEREMADE- Univ. Paris DauphineAbstract: Standard statistics is based on observations which are values of numerical or cat-

egorical variables. The Symbolic Data Analysis (SDA) framework is based ontwo sets: a ground set of observed objects called “individuals” described by stan-dard observations, and a “higher” level set of classes of individuals described bythe so-called “symbolic data” as intervals, distributions, sequence of values andthe like, in order to express the variability of the individual’s observations insideeach class. “Symbolic variables” are variables defined on classes of individualsand which values are symbolic data. The SDA aim is first to build the symbolicdata from Standard, Complex (with unpaired variables) and Big Data. Second, toextend Statistic and Data Mining to symbolic data. This paper introduce a newkind of symbolic variable called “vertical bar chart variable”. It aims to define“explanatory power” and “discriminatory power” criteria of symbolic bar chartsbased variables and to study their links and properties.

29

Title: Fuzzy and possibilistic approach to clustering of imprecise dataPresenter: Maria Brigida Ferraro, Sapienza, Università di RomaCo-author(s): Paolo GiordaniAbstract: This work focuses on clustering data affected by imprecision. The imprecision is

managed in terms of fuzzy sets, in particular, LR fuzzy data. The clustering pro-cess is approached according to the fuzzy and possibilistic approaches. In boththe approaches the objects are assigned to the clusters by means of membershipdegrees. In fuzzy clustering the membership degrees express the degrees of shar-ing of the objects to the clusters whereas, in possibilistic clustering, these givethe degrees of compatibility. These two sources of information are not exclusivebecause the former helps to discover the best fuzzy partition of the objects andthe latter gives how well the objects are described by the cluster prototypes. Wepropose an hybridization of the fuzzy and the possibilistic algorithms exploitingthe benefits of both the approaches. The adequacy of the proposal is checked bymeans of simulation and real-case studies.

Title: Symbolic data analysis approach for monitoring the stability of monuments.Presenter: Laura Grassini, Università di FirenzeCo-author(s): Bruno Bertaccini, Antonio GiustiAbstract: The paper describes the work in progress about the analysis of the behaviour of the

web cracks on the Brunelleschi’s Dome of Santa Maria del Fiore in Florence. Theweb cracks in the Dome have always given rise to concern about the stability ofthe monument. The mechanical and electronic instruments have generated morethan 6 million measurements, and the analyses performed so far, showed a steadyincrease in the size of the main cracks and, at the same time, a relationship withthe environmental variables. The paper provides an analysis of those (big) datathrough the Symbolic Data Analysis techniques.

Title: Similarity and Dissimilarity Measures for Mixed Feature-type Symbolic DataPresenter: Manabu Ichino, College of Science and Engineering Tokyo Denki University

Hatoyama, JapanAbstract: This paper presents preliminary results for the similarity and dissimilaritymea-

sures based on the Cartesian System Model (CSM) that is a mathematical modeltomanipulate mixed feature-type symbolic data. We define the notion of concept-size for the description of each object in the feature space. By extending thenotionto the concept sizes of the Cartesian join and the Cartesian meet of the de-scriptionsfor objects, we can obtain various similarity and dissimilarity measures.We presentespecially the asymmetric similarity measure, and the symmetric sim-ilarity anddissimilarity measures useful for pattern recognition problems.

30

Solicited Session SOL-09 • Wednesday 8, 17:45-19:15 • Room: SP/1

COMPOSITIONAL ANALYSISChair: Valentin Todorov

Title: Forecasting CPI weights through compositional VARIMA: an application toItalian data.

Presenter: Lisa Crosato, Università Milano-BicoccaCo-author(s): Francesca Lovisolo, Biancamaria ZavanellaAbstract: Worldwide, CPIs are mostly calculated as weighted averages of price relatives

with fixed base weights. The main source of estimation of CPI weights are Na-tional Accounts, whose complexity in terms of data collection, estimation of ag-gregates and validation procedures leads to several months of delay in the re-lease ofthe figures. This ends up in a non consistent Laspeyres formula sincethe weights donot refer to the same period as the base prices do, being older byone year. In thispaper we propose to forecast CPI weights via a compositionalVARIMA model, toobtain more updated weights and, consequently, a more up-dated measure of infla-tion through CPIs.

Title: Understanding association rules from a compositional data approachPresenter: Josep Antoni Martín-Fernández, University of GironaCo-author(s): Marina Vives-Mestres, Ron KenettAbstract: When the interest is on the analysis of relations between variables of a large

database, the association rule mining have an important role. Measures of inter-estingness are the principal element in the analysis. Because any association rulecan be expressed by a contingency table, log-ratio techniques are an appropriateapproach. We introduce compositional measures and analyse its major properties.A contrast to confirm the significance of an association rule and the interpretationof the effects between the itemsets are given. A simple example illustrates theperformance of compositional measures.

Title: Object Oriented Geostatistical Simulation of Functional Compositions via Di-mensionality Reduction in Bayes spaces

Presenter: Alessandra Menafoglio, Politecnico di MilanoCo-author(s): Alberto Guadagnini, Piercesare SecchiAbstract: We address the problem of geostatistical simulation of spatial complex data, with

emphasis on functional compositions (FCs). We pursue an object oriented geo-statistical approach and interpret FCs as random points in a Bayes Hilbert space.This enables us to deal with data dimensionality and constraints by relying ona solid geometric basis, and to develop a simulation strategy consisting of: (i)optimal dimensionality reduction of the problem through a simplicial principalcomponent analysis, and (ii) geostatistical simulation of random realizations ofFCs via an approximate multivariate problem. We illustrate our methodology ona datasetof natural soil particle-size densities collected in an alluvial aquifer.

Title: Fitting CANDECOMP-PARAFAC model for compositional data: a combinedSWATLD-ALS algorithm

Presenter: Violetta Simonacci, Università di Napoli L’OrientaleCo-author(s): Maria Anna Di Palma, Valentin Todorov

31

Abstract: Multidimensional compositional arrays require special analytical tools to be mod-eled. Specifically, the variation of the data can be captured by linear combinationsof a defined number of parameters, capable of describing the complexity of thedata. Usually these models are described as generalizations of Principal Com-ponent Analysis (PCA) to higher order cases. Here the Candecomp/Parafac(CP)model is defined for compositional data contaminated with extreme observationsby using a novel integrated SWATLD-ALS algorithm. Since the SWATLD pro-ceduredoes not find a solution in the least square sense, it is expected to havea better performance in terms of sensitivity to outliers than ALS. However, dueto the instability of its loss function, it should not be used alone: we suggest tocombine SWATLD and ALS.

Solicited Session SOL-10 • Wednesday 8, 17:45-19:15 • Room: SP/2

SUSTAINABLE DEVELOPMENT: THEORY, MEASURES AND APPLICATIONSChair: Stefania Rossetti

Title: Measuring sustainable development goals from now to 2030Presenter: Fabiola Riccardini, IstatAbstract: The United Nations last September 2015 has agreed on the sustainable devel-

opment goals that will oriented the policy actions for the next 15 years. The17 goals and the related 169 targets should be monitored and evaluated. TheNational Statistical Institutes, following a decisions of The UN Statistical Com-mission, should coordinate the process of production of the statistical data forelaboration of indicators that will be used for monitoring. Therefore it is neces-sary to develop a global, regional and national set of indicators for monitoringthe achievements of the goals and targets by countries. This paper will illustratethe strategy that Istat will develop for monitoring the SDGs goals and targetstrying to develop a coherent set of measures utilizable for this porpoise. In Italywe have approached the measuring of sustainability in the context of BES, in linewith the United Nations declaration that there is sustainable development whenthe wellbeing of people is pursued. Moreover, Istat is an observer country insidethe Inter Agency Expert Group- SDGs constituted by the UN statistical Commis-sion, where the indicators for monitoring are selected. Istat is also member of theTF UNECE for developing Sustainable Development measures. All these experi-ences allow to develop a coherent framework for Sustainable Development mea-sures in which methods, measures and concepts find clarification.

Title: How the nexus of food/water/energy can be seen with the perspective on well-being of people and the Italian BES framework

Presenter: Fabiola Riccardini, IstatCo-author(s): Dalila De RosaAbstract: According to the latest OECD Global Forum on Environment “Water, food and

energy are crucial for sustainable long term economic growth and human well-being” and crucial is the identification of the strong linkages between all three.As the UN has spotted out agriculture is currently the largest user of water at theglobal level, accounting for 70% of total withdrawal while the food productionand supply chain accounts for about 30% of total global energy consumption. Be-

32

sides, according to the global population dynamic, projected to increase by 2050up to 9 billion people along with a 70% increase in food production, the resourcecompetition is turning out to become a fundamental issue. To this extent the inter-linkages analysis within the nexus and the role-played by institutions and policiesto handle effectively this resource competition is a challenging food for thought.In addition currently discussion on SDGs is taking in consideration all the threedimensions and wellbeing of people. To this purpose the aim of this work is todeal with the nexus through the lens of the Italian BES (Benessere Equo e Sosteni-bile) modeling framework. In particular focusing on food, considered as the pri-mary need more strictly related to the wellbeing achievement, all the linkagesavailable with water and energy will be investigated. The analysis will rely onthe existing BES domains and indicators and on other possible measures, whichcan fit the nexus within the BES model. The study will also spot out the linkagesbetween food, energy and water looking at different stakeholders (individuals,firms and institutions) and at both their supply and their demand side. Finally,an attempt to think about the possible synergies between the stakeholders and thepossible ways to effectively manage the nexus using relevant indicators will becarried out.

Title: An innovative methodology for the analysis of sustainability, inclusion andsmartness of growth through Europe2020 indicators.

Presenter: Tommaso Rondinella, IstatCo-author(s): Elena GrimacciaAbstract: The comparison of different territorial areas according to multiple factors raises

the challenge of representing synthetically the complexity of multidimensionalphenomena, such as the targets of growth promoted by the Europe 2020 strategy.We considered data for ten years in order to highlight the evolution of the simi-larities and dissimilarities of the 28 European countries in the whole period. Theanalysis is centred on a technique which combines cluster analysis with the useof a composite indicator, thus permitting to identify Countries both according totheir structural characteristics and to their overall performance. We also look atconvergence processes among countries and link our results to GDP growth tobetter qualify countries models of development.

Title: The Italian population behaviours toward environmental sustainability: a studyfrom Istat surveys

Presenter: Paola Ungaro, IstatCo-author(s): Isabella Mingo, Valentina TalucciAbstract: The interest of the Scientific Community in environmental protection issues aim-

ing at guaranteeing future sustainability is constantly increasing . For this reason,the environmental social sciences, in recent years, are treating the interrelation-ships between population and environment. In Italy, an informative contributioncomes from both Istat traditional Multipurpose Survey "Aspects of daily life" andIstat more recent Survey on Energy Consumption of Households. The aim of thepaper is to analyse the determinants of pro-environmental behaviours of Italiancitizens, with particular attention to those finalized to reduce the damage on theenvironment, as well as those directed to encourage energy efficiency, by deep-ening the direction of the relationships with socio-demograpic and other relevantcharacteristics, proposing easily replicable over time measures and using a multi-variate data analysis approach.

33

Solicited Session SOL-11 • Thursday 9, 09:00-10:30 • Room: G. De Rosa

DETECTING HETEROGENEITY IN ORDINAL DATA SURVEYSChair: Domenico Piccolo

Title: CUB models: a preliminary Fuzzy approach to heterogeneityPresenter: Elvira Di Nardo, Università di TorinoCo-author(s): Rosaria SimoneAbstract: Motivated by the increasing attention paid to deal with uncertainty in ordinal data

models, we propose to combine Fuzzy analysis with CUB models within ques-tionnaire analysis. In particular, our focus is n CUB models’ uncertainty pa-rameter and its interpretation as a preliminary measure of heterogeneity in mem-bership, non-membership and uncertainty functions when referred to the moregeneral framework of Intuitionistic Fuzzy Sets. Our approach is discussed onthe basis of the Evaluation of Orientation Services survey collected at Universityof Naples Federico II. Comparisons with current literature on Fuzzy analysis forquestionnaires show the validity of the proposal.

Title: Modelling uncertainty in bivariate models for ordinal responsesPresenter: Sabrina Giordano, Università della CalabriaCo-author(s): Roberto Colombi, Anna Gottard, Maria IannarioAbstract: We consider the case of two ratings driven by two latent variables, conditionally

on one or more discrete explanatory variables. The idea is that an individual re-sponds to each rating according to his/her knowledge (feeling) or following a ran-dom response style (uncertainty). The joint distribution of the ordinal responsesresults in a mixture of four components corresponding to the cases of uncertaintyin both the answers, feeling in both the answers and uncertainty in only one ofthem.The effect of covariates can be modelled on both the marginal distributionsof the ordinal variables and the association between these ordinal variables. Es-timation is pursued by means of EM algorithm. Two case studies illustrate themain results.

Title: Treatment of ‘don’t know’ responses in rating data: effects on the heterogeneityof the CUB distribution

Presenter: Marica Manisera, Università di BresciaCo-author(s): Paola ZuccolottoAbstract: In this paper we consider the problem of measuring the heterogeneity of adis-

tribution of ratings in presence of ‘don’t know’ responses, when data are fittedbya CUB model. We show that a preliminary adjustment of the model parame-ters, ac-cording to a recently proposed method for the treatment of ‘don’t know’responses,allows to correspondingly obtain an adjusted measure of heterogeneity.

Title: Modelling a multivariate hidden Markov process on survey dataPresenter: Fulvia Pennoni, Università di Milano BicoccaAbstract: We show how to handle the information which is acquired in a dynamic frame-

work when there are multiple items in a survey collected on the same individualsat different time occasions. The response variables are commonly measured onan ordinal scale and the data may show a non-monotone missing pattern. Theunderlying phenomenon which is related to the interest of the survey may be

34

modelled by a latent stochastic process. The latter having dependences accord-ing to a Markov structure is able to capture the heterogeneity of the responsebehaviour and to account for the measurement errors that naturally arise in thesurvey. The maximum likelihood estimation of the model parameters allows usto handle the missing data and to take advantage of the information provided bythose individuals not showing complete responses.

Solicited Session SOL-12 • Thursday 9, 09:00-10:30 • Room: V. Foa

ACTIVE AGEING: AGE MANAGEMENT AND LIFELONG LEARNING STRATEGIESChair: Alessandra De Rose

Title: Age management in Italian companies. Findings of two Isfol surveysPresenter: Paolo Emilio Cardone, Isfol Università di Roma SapienzaCo-author(s): Maria Luisa Aversa, Luisa D’AgostinoAbstract: Aim of this research project is to analyze the behavior of the Italian companies

in comparison to the solutions adopted for the maintenance and reintegration inthe labour market of ageing workers, as well as the strategies adopted for theirprofessional exploitation, based on the results of two Isfol surveys: the first one,qualitative, on the age management in the large companies; the second one, quan-titative, on the attitude of the small and medium-sized enterprises (SMEs) towardsthe ageing workers. Aim of the first qualitative search is to describe and to analyzethe most meaningful experiences realized by the great enterprises that operate insome specific segments of the industrial and services sectors to face the problemof the ageing workers and the possible obsolescence of their skills and compe-tences. Aim of the second quantitative search is to analyze the characteristicsof the relationship among the enterprises development strategies and the solu-tions adopted for the maintenance, the professional exploitation and the possiblereintegration of ageing workers, according to the latest reforms of the pensionsystem. As a result, emergent trends will be underlined in the great enterprises,with a focus on similarities and differences with respect to the SMEs, in orderto individualize feasible development perspective. It will complete the researcha reflection on the virtuous behaviors adopted by the large Italian enterprises intheme of age management in order to increase workers employability, taking intoaccount needs and skills of the different generations with a focus on the wholelife cycle.

Title: Working after Retirement in EuropePresenter: Angelo Lorenti, Max Planck Institute for Demographic Research (MPIDR), Ro-

stock Germany Department of Statistical Sciences, Sapienza, Università di RomaAbstract: This paper investigates post-retirement work of European men aged 55-74. Us-

ing SHARE data, we identify individual characteristics influencing the decision-making process that leads individuals to undertake post-retirement work. Ourresults indicate that the likelihood of re-employment is higher for pensioners atthe extremes of the pension-earnings distributions. Compared to individuals inthe third quintile, the chances for those in the first are 2.6 times higher, whilefor those in the fifth are 1.4 times higher. We find that attending training coursesand/or education has a positive impact on the chances of work. Large differences

35

in re-employment among countries are observed.

Title: Older low-skilled workers and economic crisis in ItalyPresenter: Corrado Polli, Isfol, Statistical OfficeCo-author(s): Roberto AngottiAbstract: Over the last decades in Italy has taken place an increasing ageing of the work-

force. The expansion of the population of the older cohorts, as a result of thedemographic changes, requires extended working life. The ageing of the work-force could led to social and economic consequences, from the skill obsolescenceto the risk of unemployment. Extended working life requires a workforce moreand more qualified and adaptable to the needs of competitiveness of the country.Training can have a positive impact on the employability of older workers, es-pecially when it is combined with other measures. European policies, therefore,emphasize the importance of launching a strategy toward active ageing of theworkforce, in order to strengthen the skills of the older workers and increase thesupply of training opportunities. In Italy, the participation rate in education andtraining of the adult workforce is still low. Several studies showed that a similarstructure of the workforce implies a low propensity of companies to increase theiremployees’ skills. This contributes to the spread of unskilled jobs and implicates awider unemployment risk for workers that can fall into a “low-skill, bad-job trap”.The low-skilled vacancies result in bad jobs and underemployment of the overed-ucated. The evidence emerging from the data seems to confirm these hypothesesalso for Italy, and the older low-skilled workers represent the target at highest risk.The aim of this paper is to describe the main evidences regarding the characteris-tics and the training participation of the older low-skilled workers, in the contextof the ageing trends of the workforce in Italy, from 2009 to 2014 and thereforeduring the economic crisis. In particular, we intend to identify the factors thataffect the probability of participating in training activities and the probability ofbecoming unemployed. The cross-sectional and longitudinal analysis, developedthrough some logistic regression models and carried out on LFS data (Eurostat),has allowed understanding the relevance of this target and the need of developingan active ageing strategy aimed at increasing the education and training partic-ipation, in order to enhance skills and to protect employment rates of the olderlow-skilled workers.

Title: Population ageing and human resources management. A chance for AppliedDemography.

Presenter: Giulia Rivellini, Catholic UniversityCo-author(s): Francesco Marcaletti, Filomena RacioppiAbstract: The population ageing has become a global and worldwide phenomenon. Espe-

cially in Italy this process is strong and incontestable. In this scenario the workforce ageing and the challenges that make urgent to find tools and measures tomanage the oldest people can give to the “applied demographer” a significantrole. Moving from the definition of active ageing we illustrate skills, activitiesand suggestions that a demographer could easily share with human resource man-agers interested in promoting strategies and policies enhancing employability andproductivity of older workers. We report real experiences with ABB Italia, SOLGroup and Italcementi. We finally give details on AMO (Age Management Ob-servatory).

36

Solicited Session SOL-13 • Thursday 9, 09:00-10:30 • Room: SP/1

STATISTICAL MODELS FOR EVALUATING POLICY IMPACTChair: Daniele Bondonio

Title: Evaluation of Training Programs by exploiting secondary outcomes in Princi-pal Stratification frameworks: the case of Luxembourg.

Presenter: Michela Bia, Luxembourg Institute of Socio-Economic Research (LISER)Co-author(s): Fan Li, Andrea MercatantiAbstract: The Principal Stratifications framework (PS) is widely adopted in the evaluation

of public policies, in that it allows to account for the mediation effect of an inter-mediate variable (employment status) that lies in the causal pathway between theintervention and the outcome. In this paper we investigate the causal effects ofa given Training Program on wages under the PS approach, using data from theglobal social security database on labour force in Luxembourg and the adminis-trative data from the Employment Agency. We also exploit a secondary outcome(hours worked) to the purpose of sharpening inference when the primary outcomeis censored by “death”. Finally, a sensitivity analysis will be also conducted toevaluate how different departures from unconfoundedness assumptions affect ourresults.

Title: Testing Stability of Regression Discontinuity ModelsPresenter: Giovanni Cerulli, IRCrES-CNR - National Research Council of Italy Research

Institute on Sustainable Economic GrowthAbstract: Regression discontinuity (RD) models are commonly used to nonparametrically

identify and estimate a local average treatment eect (LATE). Dong and Lewbel(2015) show how a derivative of this LATE can be estimated. They use theirTreatment Eect Derivative (TED) to estimate how the LATE would change if theRD threshold changed. We argue that their estimator should be employed in mostRD applications, as a way to assess the stability and hence external validity of RDestimates. Closely related toTED, we dene the Complier Probability Derivative(CPD). Just as TED measures stability of the treatment eect, the CPD measuresstability of the complier population in fuzzy designs. We provide Stata code thatcan be used to easily implement TED and CPD estimation, and apply it to somereal data sets.

Title: Counterfactual Impact Evaluation of Vocational Education in PortugalPresenter: Ricardo Paes Mamede, ISCTE- Instituto Universitario de LisboaCo-author(s): Daniela Cruz, Teresa FernandesAbstract: The Portuguese Secondary School Public System experienced a rapid expansion

of vocational education after 2007, as a result of a policy-induced effort to reducedrop-out rates and improve the employability of secondary school graduates. Inthis paper we use an coarsened exact matching approach to assess the impacts ofthat policy on students’ academic and labour market performance for three co-horts (2008/2009, 2009/2010 and 2010/2011). Our analysis draws on a uniquedataset that merges information collected by public schools with Social Secu-rity data, which allows following the trajectory of students from the year beforeentering high school until 12 months after graduation. The performance of voca-tional education students is compared with that of similar individuals enrolled in

37

scientific-humanistic secondary school programs. We find that being enrolled invocational education has positive impacts both in school performance (transition,graduation, and dropout rates) and in labour market performance (employmentrate, average salary, and average worked months per year). We also find that vo-cational education students have lower chances of proceeding to higher education.These results are robust to variations in the estimation method.

Title: Italian public guarantees to SME: the impact on regional growthPresenter: Guido Pellegrini, Sapienza, Università di RomaCo-author(s): Marusca De CastrisAbstract: The paper evaluates the effects of the Central Guarantee Fund (CGF) on the eco-

nomic growth of the Italian provinces. Does the CGF actually support territo-rial development, taking into account crowding out and spillover effects betweentreated and not treated firms? The challenge of the empirical analysis is to cap-ture macro effects of CGF, when its intervention covered on average only 3% ofcompanies and 0.5% of funding at provincial level. Using different models basedon a “long DID” approach, the results show a positive and statistically signifi-cant, albeit modest, correlation between use of CGF and provincial, controllingfor sectoral differences, dimensional characteristics and several interactions. Ourfindings suggest that CGF has mitigated the negative regional effects of the recenteconomic crisis in Italy.

Solicited Session SOL-14 • Thursday 9, 09:00-10:30 • Room: SP/2

USAGE OF GEOCODED MICRO DATA IN THE ECONOMIC ANALYSISChair: Giuseppe Arbia

Title: Spatial sampling methods with locational errorsPresenter: Maria Michela Dickson, Università di TrentoCo-author(s): Danila FilipponiAbstract: Recently, business registers have been provided with geographical information

about units, making possible to conduct spatial studies on firms. Sometimes thesedata result perturbed respect to their real geographical locations, owing to pri-vacy and technical problems in geo-coding phase. These problems lead to non-sampling errors particularly relevant in many fields of research, among otherseconomy. In the present work, the attention is concentrated on errors about exactlocations of units and particularly on consequences in spatial sampling methods.By means of a simulation study, the aim is to investigate the effect of differentproportions of erroneous location in the selection of units, using different spatialsampling methods. The efficiency of these designs have been evaluated in termsof estimation of population totals and variances between and after the perturba-tion.

Title: Spatial Micro-Econometrics Models with Locational ErrorsPresenter: Diego Giuliani, Università di TrentoCo-author(s): Simonetta Cozzi, Giuseppe EspaAbstract: The state-of-the-art methods and models for the analysis of micro-geographic data

are based on the assumption that the information about the spatial coordinates

38

of the statistical units is completely accurate. In many practical circumstances,however, such information is inevitably affected by locational errors, that can begenerated intentionally by the data producer for privacy protection or can be dueto inaccuracy of the geocoding procedures. Some recent contributions (Arbia etal., 2015; 2016) have shown that the presence of locational errors may have anon-negligible impact on the results. In particular, wrong spatial coordinates canlead to downward bias and increased variance in the estimation of the parametersof spatial econometrics models. This contribution aims at developing a strategyto reduce the bias and produce reliable inference for spatial econometrics modelswith location errors. The validity of the proposed approach is assessed by meansof Monte Carlo experiments and its usability is illustrated by an application toreal data.

Title: Three-Year Survival Probability of Italian Start-up Businesses in HealthcareIndustry: an Empirical Investigation through Logistic Multilevel Modelling

Presenter: Flavio Santi, Sapienza, Università di RomaCo-author(s): Maria Michela Dickson, Diego Giuliani, Davide PiacentinoAbstract: The purpose of this contribution is to provide novel evidence about the main de-

terminants of the short-run survival of pharmaceutical and medical device man-ufacturing start-up firms in Italy. In order to assess both the firm-specific deter-minants and the observed and unobserved regional and contextual characteris-tics, we model the three-year firm survival probability by means of a multilevellogistic framework. The empirical analysis focuses on an internationally com-parable database of the population of firms built up and managed by the ItalianNational Institute of Statistics (ISTAT), in accordance with the procedures sug-gested by OECD and EUROSTAT, which guarantee that data are not affected bythe typical inconsistencies of the National Business Registers and hence providethe true information about firm entries and exits. The size of this dataset andthe high number of regional random effects, however, makes the standard esti-mation techniques of the multilevel logistic model computationally unfeasible.The estimation is then performed by means of the cross-entropy method for noisyoptimization suggested by Bee et al. (2015).

Solicited Session SOL-15 • Thursday 9, 09:00-10:30 • Room: A

STATISTICAL MODELS IN FUNCTIONAL DATA ANALYSISChair: Elvira Romano

Title: Space-time FPCA Algorithm for clustering ofmultidimensional curvesPresenter: Giada Adelfio, Università di PalermoCo-author(s): Francesca Di Salvo, Marcello ChiodiAbstract: In this paper we focus on finding clusters of multidimensional curveswith spatio-

temporal structure, applying a variant of a k-means algorithm basedon the princi-pal component rotation of data. The main advantage of this approachis to combinethe clustering functional analysis of the multidimensional data, withsmoothingmethods based on generalized additive models (GAM), that cope withboth thespatial and the temporal variability, and with functional principal componentsthattakes into account the dependency between the curves

39

Title: Functional data analysis approaches for satellite remote sensing applicationsPresenter: Claire Miller, University of GlasgowCo-author(s): Ruth O’Donnell, Mengyi Gong, Marian ScottAbstract: Functional data analysis provides an attractive approach for investigatingspatiotem-

poral lake images providing efficient dimensionality reduction, appropriatetreat-ment of sparse data and clustering. Specifically, we propose mixed modelfunc-tional principal components analysis (PCA) for bivariate images to reduce di-mensionalityand provide imputations for missing data, and a within-lake func-tionalclustering algorithm to identify within-lake coherence. Temporal coherencecan bedefined as the synchrony between major fluctuations in a set of time serieswithidentification of coherence providing evidence of large scale environmen-tal pressuresand change. Environmental and climatological drivers of coherencewithinand between lakes can then be investigated to explore the drivers of envi-ronmentalchange at a global scale.

Title: Order statistics for spatially dependent functional dataPresenter: Elvira Romano, Seconda Università degli Studi di NapoliCo-author(s): Antonio Balzanella, Rosanna VerdeAbstract: In this paper, we address the problem of getting order statistics for spatiallyde-

pendent functional data by means of depth functions. Given a set of spatially-dependent curves, we generalize a simple criterion for ordering locations fromcenterto outward by considering the spatial variability of the curves. To reach thisaim,we introduce the concept of spatial dispersion function defined on a specificlocationof the space. Then we extend the half region depth for functional data tospatialdispersion functions.

Title: A penalized regression model for functional data with spatial dependencePresenter: Laura Maria Sangalli, Politecnico di MilanoAbstract: This work presents a penalized regression model for the analysis of functional

data with complex dependencies, such as spatially dependent curves and timedependent surfaces, observed over complex domains. The methodology is illus-trated via an application to the study of the annual production of waste in themunicipalities of Venice province.

40

Plenary Session B • Thursday 9, 10:30-11:30 • Room: Aula Magna

PROBABILISTIC INFERENCE FROM BIG AND COMPLEX DATA

by David DunsonDuke University

Chair: Antonio Canale (Università di Torino)Discussant: Pier Cesari Secchi (Politecnico di Milano)

Abstract: Big, high-dimensional and complex data are now routinely collected across fieldsranging broadly from industry to government to the sciences. Although there isan increasingly rich literature on algorithms for analyzing such data, almost allsolutions have taken one of two types of approaches. The first breaks data intosmaller chunks and analyzes these chunks completely separately, ignoring depen-dence in the analysis. The second relies on optimization algorithms to obtain asingle “point” estimate of quantities of interest, typically without any measure ofhow uncertain this estimate is. There remains a disturbing lack of methods forcharacterizing statistical uncertainties in these settings, leading to fundamentalproblems in many applications areas (e.g., the sciences). This talk will providean overview of some recent approaches for scaling up probabilistic and Bayesianinferences to massive scale data (e.g., from computational advertising, genomicsand neurosciences). Often in massive data settings, probability models and com-putational algorithms need to be specifically designed to allow scaling up. I alsodiscuss promising ongoing directions, from both applied and theoretical perspec-tives.

Thursday 9, 12:00-13:00 • Room: Aula Magna

POSTER SPEED SESSION

Title: Non-conjugate Variational Bayes ApproximationAuthor(s): Mauro Bernardi, Erlis RuliAbstract: Recently, the increasing interest on big data is posing many new challenges to

the mainstream of statistics. On the Bayesian side, commonly used stochastic ap-proximation methods, such as MCMC methods do not scale with the sample size.To address this issue, several MCMC algorithms have been proposed which es-sentially reduce the original problem into smaller pieces and then combine themin order to obtain an approximation of the target. On the other hand, deterministicapproximations such as the Laplace method or Variational Bayes (VB) methodsscale with sample size, but they are not without limitations and they rely on strongassumptions. We present a VB algorithm that overcomes the main limitations oftraditional VB algorithms since it does not require block independence or modelconjugacy. The proposed algorithm finds the multivariate skew-t density thatminimises a slightly modified version of the Kullback–Leibler (KL) divergence

41

between this density and the target. We approximate the multivariate integral in-volved in the modified KL divergence with the improved Laplace method of Ruliet al. (2015). With respect to typical VB approximations, the proposed methodis able to capture the non–Gaussianity of the target without imposing indepen-dence constraints among blocks. The usefulness of the method is illustrated bysimulated and real–life datasets.

Title: The Multivariate Fuzzy Skew Student–t distributionAuthor(s): Mauro BernardiAbstract: In the statistical literature, truncated distributions can be used for modelling real

data. In this paper, we define the fuzzy multivariate Skew Student–t distributionthat is an extension of the selection distributions introduced by Arellano–Valleand Azzalini (2006) using fuzzy threshold. Multivariate fuzzy Skew Student–tdistributions naturally arises in the contest of conditional risk measures when theconditioning generating mechanism is defined over a fuzzy set having a proba-bilistic interpretation.

Title: Quality of Educational Services, Institutional Image, Students’ Satisfactionand Loyalty in Higher Education

Author(s): Matilde Bini, Lucio Masserini, Monica PratesiAbstract: In today’s increasingly competitive higher education environment, concepts such

as perceived quality of educational service, institutional image, students’ satisfac-tion and students’ loyalty have become of strategic concern in both public andprivate universities. The objective of this study is to investigate whether the per-ceived Quality of Educational Services and University Image influence students’overall Satisfaction with their university experience, and to examine the possi-ble consequences of these relationships on students’ Loyalty. For this purpose,several hypotheses were formulated and tested using a structural equation model.Data were collected through a web questionnaire handed out to 14,870 studentsenrolled at the University of Pisa.

Title: Estimation of INAR(p) models using bootstrapAuthor(s): Luisa Bisaglia, Margherita GerolimettoAbstract: Recently, there has been a growing interest in studying nonnegative integer-valued

time series and, in particular, time series of counts.However, when the values ofthe time series are small, as in the case of counting processes, the usual linearARMA processes are of limited use for modeling and forecasting purposes. Themost common approach to build an integer-valued autoregressive (INAR) processis based on a probabilistic operation called binomial thinning. While theoreti-cal properties of INAR models with Poisson innovations have been extensivelystudied in the literature, relatively few contributions discuss the developmentof estimation methodology when the distribution of the error terms is differentfrom Poisson. The Poisson distribution, however, has the disadvantage of allow-ing only for equi-dispersion. Thus, unlike the usual applications where the errorterms are Poisson, in order to investigate over and under-dispersion cases, we haveto assume that the error terms follow different integer distributions. When the dis-tribution of the error term is not know, it is not possible calculate the likelihoodand so to estimate the parameters of the model. Other methods, like for instance,the Yule-Walker (YW) method, are not good as the maximum likelihood (ML)estimator. In this work we propose a bootstrap methodology to estimate the un-

42

known parameters of INAR(p) processes. Extensive finite sample Monte Carloexperiments are carried out to compare the performance of bootstrap estimatorwith those of ML and YW estimators.

Title: Effect of internet-based cognitive therapy on children anxiety disorders: resultsfrom a marginal logistic quantile regression

Author(s): Davide BossoliAbstract: Cognitive behaviour therapy (CBT) has been shown to be an effective treatment

for anxiety disorders in children, but few affected patients seek or receive treat-ment. The aim of this study was to assess the relationship between anxiety disor-ders and an internet-delivered CBT, a more accessible alternative to CBT. The dataconsisted of a randomized controlled trial of 93 families with a child aged 8-12years with a principal diagnosis of generalized anxiety disorder. Participants wererandomized to 10 weeks of ICBT with therapist support, or to a waitlist controlcondition. The main response variable was related to difficulties in emotion reg-ulation scale (DERS) and it was measured weekly on a 0-100 scale. Because theresponse variable was bounded, we used logistic quantile regression. This methodprovided more information on the conditional distribution of the response thantraditional mean methods. The dependence induced by repeated measurementswas handled using an extension of generalized estimating equations to quantileregression.

Title: Some mathematical properties of the ROC curveAuthor(s): Camilla Calì, Maria LongobardiAbstract: In our poster we present ROC methodology and analyze the ROC curve. We

describe first the historical background, its relation with signal detection theoryand the further application to radar detection theory, psychophysical researchesand finally to medicine. We start with the description of the so-called “yes-no”tests and introduce the definitions of sensitivity and specificity. The ROC curveis considered as plot of sensitivity (TPR) versus 1-specificity (FPR) consideringall possible values of the cut-off c. Then we analyze the terminology used tointroduce the ROC curve in relation to statistical hypotheses testing, focusingon the definitions of False Positive Rate and False Negative Rate as error of Iand II type, respectively. Finally some mathematical properties of this curve aregiven and studied, especially in relation with stochastic orders. This relation isused to obtain a result about the concavity of the ROC curve when the randomvariables considered are in relationship with particular type of stochastic order.

Title: Machine learning for the estimation of the propensity score: a simulation studyAuthor(s): Massimo Cannas, Bruno ArpinoAbstract: Despite the extensive literature on propensity score (PS) methods there are still

several open questions for their implementation. Based on the results of an exten-sive simulation exercise, we try to address some of these questions and provideguidelines for applicants. The first question we consider is which method shouldbe preferred to estimate the PS. We compare machine learning techniques (MLT)with standard logit models by analyzing the performance of the different PS esti-mators in matching (M) and weighting (W) via MonteCarlo simulations. Second,we profit of the simulation framework to assess the efficacy of several measuresof covariate balance in predicting the quality of the propensity score weightingand matching estimators in terms of ATT bias reduction. With few exception

43

weighting estimators outperform matching estimators in all simulation scenariosin terms of bias reduction. Conditional on M or W random forests, follow bylogit, gave the lower bias while tree methods were competitive only when weight-ing and neural networks and naive bayes only with large data sets. The balancediagnostics with the highest association with the BIAS was the asam with theinclusion of interaction terms but the association was not significantly differentfrom that of classic asam. Less commonly used metrics (auc, ecdf, var ratio)resulted only weakly associated to the bias.

Title: An R package for propensity score matching with clustered dataAuthor(s): Massimo Cannas, Bruno Arpino, Claudio ConversanoAbstract: MatchIt and Matching are two main packages for propensity score analysis but

currently there is not a package handling clustered data. Recently, several ap-proaches to reduce bias due to cluster-level confounders were considered andcompared using Monte Carlo simulations by Arpino and Cannas. These meth-ods exploit the clustered structure of data in two ways: in the estimation of thepropensity score model (through the inclusion of fixed or random effects) or in theimplementation of the matching algorithm. We created R functions implementingthe strategies described in Aprino and Cannas and we illustrate their use throughthe analysis of a clinical data set.

Title: Mapping local social protection data in ItalyAuthor(s): Alessandra Coli, Barbara Pacini, Alessandro Valentini, Silvia VenturiAbstract: People living in close geographic areas can experience different quality of life

standards. This depends on many factors among which the performance of localwelfare systems may play a relevant role. In this paper, we examine Italian offi-cial statistics in order to identify available information on local social protectionactivities. The main objective is to assess if it is possible to convey a completeview of the level and quality of social protection services delivered by local ac-tors. Furthermore, we present some analysis of municipalities’ social expenditureat Nuts 2 and 3 levels to show evidence of disparities among territories.

Title: Indirect inference for nonlinear panel dataAuthor(s): Antonio CosmaAbstract: In this article, we estimate nonlinear dynamic panel models by indirect inference.

We focus in particular on short panels. Indirect inference uses an auxiliary modelto estimate the parameters of the model of interest, that we call the ’true’ model.The parameters of the auxiliary model can be estimated using either the observeddata or data simulated from the true model. Indirect inference estimates are thosevalues of the parameters of the true model such that these two estimates of theparameters of the auxiliary model are as close as possible. We use the simple lin-ear panel autoregression (AR1 model) as auxiliary model. Maximum likelihoodestimation is known to deliver biased estimates of the AR1 even if it were the truedata generating process, but we use these estimates as a criterion to estimate theparameters of the model of interest, as in Gourieroux C., Monfort A., Renault E.(1993). In the dynamic logit case, we compare our estimator with the pseudo con-ditional likelihood one suggested by Bartolucci and Nigro (2012). Monte Carlosimulations show that our method performs well, especially in short panels. Forinstance, with T=3 time periods, the mean square root error on the parameter rela-tive to the autoregressive term, decreases of about 20%, and this for different sizes

44

of N, the number of individuals in the panel. In the dynamic probit case there isno likelihood based estimator equivalent to the one suggested by Bartolucci andNigro. The values of the mean square root error we obtain are comparable to theones we get in the dynamic logit case.

Title: Cluster Analysis of Transactional Data in the Frequency DomainAuthor(s): Ivan Luciano Danesi, Flavio Maria Emanuele Pons, Cristina ReaAbstract: We consider the problem of clusterization and discrimination of stochasticpro-

cesses with an application to transactional data. The analysed dataset is a col-lectionof payment card transactions, referred to a selected number of Point ofSales(opportunely discretized). The aim is to identify homogeneous groups ofPoint ofSales starting by the received payments. We use an approach based on thevarianceprofile. Estimation is carried out in the frequency

Title: Using purchase market behavior to estimate collective well-beingAuthor(s): Lorenzo Gabrielli, Giovanni Riccardi, Luca PappalardoAbstract: Several works in literature showed that the purchase behavior of an individual

is a reliable proxy for individual wealth and well-being. Official statistics insti-tutes, indeed, base their well-being indicators on surveys regarding the purchaseexpenditure of individuals within a given territory. These surveys, however, areoften not up-to-date, expensive to carry out, and only cover a small sample ofpopulation. We propose to use massive retail market data to construct, in a fastand automatic way, indicators for collective well-being. Starting from purchasedata of individuals during one year, we select all the products for the purchasecategories in the "food and beverage" sector, according to the directions sugges-tions by Italian National Institute of Statistics (ISTAT). We then construct severalindividual variables by comparing the average purchase expenditure of each indi-vidual for a given purchase category with the global purchase expenditure of thesame purchase category. We summarize the obtained individual variables into anindividual well-being indicator, and aggregate the indicators of individuals livingin the same territory into a collective well-being indicator. We compare the ob-tained collective indicator with several well-being indicators produced by officialstatistics institutes, showing that our approach produces a fast and reliable proxyfor collective well-being and socio-economic development.

Title: The Bifactor Item Response Theory Model for the analysis of repeated mea-surements. An application to the measurement of Italian households’ well-being

Author(s): Francesca Giambona, Mariano Porcu, Isabella SulisAbstract: Repeated measures on a scale of items are used in longitudinal design surveys for

studying individuals’ change over time of a latent trait of interest. The BifactorItem Response Theory model specifies individuals’ item responses as a functionof a general and a specific factor. The general factor measures the main latent traitof interest (e.g. household well-being), whereas each specific factor captures theresidual variability shared by the cluster of items which refer to the same waveand which contribute to the definition of the specific aspect of the latent trait(e.g. the variation of household well-being at each time). Within this approach,the dependency between repeated measurements on the same unit observed inlongitudinal analysis is specifically taken into account by introducing a number ofspecific factors equal to number of waves observed. An application to longitudinal

45

Italian SILC data is discussed.

Title: A Bayesian short-term strategy for site specific wind potential assessmentAuthor(s): Antonio Lepore, Pasquale Erto, Biagio Palumbo, Massimo LeporeAbstract: In the last decade, wind energy has proven to be one of the most competitive

and fastest growing sources of renewable energy. As fast as possible evaluationof site-specific wind potential is desirable in order to reduce analysis time andeventually activate potential investments on wind energy sources. The economicprofitability of a candidate site is strongly influenced not only by the wind speed,which is needed to define the turbine type to be installed, but also by its direction,which represents a dominant parameter in the wind-farm layout design. However,wind potential assessment from short (e.g., one-month) sample can be severelypoor. However, as encouragingly shown by applications to real anemometric datafrom Southern Italian candidate sites, a Bayesian proposed approach can copewith this problem of data analysis. The attained results show the Mean SquareError of the Bayesian estimates carried out from a one-month sample is lowerthan the Maximum Likelihood one. Furthermore, a new engineering error index,namely Mean Power Integral Deviation (MPID), is proposed.

Title: Distributed-Lag Structural Equation Modelling: An Application to Impact As-sessment of Research Activity on European Agriculture

Author(s): Alessandro MagriniAbstract: Structural Equation Modelling is a class of statistical models typically employed

to analyse the dependence relationships among a set of variables. We define an ex-tension of the class where variables are related by distributed-lag linear regressionmodels, in order to account for temporal delays in the dependence relationshipsamong the variables. Our proposal is applied to impact assessment of researchactivity on European Agriculture.

Title: A multivariate circular-linear hidden Markov model for site-specific assessmentof wind predictions by an atmospheric simulation system

Author(s): Gianluca Mastrantonio, Alessio Pollice, Francesca FedeleAbstract: Winds from North-West quadrants and lack of precipitation are known to lead to

an increase of PM10 concentrations in the city of Taranto. In 2012 the ApuliaGovernment prescribed a reduction of industrial emissions by 10% every timesuch meteorological conditions are forecasted 72 hours in advance. Wind predic-tion is addressed using the Weather Research and Forecasting (WRF) atmosphericsimulation system by the Regional Environmental Protection Agency (ARPA Puglia).We investigate the ability of the WRF system to properly predict the local windspeed and direction allowing different performances for unknown weather regimes.Observed and WRF-predicted wind speed and direction at a relevant location arejointly modeled as a 4-dimensional time series with a finite number of states (windregimes) characterized by homogeneous distributional behavior. Observed andsimulated wind data are made of two circular (direction) and two linear (speed)variables, then the 4-dimensional time series is jointly modeled by a mixture ofprojected-skew normal distributions with time-independent states, where the tem-poral evolution of the state membership follows a first order Markov process. Pa-rameter estimates are obtained by a Bayesian MCMC-based method and resultsprovide useful insights on wind regimes corresponding to different performancesof WRF predictions.

46

Title: Bayesian networks for supporting the digitization process in Italian schoolsAuthor(s): Flaminia Musella, Stefania Capogna, Maria Chiara De AngelisAbstract: The wide but unbalanced spread of ICT determines a digital divide and instanti-

ates a discussion about digital and media literacy that are necessary, nowadays,for understanding texts, formats and connection templates characterizing severalcommunication channels in the knowledge society. The digital society requires tothe school to strategically approach new teaching methods and practices aiming atcontinuous skill updating and at digital competencies training. In opposition withthe most popular theories of school immobility, the hypothesis of this researchis the occurrence of workshops of research and educational novelty in Italianschools. The main goal of this research is to understand such a kind of inno-vations: firstly a qualitative approach has been adopted to catch the improvementmargins and then a survey has been carried out by a structured questionnaire foranalysing the expectations and the perceptions of teachers. Whilst the analysisof gathered data provides a snapshot of the current situation, a statistical modelbased on Bayesian networks will enable to recognize those social innovations tobe considered as best practices and will help in selecting the key strategies, byperforming probabilistic scenarios, to support the policies on the school systemdigitization.

Title: Statistical approach in aerospace industry innovationAuthor(s): Biagio Palumbo, Pasquale Erto, Flaviana Tagliaferri, Gaetano De Chiara, Roberto

Marrone, Claudio Leone, Silvio GennaAbstract: Engineering and statistical knowledge integration is a key factor for innovation

in aerospace industry. In this industrial field, the laser drilling is one of the mostimportant process for making high-quality engine components such as blades orcombustion chambers. A synergic collaboration and partnership between aca-demic statisticians and industrial practitioners has allowed to highlight the strate-gic role that a systematic statistical approach to planning for a designed industrialplays in technological process innovation for industrial competitive advantage.The team approach is the real driving force for pre-experimental activities. More-over, it enables the integration of engineering and statistical knowledge and cat-alyzes process innovation. Besides, it allows to run a virtuous cycle of sequentiallearning for continuous improvement. Several applicative examples concerningthe effective use of statistical methodologies that enabled the industrial partner tosucceed on the worldwide market are presented.

Title: Ship fuel consumption control and engineering approach to fault-detectionAuthor(s): Biagio Palumbo, Pasquale Erto, Antonio Lepore, Luigi Vitiello, Christian Capezza,

Dario Bocchetti, Andrea D’Ambra, Biagio AntonelliAbstract: The competitiveness of the shipping market and the Kyoto Protocol have been

urging shipping companies to pay increasing attention to ship energy efficiencyimprovement and CO2 emission reduction. At the same time, new monitoringdata acquisition technologies on modern ships have brought to a navigation dataoverload that need to be correctly handled by shipping companies to make deci-sions. For this purpose, a novel approach based on the Partial Least-Squares (PLS)regression is introduced for ship fuel consumption monitoring at each voyage.PLS regression naturally deals with complex, correlated and noisy data that usu-ally do not fulfil classical regression model requirements. Moreover, via the con-tribution plot and the Hotelling control chart for scores , an engineering approach

47

to fault-detection is presented in order to give practical indications on anomaliesand/or further navigation variables that need to be technologically investigated.

Title: Small area model-based direct estimator for spatial dataAuthor(s): Alessandra Petrucci, Chiara Bocci, Emilia RoccoAbstract: The interest in spatial data analysis is increased in every area of statistical re-

search. Geographical information is frequently available in many areas of ob-servational sciences, and the use of specific techniques of spatial data analysiscan improve our understanding of the studied phenomena. Geoadditive modelsanswer this issue by analysing the spatial distribution of a study variable whileaccounting for possible covariates effects through a linear mixed model represen-tation. In addition, if we are interested in producing small area estimates for askewed variable, small area estimation (SAE) methods based on log-transformedmodels are required. Under the mixed model representation, we combine ourtwo goals in a geoadditive small area log-transformed model and we extend themodel-based direct estimator for skewed data, recently introduced in literature, inorder to be able to account for data that are both skewed and with a spatial trend.A real application of the suggested approach is presented. Moreover, in order toestimate the mean squared error of the suggested small area estimator, the recentbias-robust MSE estimator for pseudo-linear estimators is used. Finally, we dis-cuss a possible imputation approach for the case in which the information aboutthe population units’ location is incomplete.

Title: Dynamic Quantile Lasso RegressionAuthor(s): Fabrizio Poggioni, Lea Petrella, Mauro BernardiAbstract: Quantile regression has been becoming a relevant and powerfult echnique to study

the whole conditional distribution of a response variable without relying on strongassumptions about the underlying data generating process. Furthermore, quantileregression has been effectively used in many real applications, providing a rep-resentation of the relation between the response variable and the covariates, thatovercomes traditional mean regression. In this paper, we consider a quantile re-gression model in which the regression coefficients are assumed to evolve overtime, following a stationary stochastic process. Furthermore, since homoskedasticquantile regression models result in location shifts of the regression hyperplane,we extend the time–varying parameter model to allow for heteroskedastic inno-vations. A dynamic version of the adaptive–Lasso penalty is then introduced toforce the dynamic evolution of non relevant parameters to shrink towards zero. Asimulation study is carried out to illustrate the model performances.

Title: Estimating dependence within neuropsychological models for designing riskprofiles of decision-makers.

Author(s): Aleksandar PramovAbstract: The main goal of this contribution is to study the dependence among measures

of spatial working memory and a set of clinical and neuropsychological variablesfrom a statistical perspective. In particular, by employing a statistical model fordependence, we aim to investigate neuropsychological models assessed by thepsychological literature through a systematic review related to Obsessive Com-pulsive Disorders. We extend the existing analyses by considering a larger set ofvariables (both clinical and neuropsychological) and provide a robust statisticalevaluation of several theory-driven models about the effect of OCD on decision-

48

making. Our dataset consists of a large set of clinical and neuropsychologicalvariables (binary, categorical and continuous) collected on 160 subjects (80 withOCD and 80 control) which we use to build our statistical model. We assess ro-bustness of the neuropsychological models by comparing a hypothesis-driven ap-proach mainly based on Structural Equation Models with a data-driven approachbased on Bayesian Networks. A main methodological challenge concerns thecomparison different model selection strategies and indicators. While there is no“unique” measure model fit, we give a systematic review of said measures anddiscuss their adequateness for the task at hand.

Title: Confidence intervals for a partially identified parameter with bounds estimatedby the minimum and the maximum of two correlated and normally distributedstatistics

Author(s): Aleksandar PramovAbstract: We study the problem of constructing a Confidence Interval (CI) for a partially

identified parameter, when the lower and upper bounds of the identification re-gion are the minimum and the maximum of two functions of the parameters ofthe underlying distribution. The motivation comes from a problem of comparingthe relative accuracy of two binary diagnostic tests in presence of noncompliancewith a subsequent nonignorably missing disease indicator. Thus, some parame-ters of the observed data distribution are many-to-one functions of the originalparameters. Relevant measures of relative accuracy, such as the relative true pos-itive rate or relative false positive rate may be only partially identified [1], withthe bounds of the identification region estimated by the min and the max of twoasymptotically jointly normal estimators. We present some existing methods forpartial identification and propose two extensions to deal with the case of usingsuch a particular type of non-differentiable functions as bounds. We providesome analytic results which deliver asymptotically the desired coverage for theCI pointwise. We also discuss a Bayesian approach, and compare our results tosome bootstrap alternatives for the construction of CIs.

Title: Semantic Knowledge Detection in Open-ended QuestionnaireAuthor(s): Ilaria Primerano, Giuseppe GiordanoAbstract: In this work we focus on the definition and analysis of a text-graph structure as

a device for the interpretation of a theoretical concept. Concepts are multidimen-sional in their nature and in surveys they can be investigated by the use of open-ended questionnaires.We consider as many open-ended questions as the numberof a priori defined dimension of a generic concept under investigation. The set ofanswers to any question become a bag of words provided from all respondents. Inthe framework of network analysis, it is a graph and constitutes a layer of a mul-tiplex network structure. The use of analytic tools of Social Network Analysisallows to explore and interpret such complex data structure. The network repre-sentation provides a simple way to visualize the connections among texts, whiletheir statistical analysis may highlight a useful description of a concept throughthe investigation of such relations. Any pairs of words are connected whereasthey are jointly used in the same answers. The number of co-occurrence can beused as a weight of the links and can be normalized according to different criteria(e.g. Term Frequency or TF-IDF index). The main goal is to identify patterns ofclosely related words and the semantic core network of all the words for all thelayers jointly considered. In such a way, we will discover the latent meaning in a

49

coherently way with the definition of concept. A simulation study and a real casestudy will be used to assess the validity of the proposed approach.

Title: A joint approach to the analysis of time-varying affiliation networksAuthor(s): Giancarlo Ragozini, Daniela D’Ambrosio, Marco SerinoAbstract: In this paper, we analyze a time-varying affiliation network formed through var-

ious co-productions that involve the theatre companies located in Campania re-gion, Italy, over four theatre seasons (2011-2014). Affiliation (two-mode) net-works (Wasserman and Faust, 1994) are characterized by a set of actors (the co-producing organizations) and a set of events or affiliations (the co-productions).Therefore, two or more theatres can be seen as affiliated when jointly collaborat-ing to one stage co-production or even more, during one or more seasons. Weshow how this complex data frame can be analyzed by multi-way analysis (Kroo-nenberg, 2008), proposing a joint approach that combines multidimensional anal-ysis and network-analytic techniques. We make use of Multiple CorrespondenceAnalysis (MCA) (Blasius and Greenacre, 1994, 2006) – also applying the com-plete disjunctive coding in Greenacre’s doubling perspective (Greenacre, 1984) –and consider the affiliation matrix as a two-way case-by-variable matrix, thanks tothe similarity between the structures of the two matrices (D’Esposito, De Stefanoand Ragozini, 2014). This technique permits us assess the relational patterns ofnetwork units, along with their categorical attributes. MCA is performed as a suit-able factorial method within Multiple Factor Analysis (MFA) (Escofier and Pagès,1998; Pagès, 2002), a technique apt to handle multiple data tables. Thereby, MFAis used to analyse the variation of relational patterns over time (Ragozini, De Ste-fano and D’Esposito, 2015). In addition, we adopt generalized blockmodeling fortwo-mode networks (Doreian, Batajeli, and Ferligoj, 2005) in order to conduct adetailed analysis of the structural patterns of this network.

Title: Non-parametric estimators for estimating bivariate survival function under ran-domly censored and truncated data

Author(s): Marialuisa Restaino, Hongsheng Dai, Huan WangAbstract: In bivariate survival analysis it is common to dealt with incomplete information

of the data, due to random censoring and random truncation. Such kind of dataoccurs in many research areas, such as medicine, economics, insurance and so-cial sciences. Most existing research papers focused on bivariate survival analysiswhen components are either censoring or truncation or where one component iscensored and truncated, but the other one is fully observed. Bivariate survivalfunction estimation when both components are censored and truncated has re-ceived considerable attention recently. These methods, however, used an iterativecomputing method which is computationally heavy. Some authors proposed anestimator based on a polar coordinate transformation, which does not require it-erative calculations and its large sample properties are established. Starting fromtheir paper, we extend their methods to a class of estimators, based on differentdata transformations. In particular assuming that the components are both randomtruncation and random censoring, we propose a class of nonparametric estimatorsfor the bivariate survival function. In practice, the pair of random variables un-der consideration may have certain parametric relationship. The proposed classof nonparametric estimators uses such parametric information via a data transfor-mation approach and thus provides more accurate estimates than existing methodswithout using such information. The large sample properties of the new class of

50

estimators and a general guidance of how to find a good data transformation aregiven. The proposed method is also justified via a simulation study and an appli-cation on an economic data set.

Title: Bayesian M-quantile regression in Small Area EstimationAuthor(s): Giovanni RiccardiAbstract: Mixed effects models are widely used in SAE in order to estimate the effect for

a particular area. However, such models depend on parametric and distributionalassumptions as well as requiring specification of the random part of the model.An alternative approach to this regression estimation is the M-quantile regression.Chambers and Tzavidis (2006) applied the M-quantile regression as basis of theirsmall area estimation method. A number of papers on M-quantile regression ap-plied to SAE has been published since 2006 (Tzavidis et al., 2010; Fabrizi et al.,2014), concerning theoretical developments, extension to non-linear models andvarious applications. But it has not been proposed any Bayesian approach. Themain objective of this work is to propose a Bayesian M-quantile regression modelspecifying an asymmetric likelihood function based on Generalized AsymmetricLeast Informative distribution (Bianchi et al., 2015). Furthermore, the BayesianM-quantile regression will be applied to SAE in order to obtain small area esti-mators and their standard errors.

Title: Optimal B-robust posterior distributions for operational riskAuthor(s): Erlis Ruli, Ivan Luciano Danesi, Fabio Piacenza, Laura VenturaAbstract: One of the aims of operational risk modelling is to generate sound and reliable

quantifications of the risk exposure, including a level of volatility that is consistentwith the changes of the risk profile. One way for assuring this is by means ofrobust procedures, such as Optimal B-Robust estimating equations. In bankingpractice more than one dataset should be incorporated in the risk modelling anda coherent way to proceed to such a data integration is via Bayesian procedures.However, Bayesian inference via estimating equations in general is problematicsince the likelihood function is not available. We illustrate that this issue canbe dealt with using approximate Bayesian computation methods with the robustestimating function as a summary of the data. The method is illustrated by a realdataset.

Title: Estimating income of immigrant communities in Italy using small area estima-tion methods

Author(s): Francesco Schirripa Spagnolo, Nicola Salvati, Antonella D’AgostinoAbstract: In the last decades, Italy has experienced huge migration flows. Studying eco-

nomic conditions of immigrant communities is a crucial step for implementingpolicies targeted on the most disadvantaged groups. The small area estimationmethodology has a wide range of application in this framework. Indeed, immi-grant communities are characterized by few households. As a consequence, weapply the SAE methods in order to compute the mean of the household income,which allows us to improve the information based on a few units and achieve moreaccurate and reliable estimates than the direct estimates. Furthermore, we test thepotentialities of the “small area estimators” (in particular we compute estimatesbased on M-quantile regression model, which is robust method) in a field wherethey have not yet been used. The empirical analysis is based on the Survey onIncome and Living Conditions of Households with Foreign People conducted in

51

2009 by ISTAT and Population Census (2011). Obviously, the lack of informationon non-resident immigrants is a limitation when aiming at understanding the eco-nomic disparities, but it does not make the SAE methods any less suitable in thisframework. Therefore, this analysis is a first empirical evidence and may provehelpful for future research.

Title: The Switching Skew–GARCH ModelAuthor(s): Matteo Soscia, Mauro Bernardi, Lea PetrellaAbstract: The purpose of the paper is to extend the basic SN–SGARCH model introduced

by De Luca andLo Perfido (2004) in order to include econometric features whichare more compliant to financialstylised facts. The basic SN–SGARCH model in-cludes a Skew Normal innovation term as a resultof a conditioning mechanismover the set of positive or negative returns of an exogenous market.Despite itsflexibility, the SN–SGARCH model is not able to reproduce the high volatilitypeakssometimes observed in financial markets because of the standard GARCHdynamics which is characterisedby a slow response to big shocks. We generalisethe SN–SGARCH model by letting themodel parameters to depend also on a la-tent Markovian process. Asymptotic stationary conditionsare derived for both theSN–SGARCH and the MS–SN–SGARCH models along with thetheoretical firstfour conditional and unconditional moments.

Title: Inference on a non-homogeneous Gompertz process with jumps as model oftumor dynamics

Author(s): Serena Spina, Virginia Giorno, Patricia Román-Román, Francisco Torres-RuizAbstract: We propose an inhomogeneous stochastic process with jumps based on the Gom-

pertz diffusion to describe the evolution of a solid tumor subject to an intermittenttherapeutic program. Each jumps represents an application of a therapy that shiftsthe cancer mass to a return state. The intermittent treatment leads to a reductionin tumor size, but produces also an increase in the growth rate represented inthemodel by the inclusion of time functions depending on the cycle of applicationofthe therapy. To find a compromise between these two aspects a strategy to selectthe inter-jump intervals is proposed and an estimation procedure of the model isprovided, supported by several simulations illustrating the validity of the proposedprocedure.

Title: Combining multiple frequencies in multivariate volatility forecastingAuthor(s): Giuseppe Storti, Alessandra Amendola, Vincenzo CandilaAbstract: In a multivariate volatility framework, several options are available to estimate

the conditional covariance matrix of returns. Some models, like the multivariateGARCH (MGARCH) ones, rely on daily returns while others exploit the addi-tional information provided by the analysis of intra-daily prices, like the realizedcovariance (RC) specifications. An additional source of uncertainty is related tothe choice of the frequency at which the intradaily returns, used to construct theRC matrices, are observed. In this paper our interest is in analyzing the impactof these two sources of uncertainty on volatility prediction. In particular, we in-vestigate the profitability of a prediction strategy based on combining forecastscoming from different model structures that are estimated using information atvarious frequencies. In order to illustrate the benefits of our approach we carryout an extensive application to portfolio allocation for a panel of U.S. stocks.

Title: An Enhanced Measure of Well-being through Structural Equation Modeling:

52

a Cross-Country ApproachAuthor(s): Daniele Toninelli, Michela CamelettiAbstract: The measurement of well-being in a country is not an easy task, despite being

very important. Previous research has already proven that quality of life is a mul-tifaceted topic and its measure cannot rely anymore on one index only (GDP)or on few economic indicators. It rather involves several aspects of human life.Furthermore, the big availability of official and survey data represents a great po-tential for the implementation of new statistical methods aimed at summarizingdifferent indicators into a multidimensional measure of the phenomenon. Themain purpose of our work is to apply the Structural Equation Modelling methodto a wide group of variables belonging to three main domains (economic, socio-demographic, environmental). Our research is meant as a cross-country and lon-gitudinal study, allowing us to evaluate at what extent our findings can be gener-alized and if the relative importance of the selected variables is varying over timeand/or over countries. Our work can help in obtaining a more accurate method tomeasure well-being and in driving policy makers in order to improve the citizens’quality of life.

Title: Statistics for knowledge improvement of an innovative manufacturing processand quality cost management

Author(s): Amalia Vanacore, Biagio Palumbo, Francesco Del Re, Pasquale Corrado, MariaRosaria Lanza, Giuseppe La Sala, Mimmo Mastrovita

Abstract: The activities developed in collaboration with MBDA have followed two differ-ent research lines. In the first line a framework of statistical methods useful in theindustrial qualification phase of an innovative process based on Additive Manu-facturing technologies has been proposed. This technology has been recognizedas the next chapter in the industrial revolution. Exactly in the qualification processof a new industrial technology, when is inadequate the practitioner’s experience,statistics play a strategic role to help management in making-decision by planningexperimental activities, analyzing statistical results and catalyzing technologicalinterpretations. The second research line aims at providing a suitable statisti-cal quality control methodology to monitor the non conformance management(NCM) process applied to a MBDA dedicated production line. The NCM processdata are highly autocorrelated and cannot be assumed homogeneous because ofthe significantly different impact on non-quality costs. A demerit system based ona cost severity weighting scheme has been ad hoc developed and a suitable statis-tical process control methodology has been adopted to handle the autocorrelationstructure in NCM process data.

Title: Statistics for Safety and Ergonomics in DesignAuthor(s): Amalia Vanacore, Antonio Lanzotti, Chiara Percuoco, Agostino Capasso, Fabio

Liccardo, Bonaventura VitoloAbstract: The research aims at developing robust assessment and design strategies to sup-

port industrial engineers in the selection of optimal solutions for safety and er-gonomics. It is realized through a successful integration of knowledge in experi-mental statistics, biomechanical modelling and advanced engineering design. Therobust assessment and design strategies have been applied to the context of air-craft and automotive seat design, respectively. They rely on both physical and VRsimulated experiments. In the former case, seat comfort is assessed via subjectiveperceptions and postural responses to seat exposures measured via interface pres-

53

sure maps; whereas in simulated experiments, ergonomic evaluations are based onpostural indexes based on joint angles. The adoption of the proposed strategieshas provided interesting results in deepening the knowledge on seat ergonomicswith focus on three main critical aspects: 1) differences in postural responses toseated exposures; 2) the impact of gender-based postural differences on objec-tive measures of seat discomfort; 3) the usefulness of postural measurements (i.e.seat-interface pressure and joint angles) in finding significant differences in seatdesigns across different target populations of users.

Title: Modelling transition probabilities in a flexible hierarchical logit framework:evidence from the Italian labour market

Author(s): Luca Zanin, Raffaella CalabreseAbstract: The aim of the study is to estimate the transition probabilities in the Italian labour

market with a focus on the joint interaction of the age of the population and theeconomic well-being. The estimation of the probabilities of transition has beencarried out using a pooled set of longitudinal data from the Labour Force Surveycollected by the Italian Institute for National Statistics. The dataset covers the 10year period from 2004 to 2013. We compare different occupational states of malesand females over twelve months using a Markovian modelling framework of or-der one. We propose using a logit model with a hierarchically structured responseembedded in a generalised additive model framework. This flexible estimationallows relaxing assumptions on the relationship between response and predictorand simultaneously deals with the necessary estimation of the combined effect ofage of individuals and gross domestic product per capita using a tensor productinteraction. In general, we observe spatial heterogeneity in the labour market withbetter opportunities in the north than the south of Italy. In particular, we find cru-cial differences in the magnitude of the risks of job loss and difficulties in findinga new job by age cohorts, especially among the youngest and the oldest individ-uals living in the areas with low economic well-being. We conclude the analysiswith a discussion of possible uses of the estimated probabilities to improve thedecision-making of policy-makers and practitioners (such as banks and insurancecompanies).

Specialized Session SPE-06 • Thursday 9, 14:45-16:15 • Room: V. Foa

SPATIAL ANALYSES IN DEMOGRAPHYChair: Oliviero Casacchia; Discussant: Alessandra Petrucci

Title: Measuring residential segregation with spatial indices: an appraisal and appli-cations for the metropolitan area of Rome

Presenter: Frank Heins, Irpps-CnrCo-author(s): Federico Benassi, Fabio Lipizzi, Evelina PaluzziAbstract: Over the last decades there have been important methodological advances in mea-

suring residential segregation, especially concerning spatial indices. After a dis-cussion of the fundamental concepts some of the numerous indices are introduced.The contribution is based on data on the number of foreign citizens resident inthe census enumeration area that form metropolitan area of Rome. Data refer tothe population censuses 2001 and 2011. The focus is on indices that consider

54

the spatial location of the single enumeration area. Applying the indices to themetropolitan area of Rome serves as a test of the practical and potential useful-ness of the proposed measures and their possible interpretation. We will showthat few indices suffice to describe in an exhaustive way the segregation of for-eign communities.

Title: Immigrants’ settlement patterns in the city of Naples - I modelli insediativi degliimmigrati nella città di Napoli

Presenter: Angelo Mazza, Università di CataniaCo-author(s): Giuseppe Gabrielli, Salvatore StrozzaAbstract: Residential segregation is the outcome of both economic inhomogeneitieswithin

the urban space and attraction among individuals sharing the same ethnicity. Herewe focus on the settlement patterns originated by different groups of immi-grantsin the city of Naples. We use the inhomogeneous L-function for measuringsegre-gation due to spatial attraction, while adjusting for the effects of inhomogene-ity.Monte Carlo simulations have been used to build confidence envelopes for thenullhypothesis of absence of spatial attraction. All nationalities exhibited signifi-cant spatial attraction at all considered distances, except for Romania and Poland.However, spatial attraction resulted much stronger for immigrants from Pakistan,China, and Sri Lanka.

Title: Native Immigration and Pull Factor Evolution in Italy: a Spatial ApproachPresenter: Luisa Natale, Università degli Studi di Cassino e del Lazio MeriodionaleAbstract: In Italy several small and medium towns - not belonging to metropolitan areas -

cold be seen as weak from both the demographic and economic point of view. Inparticular, the dynamic and level of migration could be seen as n important signalof socio-economic factors. Our aim is to examine the pull force decline (in termsof immigration rate) for four Italian regions at municipatilites level. Economicand demographic crisis go side-by-side with the abadonment of small towns bythe Italian population and the arrival of the foreign immigrants. Our study con-siders variables take into account the "suburbanity" of the areas with their neigh-bourhoods to bring into better focus the existence of a spatial link between im-migration in one area and the bordering areas. In the paper some technical toolstypical of the Exploratory Spatial Data Analysis (ESDA) are used. Our methodimplies different spatial models that follow classic econometric approach.

Specialized Session SPE-07 • Thursday 9, 14:45-16:15 • Room: G. De Rosa

RECENT DEVELOPMENTS IN VOLATILITY MODELINGChair: Giampiero Gallo; Discussant: Peter Boswijk

Title: Dynamic Model Averaging for Quantile RegressionPresenter: Roberto Casarin, University Ca’ Foscari di VeneziaCo-author(s): Mauro Bernardi, Bertrand Maillet, Lea PetrellaAbstract: We propose a general dynamic model averaging (DMA) approach based on Markov-

Chain Monte Carlo for the sequential combination and estimation of quantile re-gression models with time-varying parameters. The efficiency and the effective-ness of the proposed DMA approach and MCMC algorithm are shown through

55

simulation studies and applications to macro-economics and finance.

Title: Testing volatility: consistency of bootstrap testing for a parameter on the bound-ary of the parameter space

Presenter: Anders Rahbek, University of CopenhagenAbstract: It is well-known that with a parameter on the boundary of the parameter space,

such as in the classic cases of testing for a zero location parameter or no ARCHeffects, the classic nonparametric bootstrap, based on unrestricted parameter esti-mates, leads to inconsistent testing. In contrast, we show here that for the twoaforementioned cases a nonparametric bootstrap test based on parameter esti-mates obtained under the null, referred to as restricted bootstrap, is indeed con-sistent. While the restricted bootstrap is simple to implement in practice, noveltheoretical arguments are required in order to establish consistency. In particu-lar, since the bootstrap is analyzed both under the null hypothesis and under thealternative, non-standard asymptotic expansions are required to deal with param-eters on the boundary. Detailed proofs of the asymptotic validity of the restrictedbootstrap are given and, for the leading case of testing for no ARCH, a MonteCarlo study demonstrates that the bootstrap quasi-likelihood ratio statistic per-forms extremely well in terms of empirical size and power for even remarkablysmall samples, outperforming both the standard Lagrange multiplier test and theasymptotic quasi-likelihood ratio test.

Title: Asymmetric Stochastic Volatility Models: Properties and EstimationPresenter: Esther Ruiz, Universidad Carlos III de MadridCo-author(s): Veronika Czellar, Xiuping Mao, Helena VeigaAbstract: In this paper, we derive the statistical properties of a general family of Stochastic

Volatility (SV) models with leverage effect which capture the dynamic evolutionof asymmetric volatility in financial returns. We provide analytical expressions ofmoments and autocorrelations of power-transformed absolute returns. Moreover,we analyze the finite sample performance of an Approximate Bayesian Compu-tation (ABC) filter-based Maximum Likelihood (ML) estimator, a technique sim-ilar to indirect inference as it requires simulation of pseudo-observations whichare weighted according to their distance to the true observations. We show thatthe ABC filter-based ML estimator does a remarkably good job in estimating theparameters of a very general specification of the log-volatility with standardizedreturns following the Generalized Error Distribution (GED). The results are illus-trated by analyzing a series of daily S&P500 returns.

Specialized Session SPE-08 • Thursday 9, 14:45-16:15 • Room: SP/1

ADVANCES IN ORDINAL CONTINGENCY TABLE ANALYSISChair: Maurizio Carpita; Discussant: Francesco Palumbo

Title: Dimensionality reduction methods for contingency tables with ordinal variablesPresenter: Luigi D’Ambra, Università di Napoli Federico IICo-author(s): Pietro Amenta, Antonello D’AmbraAbstract: Correspondence analysis is a widely used tool for obtaining a graphical repre-

sentation of the interdependence between the rows and columns of a contingency

56

table, by using a dimensionality reduction of the spaces. The maximum infor-mation regarding the association between the two categorical variables is thenvisualized allowing to understand its nature. Several extensions of this methodtake directly into account the possible ordinal structure of the variables by usingdifferent dimensionality reduction tools. Aim of this paper is to present an uni-fied theoretical framework of several methods of correspondence analysis withordinal variables.

Title: Modelling Trends in Ordered Three-Way Non-Symmetrical CorrespondenceAnalysis

Presenter: Rosaria Lombardo, Seconda Università di NapoliCo-author(s): Pieter Kroonenberg, Eric BehAbstract: This paper presents three-way polynomial non-symmetrical correspondence anal-

ysis as a method to analyse associations in three-way contingency tables thatare constructed from ordinal variables with a dependence relationship. Histor-ically, non-symmetrical three-way correspondence analysis (Lombardo, Carlierand D’Ambra, 1996; Kroonenberg, 2008, Chap. 17) has been used for this pur-pose without regard to the ordinal structure of the variables, which may, however,be modelled via orthogonal polynomials. Such polynomials constitute an alter-nate orthogonal basis for modelling interactions in three-way contingency tables(see, also, Beh and Davy, 1998).

Title: Using Collapsing and Multiple Comparisons to Detect Association in Two WayContingency Tables

Presenter: Marco Riani, Università di ParmaCo-author(s): Spyros ArsenisAbstract: In order to test the presence of association between two categorical variables we

can use the χ2 test, the G2-test or generalizations of the Fisher’s exact test. Theχ2 and G2-test can be applied to two way tables of any size and their samplingdistribution is approximated, under the null hypothesis of independence, by χ2

distributions. On the other hand, Fisher’s exact test is defined for 2× 2 tables andenables to compute exact p-values. In this paper we analyze the properties of aprocedure which collapses the original contingency table in 2 × 2 tables and toeach of them applies the Fisher’s exact test. The result of the procedure enables usto highlight the contribution to the association of each single entry of the originaltwo way table. We compare this approach with the standard one which is basedon the inertia decomposition.

Specialized Session SPE-09 • Thursday 9, 14:45-16:15 • Room: SP/2

STATISTICAL MODELS FOR DIRECTIONAL AND CIRCULAR DATAChair: Antonello Maruotti; Discussant: Francesco Lagona

Title: The WeiSSVM: a tractable, parsimonious and flexible model for cylindricaldata

Presenter: Christophe Ley, Ghent UniversityAbstract: Cylindrical data appear in a plethora of fields such as climatology, oceanogra-

phy, studies of animal behavior and environmental sciences. In order to perform

57

meaningful statistical analyses, it is thus of paramount importance to possesscylindrical distributions that (i) are able to model very diverse data shapes, but(ii) remainof a tractable form. In this paper, we present one such distribution,the WeiSSVM, obtained by combining the sine-skewed von Mises distribution(circular part) withthe Weibull distribution (linear part).

Title: The multivariate projected-skew normal distribution: Bayesian estimation anda hidden Markov model application

Presenter: Gianluca Mastrantonio, Politecnico di TorinoAbstract: Often circular variables are recorded along with linear ones, e.g. wind direction

and speed, and hence a joint modeling of mixed type variables is needed. Thecircular and linear variables live on different spaces and there is not an obviousway to define multivariate circular-linear distributions. Up to now there are fewproposals and among all of them, the most interesting are capable to model onlycylindrical data.In this work we propose a flexible distribution for circular-linearvariables, based on the projected and the skew normal and we show how pa-rameters estimate is easy within a Bayesian framework. The proposal is used asemission distribution in a hidden Markov model applied to an animal movementdataset with multivariate circular and linear variables.Often circular variables arerecorded along with linear ones, e.g. wind direction and speed, and hence a jointmodeling of mixed type variables is needed. The circular and linear variables liveon different spaces and there is not an obvious way to define multivariate circular-linear distributions. Up to now there are few proposals and among all of them,the most interesting are capable to model only cylindrical data.In this work wepropose a flexible distribution for circular-linear variables, based on the projectedand the skew normal and we show how parameters estimate is easy within aBayesian framework. The proposal is used as emission distribution in a hiddenMarkov model applied to an animal movement dataset with multivariate circularand linear variables.

Title: Circular density estimation via matching local trigonometric momentsPresenter: Agnese Panzera, Università di FirenzeCo-author(s): Marco Di Marzio, Stefania Fensore, Charles C. TaylorAbstract: We propose the matching between population trigonometric moments and empir-

ical ones in order to estimate a circular density and its derivatives at a point. Themethod is non-parametric in nature, as the moments are locally weighted and thepopulation density function is locally approximated by a pth degree polynomial.

Specialized Session SPE-10 • Thursday 9, 14:45-16:15 • Room: A

THE INTERPLAY BETWEEN FREQUENTIST AND BAYESIAN INFERENCEChair: Walter Racugno; Discussant: Maria Eugenia Castellanos

Title: Classical inference for intractable likelihoodsPresenter: Clara Grazian, Sapienza, Università di RomaAbstract: Approximate Bayesian computation (ABC) and likelihood-free methods have been

recently proposed in a Bayesian setting to produce inference for complex mod-els for which the likelihood function may be considered intractable, i.e. difficult

58

or even impossible to evaluate, both analitycally and computationally. We pro-pose to use the ABC methodology within a classical approach, in order to obtaina likelihood function for (few) parameters of interest in presence of (potentiallymany) nuisance parameters when the model is particularly complex. We study theproposed method in some theoretical and practical examples.

Title: Fusion learning for Interlaboratory ComparisonPresenter: Jan Hannig, University of North Carolina at Chapel HillCo-author(s): Qing Feng, Hari Iyer, C. M. Wang, Xuhua LiuAbstract: In this paper we propose a Generalized Fiducial Inference inspired m ethod for

finding a robust consensus of several independently derived collection of confidence distributions (CDs) for a quantity of interest. The resulting fused CD is robustto the existence of pote ntially discrepant CDs in the collec- tion. The method usescomputationally efficient fiducial model avera ging to obtain a robust consensusdistribution without the need to eliminate discrepant CDs from the an alysis. Thiswork is motivated by a commonly occurring problem in interlaboratory trials,where diffe rent national laboratories all measure same unknown true value of aquantity and report their CD s. These CDs need to be fused to obtain a consensusCD for the quantity of interest. When some of t he CDs appear to be discrepant,simply eliminating them from the analysis is often not an acceptable app roach,particularly so in view of the fact that the true value being measured is not knownand a discrepant result from a lab may be closer to the true value than the restof the results. Additionally, e liminating one or more labs from the analysis canlead to political complications since all labs are regard ed as equally competent.These considerations make the proposed method well suited for the task since nolaboratory is explicitly elimi- nated from consideration. We report results of threesimulation ex periments showing that the proposed fiducial approach has bettersmall sample properties than the cur rently used naive approaches. Finally, weapply the proposed method to obtain consensus CDs for gauge b lock calibrationinterlaboratory trials and measurements of Newton’s constant of gravitation ( G )by several laboratories.

Title: p-value in science: a review of issues and proposed solutionsPresenter: Francesco Pauli, DEAMS – Università di TriesteAbstract: We review the recent debate on the lack of reliability of scientific results and its

connections to the statistical methodologies at the core of the discovery paradigm.Contributions to the debate include recommendations for paradigm changes aswell as methods to quantify the reliability of the present paradigm.We reviewthe main proposals within the two approaches and we propose a new modelingstrategy for assessment purposes to exploit data from replication experiments.

59

Specialized Session SPE-11 • Thursday 9, 14:45-16:15 • Room: B

SOCIÉTÉ FRANÇAISE DE STATISTIQUEChair: Avner Bar-Hen; Discussant: –

Title: Stochastic Block Model for Multiplex network: an application to a multilevelnetwork of researchers.

Presenter: Bar-Hen Avner, Université Paris DescartesAbstract: Modeling relations between individuals is a classical question in social sciences

and clustering individuals according to the observed patterns of interactions al-lows to uncover a latent structure in the data. Stochastic block model (SBM) isa popular approach for grouping the individuals with respect to their social com-portment. When several relationships of various types can occur jointly betweenthe individuals, the data are represented by multiplex networks where more thanone edge can exist between the nodes. We extend the SBM to multiplex networksin order to obtain a clustering based on more than one kind of relationship. Wepropose to estimate the parameters –such as the marginal probabilities of assign-ment to groups (blocks) and the matrix of probabilities of connections betweengroups– through a variational Expectation-Maximization procedure. Consistencyof the estimates is studied. The number of groups is chosen thanks to the Inte-grated Completed Likelihood criterion, a penalized likelihood criterion. Multi-plex Stochastic Block Model arises in many situations but our applied exampleis motivated by a network of French cancer researchers. The two possible links(edges) between researchers are a direct connection or a connection through theirlabs. Our results show strong interactions between these two kinds of connectionsand the groups that are obtained are discussed to emphasize the common featuresof researchers grouped together.

Title: Nonnegative Matrix Factorization for Transfer LearningPresenter: Younes Bennani, Université Paris 13 - Sorbonne Paris CitéCo-author(s): Ievgen RedkoAbstract: The ability of a human being to extrapolate previously gained knowledge to other

domains inspired a new family of methods in machine learning called transferlearning. Transfer learning is often based on the assumption that objects in bothtarget and source domains share some common feature and/or data space. If thisassumption is false, most of transfer learning algorithms are likely to fail. In thistalk we propose to investigate the problem of transfer learning from both theoret-ical and applicational points of view. First, we present two different methods tosolve the problem of unsupervised transfer learning based on Non-negative ma-trix factorization techniques. First one proceeds using an iterative optimizationprocedure that aims at aligning the kernel matrices calculated based on the datafrom two tasks. Second one represents a linear approach that aims at discoveringan embedding for two tasks that decreases the distance between the correspond-ing probability distributions while preserving the non-negativity property. Wealso introduce a theoretical framework based on the Hilbert-Schmidt embeddingsthat allows us to improve the current state-of-the-art theoretical results on transferlearning by introducing a natural and intuitive distance measure with strong com-putational guarantees for its estimation. The proposed results combine the tight-

60

ness of data-dependent bounds derived from Rademacher learning theory whileensuring the efficient estimation of its key factors. Both theoretical contributionsand the proposed methods were evaluated on a benchmark computer vision dataset with promising results. Finally, we believe that the research direction chosenin this study may have fruitful implications in the nearest future

Title: Detection of dependence patterns with delayPresenter: Thomas Laloe, Université de Nice - Sophia AntipolisAbstract: The Unitary Events (UE) method is a popular and efficient method used this last

decade to detect dependence patterns of joint spike activity among simultaneouslyrecorded neurons. The first introduced method is based on binned coincidencecount (Grün, 1996) and can be applied on two or more simultaneously recordedneurons. Among the improvements of the methods, a transposition to the con-tinuous framework has recently been proposed in (Muino and Borgelt, 2014) andfully investigated in (Tuleau-Malot et al., 2014) for two neurons. The goal ofthe present paper is to extend this study to more than two neurons. The mainresult is the determination of the limit distribution of the coincidence count. Thisleads to the construction of an independence test between L = 2 neurons. Finallywe propose a multiple test procedure via a Benjamini and Hochberg approach(Benjamini and Hochberg, 1995). All the theoretical results are illustrated by asimulation study, and compared to the UE method proposed in (Grün et al., 2002).Furthermore our method is applied on real data.

Title: Disaggregated Electricity Forecasting using Wavelet-Based Clustering of Indi-vidual Consumers

Presenter: Jean-Michel Poggi, Univ. Paris Descartes, France and LMO, Univ. Paris-SudOrsay, France

Co-author(s): Jairo Cugliari, Yannig GoudeAbstract: Electricity load forecasting is crucial for utilities for production planning as well

as marketing offers. Recently, the increasing deployment of smart grids infras-tructure requires the development of more flexible data driven forecasting meth-ods adapting quite automatically to new data sets. We propose to build clusteringtools useful for forecasting the load consumption. The idea is to disaggregate theglobal signal in such a way that the sum of disaggregated forecasts significantlyimproves the prediction of the whole global signal. The strategy is in three steps:first we cluster curves defining super-consumers, then we build a hierarchy ofpartitions within which the best one is finally selected with respect to a disaggre-gated forecast criterion. The proposed strategy is applied to a dataset of individualconsumers from the French electricity provider EDF.

61

Thursday 9, 16:45-17:30 • Room: Aula Magna

SIS PRESIDENT’S LECTURE

by Nicola TorelliUniversità di Trieste

Solicited Session SOL-16 • Friday 10, 09:00-10:30 • Room: G. De Rosa

FORECASTING ECONOMIC AND FINANCIAL TIME SERIESChair: Tommaso Proietti

Title: Asymptotics and power of entropy based tests of dependence for categoricaldata.

Presenter: Simone Giannerini, Università di BolognaCo-author(s): Greta GoracciAbstract: The literature on measures of dependence and on tests for independence for cate-

gorical data is very wide. In this work we focus on the power-divergence family ofstatistics (Read, T., Cressie, N.: Goodness-of-Fit Statistics for Discrete Multivari-ate Data. Springer Series in Statistics. Springer New York (1988)), that includesas a special case both the measure Sρ and Pearson’s chi square. Nevertheless,the theoretical derivations concern the null hypothesis of stochastic independenceagainst the alternative of some sort of local deviation from it. We derive an asymp-totic approximation valid for the whole family and for every given level of depen-dence. This allows to build tests where H0 : Sρ = S0 ≥ 0 against H1 : Sρ 6= S0.Moreover, we can compute analytically the power of the tests for independencethat rely on the power-divergence family. We compare the performance of testsbased on Sρ with that of classical χ2 both analytically and by means of MonteCarlo studies.

Title: Forecasting electricity load and price: a comparison of different approachesPresenter: Matteo Maria Pelagatti, Università di Milano-BicoccaAbstract: Electricity market time series include several components describing the long-

term dynamics, annual, weekly and daily periodicities, calendar effects, jumps,and so on.Thus, predicting electricity variables requires to estimate and forecastall these components.Two different models are applied to the same dataset: thefirst one treats the components as deterministic by representing them throughsplines and dummy variables, while the second uses stochastic (time-varying) ver-sions of these components in state-space form. We compare the forecasting accu-racy of these two approaches also with predictions produced by machine learningalgorithms such as random forests and support vector machines.

62

Title: Flexible Realized GARCH ModelsPresenter: Giuseppe Storti, Università di SalernoCo-author(s): Richard GerlachAbstract: We introduce a new class of flexible Realized GARCH models. Our model gen-

eralizes the original specification of Hansen, Huang and Shek (2012) along threedifferent directions.First, it allows for heteroskedasticity of the noise componentin the measurement equation. Second it features a time varying volatility persis-tence. Namely, the shock response coefficient in the volatility equation adjusts tothe time varying accuracy of the associated realized measure. Finally, our frame-work allows to consider, in a parsimonious way, the inclusion of multiplerealizedmeasures. The appropriateness of the proposed class of models is appraised bymeans of an application to a set of stock returns data.

Solicited Session SOL-17 • Friday 10, 09:00-10:30 • Room: V. Foa

IMMIGRATIONS AND INTEGRATION IN ITALYChair: Anna Paterno

Title: Minorities internal migration in Italy: an analysis based on gravity modelsPresenter: Oliviero Casacchia, Sapienza, Università di RomaCo-author(s): Cecilia Reynaud, Salvatore Strozza, Enrico TucciAbstract: Italians and foreigners internal migration assume different behaviour in terms of

intensity, geography, type. The levels of the mobility, the propensity to move ina short or in a long range, the propensity to cluster or to disseminate in the hostcountry represent important differential characteristics between the two popula-tions. Actually the foreign population seems as a mosaic made up of minoritiesshowing different propensities. This is the reason why an analysis consideringforeign population as a whole could reach biased outpcomes. In the paper somegravity models applied to migratory movements among the 110 Italian provincesconcerning the most consistent minorities groups are used. The Poissonian ef-fects (regarding various typologies of masses and distance) show in a syntheticway the main differences among the minorities internal mobility. Moreover, theinterpretation of these parameters allows an original interpretation of the minori-ties mobility structure inside Italy: the sign and level of the estimates derived fromthe gravity model permit to better illustrate the residential model of the minoritiesreflecting how different theories in this domain act.

Title: Growing generations and new models of integrationPresenter: Cinzia Conti, IstatAbstract: The survey on “Integration of the second generation”, carried out in 2015 by Istat

(financed by EIF), involved lower and upper secondary schools. Nearly 38% ofthe foreign students declared to feel “Italian”; 33% felt foreigner and more than29% preferred not to answer. Among the students arrived at the age of 10 or olderabout the 53% felt “foreigner”. The situation for foreign students born in Italy isvery different: only 23.7% of respondents felt foreigner, while 47.5% felt Italian.Many Italian students (42.6%) and foreign students (46.5%) would like to liveabroad when they will be adult. Regression models have been applied to studyceteris paribus the propensity of the students (Italians and foreigners) to live in

63

Italy when they will be adult.

Title: Measurement of segregation in the labour market. An alternative approachPresenter: Nicola Tedesco, Università di CagliariCo-author(s): Luisa SalarisAbstract: In the last years different comparative studies have shown that foreigners in Italy

experience difficulties in entering regular and skilled jobs, whereas have fairlyeasy access to irregular, unskilled and semi-skilled jobs, with no chances of afuture upward promotion. It was showed that ethnicization and specializationprocesses lead to occupational segregation in specific sectors of the economy formany different foreign community. In this job we want to estimate level of segre-gation of foreign workers vs italians, using data of RCFL of ISTAT for the years2007 and 2010. Usually the level of segregation is measured by index of dissim-ilarity (ID) or others measures. As recently shown in literature these measurestend to over-estimate segregation in small groups, while this bias decreases ac-cording to increasing the dimension of groups. In this job we apply an alternativemethod to measure the segregation based on a multilevel approach that permitus to consider time and spatial effects and can adjust and explain differences forcharacteristics of the individuals. This alternative approach is based on a simula-tion method according to the estimated segregation parameters are expression ofany desired n-groups segregation indexes.

Title: Family behaviours among first generation migrantsPresenter: Laura Terzera, Università degli Studi di Milano-BicoccaCo-author(s): Elisa Barbiano di BelgiojosoAbstract: Over the last decade the number of families in Italy with foreign members has

increased considerably transforming immigrants from ‘labour force’ to ‘popula-tion’. However, the family settlement in the host country has taken place differ-ently among migrants depending on their countries of origin. Focusing on genderand origin, we identified different family migration patterns. The analysis is basedon a pooled dataset obtained from 12 repeated cross-sectional surveys (Osservato-rio Regionale per l’integrazione e la Multietnicità, ORIM 2001-2012) on migrantsliving in Lombardy. In profiling the differences in family behaviours of migrantsof different origin, the results highlight the clear role of gender. The behavioursof women and men with the same origin match revealing a specific profile.

Solicited Session SOL-18 • Friday 10, 09:00-10:30 • Room: SP/1

OPEN DATA, LINKED DATA AND BIG DATA IN PUBLIC ADMINISTRATION ANDOFFICIAL STATISTICS

Chair: Cristina Martelli

Title: Linked Administrative Data in Official Statistics: a Positive Feedback for theQuality?

Presenter: Grazia Di Bella, IstatCo-author(s): Giuseppe GarofaloAbstract: From the organizational point of view, Istat determined that the acquisition of

administrative data must generally be made at a central level. The centralized

64

functions are: i) information needs collection from all Istat production units; ii)formulation of the data request to the owners of administrative data; iii) man-agement of the administrative data repository having the aim to standardize andintegrate data; iv) data and metadata dissemination to internal users. As part ofthis process, an evaluation function is needed to optimize quality. This functionis articulated in terms of data input, processing and output (statistics producer-oriented). Starting from the experience of the international project BlueEts, ageneralized Quality Report Card is going to be implemented in order to docu-ment information about usability and quality monitoring of administrative datareceived by Istat. The source holder and the administrative data users play an im-portant role. It is crucial to coordinate and share information among them in orderto improve as much as possible the administrative data quality, mitigate impacts ofpossible changes in the data supplies and finally enhance the use of administrativedata over time.

Title: Generating high quality administrative data: new technologies in a nationalstatistical reuse perspective

Presenter: Cristina Martelli, Università di FirenzeCo-author(s): Manlio Calzaroni, Antonio SamaritaniAbstract: The statistical reuse of administrative data is limited by serious issues about data

quality: these concerns are particularly serious in innovative context, in whichonly administrative data provide the required granularity and fit to the real pro-cesses. Administrative data semantic and meta-information are hardly ascribableto general notation standards: PA, as a services purchaser, is often not able toprovide to its suppliers efficient clues on the way data must be denoted for theirstatistical reuse. In this paper we discuss how new technologies and semantic web,may provide unprecedented methods and instruments to support PA in orientingtheir suppliers to high quality administrative data: the joint role of the NationalStatistical Institute and Public bodies, as owners of administrative data, will bediscussed.

Title: An innovative approach about the analysis of quality and efficiency in Italianlaw

Presenter: Vito Santarcangelo, Università di CataniaCo-author(s): Antonio Buondonno, Angelo Romano, Massimiliano Giacalone, Carlo CusatelliAbstract: A lot of information about processes is today available on internet, then it is pos-

sible to retrieve documents and produce a digital repository. Thanks to techniquesof text mining is possible to extract information useful to reach the task of definea statistical index to describe the quality and efficiency in law. It is also possi-ble to produce an intelligent knowledge base about facts and judgments. In thispaper we propose an useful tool to fulfill the essential constitutional principle inthe article 111 of Italian Constitution, that concerns the "reasonable duration" ofthe process, realising also the precept of the art. 6 of the European Conventionof Human Rights. This work aims to open some new conjoint debates about thestudy and application of statistical and computational methods to web data on newforensic topics.

65

Title: Prato municipality experience towards a high integration between administra-tive and statistical data L’esperienza del Comune di Prato per l’integrazione deidati amministrativi e statistici

Presenter: Benedetta Squittieri, Comune di PratoAbstract: Prato municipality has an important tradition in innovative computer aided ser-

vices delivery: in this perspective, when electronic invoicing has become com-pulsory for Italy, the Municipality has opted for a complete process demateri-alization, to enhance the storage and processing purchasing efficiency. Havingdigitized the whole process, Prato municipality has empowered its informationand knowledge base; which is now available both for management control andfor dissemination, transparency, for feeding the public statistical information sys-tem. As Prato is an Italian excellence, in adopting electronic invoicing, in thispaper management difficulties and statistical potentialities will be presented anddiscussed.

Solicited Session SOL-19 • Friday 10, 09:00-10:30 • Room: SP/2

EVALUATION OF PROGNOSTIC BIOMARKERSChair: Maria Grazia Valsecchi

Title: Combining Clinical and Omics data: hope or illusion?Presenter: Federico Ambrogi, Università degli Studi di MilanoCo-author(s): Patrizia BoracchiAbstract: In modern biomedicine the combination of clinical information with that of mul-

tiple biomarkers coming form transcriptomic-wide experiments is a theme of cen-tral interest. New clinical trials design try to combine the evaluation of the effectof new drugs with that of the identification of subgroups with maximum bene-fit. The subgroups are identified with measurements from omic data (genomics,proteomics, lipidomics, etc.). A general concern regards the magnitude of theeffects from omic data which is a central point when designing a trial. In thiswork we present a simulation strategy to investigate the impact on some measuresof prognostic impact (namely the (integrated) prediction error and the C-index)of the magnitude of the effects from omic covariates. We hypothesise the pres-ence of clinical covariates with large impact on prognosis and not correlated withthe omic data. We adopt as a method of analysis the Cox regression with lassoregularisation.

Title: Graphical representations and summary indicators to assess the performanceof risk predictors

Presenter: Laura Antolini, Università di Milano BicoccaCo-author(s): Davide Paolo BernasconiAbstract: The availability of novel biomarkers in several branch of medicine opensroom for

refining prognosis by adding factors on top of those having an establishedrole.It is accepted that the impact of novel factors should not rely solely on regres-sioncoefficients and their significance. This motivated the fruitful literature inthelast decades proposing predictive power measures, such as Brier Score, ROCbasedquantities, net benefit and related inference. This work reviews the con-ceptual formulationsand interpretations of the available graphical methods and

66

summary measuresfor evaluating risk predictor models. The aim is to provideguidance in theevaluation process that from the model development brings therisk predictor to beused in practice.

Title: Multivariable prognostic model: external validation and model recalibrationwith application to non-metastatic renal cell carcinoma

Presenter: Paolo Chiodini, Seconda Università di NapoliCo-author(s): Luca CindoloAbstract: To define a useful prognostic model is not sufficient to show that it accurately

predicts outcome in the initial development data, but evidence is necessary thatthe model performs well also in other groups of individuals. External validationtests the accuracy of the system in a different setting to assess the transportabilityand generalizability of the model. Different measures are available to test modelaccuracy in term of discrimination and calibration and several methods have beenproposed to update prognostic models and improve model accuracy. In the ap-plication to non-metastatic renal cell carcinoma external validation of the Kattannomogram (KN) was performed in three European centres. KN predicts the 5-year probability of recurrence free survival. Results evidenced a good discrimina-tion of KN calculated by means of the Harrell c-index for censored data (c-index:0.807, 95%CI 0.777-0.835). However, the prognostic model was not well cali-brated. To improve prediction accuracy three different recalibration methods areused and compared to derive new recalibrated predicted probabilities. Amongthese methods, a modified version of the “validation by calibration” approach hadthe best result.

Solicited Session SOL-20 • Friday 10, 09:00-10:30 • Room: A

MODELS FOR STUDYING THE MOBILITY OF STUDENTSChair: Dalit Contini

Title: Modelling inter-regional patient mobility: evidence from the Italian NHSPresenter: Silvia Balia, CRENoS Università di CagliariAbstract: Free patient mobility among regions has been often considered a useful stimulus

for enhancing healthcare. However, some jurisdictions may underperform be-cause of economies of scale and spatial spillovers. This could challenge the sus-tainability of regional budgets in a decentralised National Health Service (NHS)where regions bear the costs of providing care to residents, and could put at riskuniversalism and equity of health care. This work aims at enhancing the un-derstanding the determinants of patient mobility by providing a comprehensivepicture of the patterns of inter-regional patient flows. Using a ten years (2001-2010) panel of Italian data on hospital discharges, we model bilateral Origin-to-Destination (OD) flows between any two regions within a gravity framework. Weestimate a dynamic panel model that includes a rich set of push and pull factorsand addresses region-pair-specific unobserved heterogeneity and spatial depen-dence. In addition to significant scale effects and spatial spillovers, which are notunder the control of the regional health authorities, our findings highlight the roleof local supply factors, such as the hospital capacity, technology, specializationand performance indicators.

67

Title: University mobility at enrollment: geographical disparities in ItalyPresenter: Antonella D’Agostino, Università di Napoli ParthenopeCo-author(s): Giulio Ghellini, Sergio LongobardiAbstract: Using a micro-dataset of Italian students enrolled at university for the first time in

2008 and derived from the Anagrafe Nazionale degli Studenti (ANS), we modelinternal student mobility as a function of both individual-level and territorial char-acteristics. We use multilevel modelling to explicitly account for the hierarchicalnature of our data (students nested within Italian districts - NUTS-3 geographicaggregation level) and to understand whether there are significant variations inmobility patterns within and between districts. District differences in student mo-bility remain significant even after controlling for individual characteristics: thisresult confirms that the geographical dimension is relevant for student mobility.

Title: From South to North? Mobility of Southern Italian students at the transitionfrom the first to the second level university degree

Presenter: Marco Enea, CRENoS Università di CagliariAbstract: In the last decades, the Italian University System has encountered several struc-

tural reforms aimed at making it more internationally competitive. Among them,the introduction of the University financial autonomy has triggered an “internal”competition among Universities to attract students from the entire country. Stu-dents’ enrollment at the first level has decreased significantly especially after theeconomic crisis of 2008, while the students’ migration from the South to the Cen-tral and Northern regions of the country has increased. These phenomena havecreated further inequalities within the country and a cultural and socio-economicloss for the South that does not appear to slow down. While Italian internal mo-bility at the first level has been previously investigated, second level mobility hasreceived little attention. This work attempts to fill this gap, by analyzing the tran-sition from first to second level university degree courses of the Southern Italianstudents in terms of macro-regional mobility. The data were provided by the Ital-ian Ministry of Education, University and Research. They are a national levellongitudinal administrative micro-data on educational careers of the freshmen en-rolled at the first level Italian university degree course in 2008-09 and followed upto 2014.We will use a discretetime competing risk model with the aim to detectthe determinants of the choices of Southern Italian students after their bachelordegree: discontinuing university; enrolling at the second level University degreecourse in a Southern university, or (moving) to Central or Northern universities.We will analyze the role played by demographic variables, time elapsed to get thefirst level degree, the performance in the previous schooling career, etc. in orderto provide mover or stayer profiles of Southern bachelors.

Title: Measuring territory student-attractiveness in Italy. Longitudinal evidencePresenter: Francesca Giambona, Università di CagliariAbstract: The aim of this paper is to investigate the factors affecting university student

mobility in Italy in a longitudinal perspective, by considering the flows acrosscompeting territorial areas supplying tertiary education programs. The Bradley-Terry modelling approach based on pair comparisons has been adopted to definethe attractiveness of competing territories and a range of determinants related tothe socio-economic characteristics of the areas as well as universities’ resources.Data released by the Italian Ministry of Education (MiUR) are analysed for theacademic years 2010/2011-2014/2015. The modelling approach considers score

68

values for each territory and year, allowing to evaluate whether attractiveness im-proves or deteriorates over time, and to rank areas according to their attractive-ness. To this end, an index based on ranking changes, appropriately weighed withthe differences in score values, is proposed. Empirical findings highlight that at-tractiveness depends not only on the educational programs, but also on territories’socio-economic factors, reflecting the well-known North-South divide.

Plenary Session C • Friday 10, 10:30-11:30 • Room: Aula Magna

FOREIGN IMMIGRATION IN ITALY: A FORTY-YEAR-OLD HISTORY

by Salvatore StrozzaUniversità di Napoli Federico II

Chair: Alessandra De Rose (Università di Roma “La Sapienza”)Discussant: Massimo Attanasio (Università di Palermo)

Abstract: Italy has long been a multi-ethnic and multicultural country, with over 5 millionresident foreigners displaying a great variety in terms of origins, characteristicsand behaviours. After placing our country within Europe’s context of migration,this paper describes the contribution of demography to the observation of the phe-nomenon and to the evaluation of its demographic impact. Finally, the paper dealswith a few major issues related to the integration of adult migrants and their chil-dren’s schooling.

Specialized Session SPE-12 • Friday 10, 12:00-13:30 • Room: SP/2

NATIONAL ACCOUNTSChair: Luigi Biggeri; Discussant: Vittorio Nicolardi

Title: The European Welfare State in times of crisis according to macroeconomicofficial statistics

Presenter: Alessandra Coli, Università di PisaCo-author(s): Erika Micheletti, Barbara PaciniAbstract: This paper aims at exploring changes in welfare state in Europe in the last decade,

focussing on social protection expenditure. In particular, we examine the evo-lution of social protection expenditures in four countries, namely United King-dom, Denmark, Italy and France, each one representing a specific welfare statemodel. The objective is to point out whether the crisis is pushing countries to-wards more homogeneous expenditure patterns or, at the opposite, towards evenmore polarized systems. Furthermore, we investigate the effect of social protec-tion policies on people economic well-being (disposable income, consumptionexpenditure and wealth), with a focus on social benefits. The analysis is based onmacroeconomic data, namely National Accounts and ESPROSS data.

69

Title: National Account and Open Data: a new semantic approachPresenter: Cristina Martelli, Università di FirenzeAbstract: –

Title: New information contents of the National Accounts for the monitoring of theeconomic situation

Presenter: Gian Paolo Oneto, IstatAbstract: –

Specialized Session SPE-13 • Friday 10, 12:00-13:30 • Room: G. De Rosa

STATISTICAL TOOLS FOR MONITORING THE EDUCATIONAL SYSTEM AND ASSESSINGSTUDENTS’ PERFORMANCES

Chair: Isabella Sulis; Discussant: Ornella Giambalvo

Title: Evaluation of university students’ performance through a multidimensional fi-nite mixture IRT model

Presenter: Leonardo Grilli, Università di FirenzeCo-author(s): Silvia Bacci, Francesco Bartolucci, Carla RampichiniAbstract: The paper analyzes the performance of university students, with reference to first-

year compulsory courses. We propose an Item Response Theory model that in-cludes two latent variables corresponding to the student’s ability and the pref-erence about the order of attempted exams. In this way, we explicitly accountfor nonignorable missing observations since the indicators of item response alsocontribute to measure the ability and then the model is of within-item multidimen-sional type. The two latent variables are assumed to have a discrete distributiondefining latent classes of students that are homogenous in terms of ability andpriority assigned to exams.

Title: Monitoring school performance using value-added and value-table models: Lessonsfrom the UK

Presenter: George Leckie, Centre for Multilevel Modelling and Graduate School of Educa-tion, University of Bristol

Abstract: Since 1992, the UK Government has published so-called ‘school league tables’summarizing the average attainment and progress made by pupils in each state-funded secondary school in England. In this article, we statistically critique andcompare prominent past, current and forthcoming value-added measures of schoolperformance. We discuss the advantages and disadvantages of the different mea-sures as well as their underlying statistical models.

Title: A statistical model to assess teacher performancePresenter: Pasquale Sarnacchiaro, Università di Roma Unitelma SapienzaCo-author(s): Ida Camminatiello, Raffaela PalmaAbstract: The goal of this research is to determine the drivers which affect teacher per-

formance in public services. Many studies show that public service motivationis an important predictor of public employees’ performance. Starting from thesestudies we construct a conceptual model that allows us to identify the factors influ-encing teacher performance. The model was then tested through a sample surveycarried out by delivering a questionnaire. By means of the combined applica-

70

tion of Item Response Theory and Structural Equation Modelling the effects ofpublic service motivation and other key factors, such as Organizational Citizen-ship Behaviour, Job Satisfaction, Person-Organization Fit and User Orientationon teacher performance have been studied.

Specialized Session SPE-14 • Friday 10, 12:00-13:30 • Room: SP/1

ROBUST INFERENCE BY BOUNDED ESTIMATING FUNCTIONSChair: Luca Greco; Discussant: Marco Riani

Title: M Estimation based Inference for Ordinal Response ModelPresenter: Anna Clara Monti, Università del SannioAbstract: Outlying covariates as well as anomalous data in the response variable can jeop-

ardize the quality of inferential analyses on ordered response models based on thelikelihood function. Attention to robustness issues needs be paid in two funda-mental moments: when choosing the link function, and in the selection of the in-ferential methods which should have robustness properties adequate to the anoma-lies which are likely to arise in the data. Consequently the paper compares twoof the most popular links, and illustrates a robust M estimator. The M estimatoris compared with the Maximum Likelihood estimator in an extensive numericalexperiment where the M estimator systematically outperforms the classical esti-mators when anomalous data occur.

Title: Approximate Robust Bayesian Inference with an Application to Linear MixedModels

Presenter: Erlis Ruli, Università di PadovaCo-author(s): Nicola Sartori, Laura VenturaAbstract: We illustrate a novel approach for developing robust posterior distributions using

Approximate Bayesian Computation (ABC) methods from robust M- estimatingequations. The method is formally motivated by the use of unbiased estimatingfunctions as automatic informative summary statistics in ABC. The usefulness ofthe method is shown by a simulation study as well as by a linear mixed modelsapplied to a real-life dataset.

Title: Some robust methods using empirical likelihood for two samplesPresenter: Janis Valeinis, University of LatviaCo-author(s): Mara Velina, Edmunds Cers, George LutaAbstract: In this paper we make some review of the empirical likelihood method for the

two-sample case in a general framework. Empirical likelihood has several ap-pealing properties: it is a nonparametric procedure, the shape of the respectiveconfidence intervals or regions is data-driven and asymmetric, usually it admitsthe Bartlett correction. Regarding robust statistical inference we introduce theempirical likelihood method for the difference of smooth Huber estimators andtrimmed means. Finally, we analyze the empirical level of the tests by some smallsimulation study.

71

Contributed Session CON-09 • Friday 10, 16:15-17:30 • Room: G. De Rosa

ECONOMIC DATA ANALYSISChair: Maurizio Carpita

Title: Getting older and riskier: the effect of Medicare on household portfolio choicesPresenter: Marianna Brunetti, Università di Roma Tor Vergata, CEIS and CEFINCo-author(s): Marco Angrisani, Vincenzo AtellaAbstract: The continuous increase in health care costs has become an increasingly impor-

tant contributor to financial risk. To the extent that these costs can be large, un-predictable, and are uninsured, they form a source of background risk that couldpotentially drive households toward more safe financial portfolios. Using HRSdata we empirically test to what extent the Medicare program in US can helpsheltering this background risk. The results show that Medicare eligibility is ableto offset the negative effect of bad health status on household portfolio at boththe intensive and the extensive margin. This effect is driven by households with-out health insurance. These results are robust in terms of sign, magnitude andstatistical significance to several samples and model specifications.

Title: Modelling the Public Opinion on the European Economy with the HO-MIMICModel

Presenter: Maurizio Carpita, Università di BresciaCo-author(s): Enrico CiavolinoAbstract: Is there an effect of the news for the European economic conditions on the cit-

izens’ opinion about this economy? The study aims at measuring this relation,defining a theoretical framework based on available data from the European Of-ficial Statistics, and implementing a Higher Order Multiple Indicators MultipleCauses (HO-MIMIC) Model with parameters estimated using the Partial LeastSquares (PLS) method.

Title: Indexing the Worthiness of Social Agents. To norm index on conventional spec-ifications

Presenter: Giulio D’Epifanio, Università degli Studi di PerugiaAbstract: An index is constructed to evaluate the worthiness of social agents standardized on

the conventional reference-framework specified by the policy-maker. Applying arecursive principle of “minimum information”, the index may be adapted on thedata of the chosen standard agent, normalized on the behavior model convention-ally assumed by the policy-maker. An interdisciplinary attempt is made herein tointegrate concepts and methods scattered in various fields.

Title: An econometric model for undeclared workPresenter: Giuseppina Guagnano, Sapienza, Università di RomaCo-author(s): Maria Felice ArezzoAbstract: Tax evasion is a matter of huge concern for all European Member States as it

affects all of them. It undermines the public finance of a State and it can leadto a very severe inefficient resource allocation. It is therefore very important tounderstand which are the mechanisms underlying the decision to not comply. Wefocus on a particular facet of tax evasion: undeclared work. We use microdataon audits and link them to a vast set of firms characteristics, mainly related to the

72

economic, financial and structural dimensions. There are two main methodologi-cal issues arising from the type of data we used. The first is the non-randomnessof the inspections and the second is that the fraction of inspected firms in the pop-ulation is low (around 4%). In order to cope with these problems, we develop aprobit model with sample selection in the case-control framework.

Title: A spatial shift-share decomposition of energy consumption variationPresenter: Mauro Mussini, Università degli Studi di VeronaCo-author(s): Luigi GrossiAbstract: This paper shows a spatial shift-share decomposition of the change in regional

energy consumption. Applying the traditional shift-share approach to the analysisof energy consumption, a new decomposition explaining the spatial effects of en-ergy efficiency change is introduced. By means of this decomposition of regionalenergy consumption change, the roles played by the changes in regional output,industry mix and energy efficiency are separated. More specifically, the regionalcompetitive effect in energy efficiency is measured taking the variations in energyefficiency of neighbouring regions into account.

Contributed Session CON-10 • Friday 10, 16:15-17:30 • Room: V. Foa

QUANTILE METHODSChair: Domenico Vistocco

Title: Bayesian inference for Lp–quantile regression modelsPresenter: Mauro Bernardi, Università di PadovaCo-author(s): Valeria Bignozzi, Lea Petrella, Lea Petrella, Lea PetrellaAbstract: Lp–quantiles generalise quantiles and expectiles to account for the whole distri-

bution of the random variable of interest. In this paper, we introduce the Lp–quantile regression model, we propose a collapsed Gibbs–sampler algorithm tomake Bayesian inference on the regression parameters. We also provide sometheoretical results concerning the posterior distribution of the regression parame-ters.

Title: On the Lp-quantiles and the Student t distributionPresenter: Valeria Bignozzi, Sapienza, Università di RomaCo-author(s): Mauro Bernardi, Lea PetrellaAbstract: Lp-quantiles belong to the class of generalised quantiles. In this contribution we

show that the Lp-quantiles represent an important class of risk measures; we studytheir mathematical properties and provide some recursive equations to computethem. For the special case of the L2-quantile, also known as expectile, it has beenshowed the equivalence of the quantile and the expectile for a class of distributionsthat includes the Student t distribution with 2 degrees of freedom. We extend thisresult and show that for a general Student t distribution with p degrees of free-dom (and any affine transformation), the Lp-quantile and the quantile coincide.This result opens up new research directions for the computation of the Student tquantile that is generally not available in closed form and difficult to compute.

Title: M-quantile regression for multivariate longitudinal dataPresenter: Maria Francesca Marino, Università degli Studi di Perugia

73

Co-author(s): Marco Alfo’, Maria Giovanna Ranalli, Nicola SalvatiAbstract: We propose a M-quantile regression model for the analysis of multivariate con-

tinuous longitudinal data. M-quantile regression represents an appealing alter-native to standard regression models, as it combines the robustness of quantileand the efficiency of expectile regression, detailing a picture of the response vari-able distribution. Discrete individual-specific random parameters are consideredto account for both dependence within longitudinal profiles and association be-tween multiple responses from the same sample unit. An extended version of thestandard EM algorithm for mixed models is proposed to derive model parameterestimates.

Title: Comparing Prediction Intervals in Quantile and OLS RegressionPresenter: Domenico Vistocco, Università di Cassino e del Lazio MeridionaleCo-author(s): Cristina DavinoAbstract: In the regression framework, prediction intervals are a valuable tool to estimate

the value of the response variable. Such prediciton intervals can be formulated interms of the expected value of the response variable as well as for a single specificvalue. Bothe the type of intervals suffer of violations of the assumptions of theclassical regression models, resulting in empirical coverage levels not consistentwith nominal levels. Among the several possibilities proposed in literature to facethis problem, we consider the estimations provided by quantile regression at twodifferent quantiles to obtain prediction intervals. Exploiting the non parametricnature of quantile regression, such intervals are useful in situations characterisedby heteoscedasticity or when the response variable is skewed.

Contributed Session CON-11 • Friday 10, 16:15-17:30 • Room: SP/1

STATISTICAL ALGORITHMSChair: Nicola Loperfido

Title: An Algorithm for Finding Projections with Extreme KurtosisPresenter: Nicola Loperfido, Università degli Studi di Urbino Carlo BoCo-author(s): Cinzia FranceschiniAbstract: Projection pursuit is a multivariate statistical technique aimed at finding interest-

ing low-dimensional data projections. A projection pursuit index is a functionwhich associates a data projection to a real value measuring its interestingness:the higher the index, the more interesting the projection. Consequently, projec-tion pursuit looks for the data projection which maximizes the projection pursuitindex. The absolute value of the fourth standardized cumulant is a prominentprojection pursuit index. In the general case, a projection achieving either mini-mal or maximal kurtosis poses computational difficulties. We address them by analgorithm which converges to the global optimum.

Title: Poisson change-point models estimated by Genetic AlgorithmsPresenter: Luca Scrucca, Università degli Studi di PerugiaAbstract: Change-point analysis aims at both detecting whether or not a sharp change has

occurred, or whether several changes might have occurred, and identifying thetimes of any such changes.Numerous approaches to conduct a change-point anal-

74

ysis are available in the literature. In this paper we propose the use of GeneticAlgorithms (GAs) for estimating Poisson change-point models. GAs are stochas-tic search and optimisation technique inspired by natural evolution. They providea robust and flexible framework that can be applied to a wide range of learningand optimisation problems, in particular when traditional optimisation techniquesbreak down. A data analysis on the annual number of patients with haemolyticuremic syndrome is presented, with change-point models estimated using the GAR package.

Title: Maximum Likelihood Estimators of Brain White Matter MicrostructurePresenter: Aymeric Stamm, Politecnico di Milano, Boston Childrens Hospital, Harvard

Medical SchoolCo-author(s): Olivier Commowick, Simone Vantini, Simon Keith WarfieldAbstract: The microstructure of the brain white matter is not visible to the naked eye but

would be of invaluable help to the clinician in the diagnosis and treatment ofmany brain pathologies. Diffusion MRI is an in-vivo non invasive imaging tech-nique that probes the cyto-architecture of the white matter through the diffusionof water. However, diffusion MRI is limited in resolution, which makes for-ward models of the diffusion at the voxel level rather complex. In this paper, weprovide a statistical framework for recovering the maximum-likelihood estima-tors of the parameters of mixture models of the diffusion. We calibrate differentmethods on simulated data to guarantee convergence to the maximum likelihoodand show that profile likelihood maximization using variable projection togetherwith a Levenberg-Marquardt algorithm with analytic Jacobian is the most efficientmethod to obtain the MLE.

Contributed Session CON-12 • Friday 10, 16:15-17:30 • Room: SP/2

STATISTICS FOR MEDICINEChair: Mauro Ferrante

Title: Competing risks between mortality and heart failure hospital re-admissions: acommunity-based investigation from the Trieste area

Presenter: Giulia Barbati, Università di TriesteCo-author(s): Francesca Ieva, Arjuna Scagnetto, Gianfranco Sinagra, Andrea Di LenardaAbstract: Predictors of mortality and readmission among patients hospitalized for heart

failure (HF) were investigated in a large, unselected population of the Triestearea. The cohort of 4666 patients survived at the index admission in the pe-riod 2009-2014 was followed after discharge. Incidence of mortality and re-HFadmission within 30 days and one year were computed, by comparing cumula-tive incidence probabilities with cause- specific Kaplan-Meier curves. Competingrisks regression was used to find factors associated respectively with re-HF ad-mission and death. Two distinct risk profiles were obtained, particularly for earlyoutcomes, useful for better targeting treatment of these high-risk patients.

75

Title: Evaluating association between emotion recognition and Heart Rate Variabilityindices

Presenter: Chiara Brombin, Università Vita-Salute San Raffaele, CUSSB (University Cen-tre for Statistics in the Biomedical Sciences)

Co-author(s): Federica Cugnata, Riccardo Maria Martoni, Manuela Ferrario, Clelia Di SerioAbstract: Correctly recognizing own and others’ emotions is an essential skill to manage in-

terpersonal relationships in everyday life. It has been shown that Heart Rate Vari-ability (HRV) represents an important marker of the ability to understand other’smental states. This work aims at evaluating association between HRV indicesand performances in emotion processing, measured by means of the "Reading theMind in the Eyes" Test (RMET). To fulfill our goal, we first obtain a model-basedweighted total score by applying Rasch model on RMET data collected on a largesample from general population. Then relationships between test performance,resting-state HRV, demographics and psychopathological traits were assessed in asubsample of volunteers participating to the experimental sessions, where physio-logical measurements were recorded at rest and while completing RMET. Finally,Latent Class Mixed Models (LCMMs) have been applied (i) to evaluate whethersuccessful emotion recognition elicits a physiological activation, while control-ling for demographic and clinical characteristics, and (ii) to identify clusters ofsubjects characterized by similar longitudinal trajectories.

Title: Socio-economic deprivation, territorial inequalities and mortality for cardiovas-cular diseases in Sicily

Presenter: Mauro Ferrante, Università degli Studi di PalermoCo-author(s): Anna Maria Milito, Anna Maria ParrocoAbstract: The present work aims at analyzing the relationship among deprivation, distance

from the nearest hospital and mortality for cardiovascular diseases in Sicily. Dataderived from 2011 Census are used for the determination of the deprivation indexat the municipality level, whereas distance from the municipality of residenceand the nearest municipality with at least one hospital is considered in terms oftravel time. Results highlight association between socio-economic conditions andmortality for cardiovascular diseases, whereas it seems that the distance from thehospital is only poorly associated with mortality.

Title: The use of Permutation Tests on Large-Sized DatasetsPresenter: Massimiliano Giacalone, Università di Napoli Federico IICo-author(s): Angela Alibrandi, Agata ZirilliAbstract: The increasing availability of large-sized datasets produces a growing interest in

permutation testing methods. They represent an effective solution for problemsconcerning the testing of multidimensional hypotheses, difficult to face in a para-metric context. In this paper we propose an application of permutation test on alarge amount of data in order to show its utility into analyze a big dataset array.The analysis was performed in order to assess the existence of significant differ-ences, with reference to several variables, between two gastrointestinal illnesses.

76

Contributed Session CON-13 • Friday 10, 16:15-17:30 • Room: D

STATISTICS FOR THE EDUCATION SYSTEMChair: Alessandro Valentini

Title: Further considerations on a new indicator for higher education student perfor-mance

Presenter: Giovanni Boscaino, Università di PalermoCo-author(s): Giada Adelfio, Vincenza CapursiAbstract: This paper joins the international debate on academic achievements, inparticular

it offers some reflections about the the suitable marks system and theirsynthesis,since it is usually used as a performance academic student measure. The ItalianUniversity System is used as a starting point to make several considerationsandto suggest a new performance indicator. Then, a comparison between old and-new indicator is performed, also in terms of significance of the determinants ofthestudent performance. Finally, a generalization for other marking systems isshown.

Title: Analysis of pupils’ INVALSI achievements by means of bivariate multilevelmodels

Presenter: Chiara Masci, Politecnico di MilanoCo-author(s): Anna Maria Paganoni, Francesca Ieva, Tommaso AgasistiAbstract: The purpose of this study is to identify a relationship between pupils’mathematics

and reading test scores and the characteristics of students themselves, stratifyingfor classes, schools and geographical areas. The dataset of interest containsde-tailed information about more than 500,000 students at the first year of juniorsec-ondary school in the year 2012/2013. The innovation of this work is in the use of-multivariate multilevel models, in which the outcome variable is bivariate: readin-gand mathematics achievements. Using the bivariate outcome enables researcherstoanalyze the correlations between achievement levels in the two fields and to es-timatestatistically significant school and class effects after adjusting for pupil’scharacteristics.

Title: Promoting statistical literacy to university students: a new approach adopted byIstat

Presenter: Alessandro Valentini, IstatCo-author(s): Giulia De Candia, Monica CarbonaraAbstract: Istat, Italian NSI, has been pursuing the aim of promoting statistical literacy for

many years. Recently (2013) the constitution of a territorial network of experts indisseminating activities is a further effort towards this direction. A new project isdevoted to university students. The new approach is gradual: i) to assess statisti-cal literacy of students; ii) to intercept statistical requirements of professors; iii)to design standardized educational packages aimed at improving students abilityto read data and statistical information; iv) to guide students towards statisticalthinking through laboratories. Implications of the new approach are discussed.

77

Contributed Session CON-14 • Friday 10, 16:15-17:30 • Room: B

TESTING PROCEDURESChair: Gustavo De Santis

Title: A Reliability Problem: Censored TestsPresenter: Egidio Cascini, Italian Association for Quality (AICQ) - Italian Academy for Six

Sigma (AISS)Abstract: A device (for example, an electronic card) is likely to fail a certain number of

times during its mission time (time elapsed from start up to shut down). Thefailure time is defined as the time elapsed from start up to an observed reparaiblefailure. There are two categories of failures; casual failures, when the time tofailure is exponentially distributed and wear - out failures, when the time to failurefollows a normal or close to a normal distribution. In any given time interval,during its mission time, a failure can belong to either category. To define itsnature, laboratory tests are required. The present work proposes a novel methodby which to define the two probability distributions on the basis of data gatheredfrom laboratory tests, performed during a predetermined time rest. Additionallaboratory tests are not required after the time test. The method holds the potentialto be very useful, because determining distribution time by means of laboratorytests, sufficient to gather a significant sample, could require, paradoxically, a testtime equal to that of the device mission time.

Title: Testing the Gamma-Gompertz-Makeham modelPresenter: Gustavo De Santis, Università di FirenzeCo-author(s): Giambattista SalinariAbstract: The Gamma-Gompertz (GG) model offers an excellent description of mortality

at older ages. At a closer look, however, it also reveals a few shortcomings. Forinstance, the fact that mortality remains roughly constant at young adult ages (20-40 years) seems to suggest the inclusion of an age-independent component ofmortality, which leads to the Gamma-Gompertz-Makeham (GGM) model. Ourtests, however, indicate that this solution does not solve all the problems, and thata different way of improving the GG model should perhaps be considered.

Title: A nonparametric test of independencePresenter: Matteo Maria Pelagatti, Università degli studi di Milano-BicoccaAbstract: We introduce a nonparametric test for the independence of two random variables.

We derive the asymptotic distribution of the test under the null of independenceand verify by Monte Carlo simulations that it represent a good approximation ofthe finite sample distribution.We compare our test with the well-known Cramer-von Mises type empirical copula test and find it generally more powerful under awide range of alternative hypotheses, that include some time series models wildlyused in financial econometrics.

Title: Functional Data Analysis of Tongue ProfilesPresenter: Alessia Pini, Politecnico di MilanoCo-author(s): Lorenzo Spreafico, Simone Vantini, Alessandro ViettiAbstract: We present in this work the functional data analysis of a data set of tongue profiles

recorded for a study on Tyrolean, a German dialect spoken in South Tyrol. We test

78

pairwise differences between five different manners of articulation of the uvular/R/: vocalized /R/, approximant, fricative, tap, and trill. The analysis is based onthe Interval-Wise Testing (IWT), a non-parametric inferential procedure for func-tional data. The output of the IWT is an adjusted p-value function that can be usedto select intervals of the domain imputable for the rejection of a null hypothesis.IWT is jointly applied to the position, slope, and concavity of the tongue profiles.IWT-based comparisons result in an informative and detailed representation of theregions of the tongue where a significant difference is located.

Title: On the asymptotic power of the statistical test under Response-Adaptive ran-domization

Presenter: Alessandro Vagheggini, Università di BolognaCo-author(s): Alessandro Baldi Antognini, Maroussa ZagoraiouAbstract: We analyze the impact of Response-Adaptive (RA) randomization rules in the

case of normally response trials when the aim consists in testing the superiorityof one of two available treatments. Taking into account the asymptotic powerof the classical statistical test, we show that some target allocations may induceconsistent loss of power. Moreover we analyze the conditions under which thisproblem can be avoided, adopting suitable targets as well as correct applicationsof RA rules.

Contributed Session CON-15 • Friday 10, 16:15-17:30 • Room: A

TIME SERIES ANALYSISChair: Michele Costa

Title: Robust Atheoretical Regression Tree to detect structural breaks in financialtime series

Presenter: Carmela Cappelli, Università di Napoli Federico IIAbstract: In this paper we propose a method to locate multiple structural breaks in finan-

cial time series that accounts for the interval structure of these series as wellas for the presence of outliers. For each time unit, the upper and lower boundof the intervals depend on the closing value. Then, to locate the break dates, arobust exponential based distance, that is able to neutralize the impact of outlier,is employed in the framework of Atheoretical Regression Trees. An empiricalapplication to the prices of an asset shows the usefulness of the proposed proce-dure.

Title: Prediction intervals for heteroscedastic series by Holt-Winters methodsPresenter: Paolo Chirico, Università di TorinoAbstract: The paper illustrates a procedure for prediction intervals by Holt-Winters method

in case of heteroscedasticity. The procedure has been applied to daily electricityprices of the year 2014. The results are consistent with those obtained usingequivalent ARIMA-GARCH models.

Title: Inequality decomposition for financial variables evaluationPresenter: Michele Costa, Università di BolognaAbstract: This paper illustrates the use of the methods related to inequality decomposition

for the analysis of financial variables. By means of the overlapping component

79

and of the inequality between it is possible to detect and to assess the main factorsdetermining the cross section assets variability.

Title: Three-stage estimation for a copula-based VAR modelPresenter: Giorgia Rivieccio, Universitàdi Napoli ParthenopeCo-author(s): Giovanni De LucaAbstract: We present the estimation of a copula-based Vector Autoregressive model using a

three-stage method. The benefit of this approach is the high flexibility in assumingthe parametric distributions of the marginals, as well as the dependence structureamong the variables.

Contributed Session CON-16 • Friday 10, 16:15-17:30 • Room: Sala Studio

FORECASTING METHODSChair: Maria Simona Andreano

Title: Forecasting with Mixed Data Sampling Models (MIDAS) and Google trendsdata: the case of car sales in Italy

Presenter: Maria Simona Andreano, Universitas MercatorumCo-author(s): Roberto Benedetti, Paolo PostiglioneAbstract: In the last few years the attention focused on the use of Internet data as an in-

formation source that could improve forecasts. Google Trends data can provideweekly information that represent proxies for market expectations of the currenteconomic fundamentals and, therefore, may be helpful for short-term economicpredictions. Unfortunately, most economic activity indicators are sampled at alower frequency than the week, therefore the two datasets are unbalanced. In thepresent paper we use a Mixed Data Sampling Model to forecast monthly time se-ries with weekly real-time information obtained from Google trends. The modelis applied to car sales in Italy.

Title: Probability forecasts in the market of tennis betting: the CaSco normalizationPresenter: Vincenzo Candila, Università di SalernoCo-author(s): Antonio ScognamilloAbstract: The probability odds, obtained as the inverse of the betting odds, represent the

most accurate proxy of the bookmaker forecasts associated to each player/teamvictory. However, these probability odds cannot be defined as actual forecastsbecause they incorporate the bookmaker margin and the longshot bias such thattheir summation is greater than one. In literature, different normalization meth-ods have been proposed in order to close the gap between probability odds andthe underlying forecasts. This paper proposes a new method of normalization,named CaSco (Candila-Scognamillo) normalization, for the fixed odds offered onsports with only two possible outcomes. The performances of the new proce-dure are evaluated using the betting odds provided by one of the most importanton-line professional bookmaker on over 27.000 male tennis matches. The resultsshow that the CaSco normalization has a better forecasting ability than that of theother approaches presented in literature.

80

Title: Daily Prediction of Demand and Supply CurvesPresenter: Simone Vantini, Politecnico di MilanoCo-author(s): Antonio CanaleAbstract: We propose a model for the analysis of functional time series subjected to mono-

tonicity and bound-constraints on the codomain. In detail, we provide the spaceof constrained functions with a suitable pre-Hilbert structure and then model themby mean of a functional autoregressive model. The autoregressive lagged oper-ators and the non-centrality function of the model are estimated by minimizingthe squared L2 distance between functional data and functional predictions witha penalty term based on the Hilbert-Schmidt squared norm of the autoregressivelagged operators. Moreover, we prove the existence and uniqueness of the corre-sponding estimators. Finally, the method is successfully applied to the analysis ofdaily demand and supply curves of the Italian Natural Gas Balancing Platform.

Contributed Session CON-17 • Friday 10, 17:30-18:45 • Room: G. De Rosa

BAYESIAN STATISTICS (2)Chair: Lucia Paci

Title: Bayesian hierarchical models for analyzing and forecasting football resultsPresenter: Giovanni Marchese, Sapienza, Università di RomaCo-author(s): Pierpaolo Brutti, Stefania GubbiottiAbstract: We develop a Bayesian model for the analysis and forecasting of football match

results with a hierarchical structure based on conditionally independent Poissondistributions. The higher levels of the hierarchy reveal dependency through scor-ing intensities, which are modelled dynamically and stochastically over time. Weverify the out-of-sample performance of our model through a betting strategy thatis applied to the match outcomes of the 2013–2014 and 2014–2015 seasons of the“Serie A”, the Italian major football league. Exploiting different staking strate-gies, we show that our model is able to produce a substantial positive return overthe bookmaker’s average odds.

Title: Bayesian modeling of spatio-temporal point patterns in residential propertysales

Presenter: Lucia Paci, Università di BolognaCo-author(s): Alan E Gelfand, Mar ía Asunci ón Beamonte, Pilar Gargallo, Manuel SalvadorAbstract: It is of important economic interest to understand the market for sales of resi-

dential properties. Customary analysis focuses on explaining selling price usingproperty and neighborhood characteristics, so-called hedonic models. Here, ourinterest is in understanding the locations of the sales. The set of such locationsforms a space-time point pattern. With interest in comparing different types ofproperty sales, we obtain marked space-time point patterns according to the sizeof property. We focus on sales in the city of Zaragoza in Spain during a tenyearperiod. We employ nonhomogeneous Poisson process models as well as logGaussian Cox process models, fitted within a Bayesian framework, with investi-gation of model adequacy and model comparison.

81

Title: Non-parametric Bayesian Networks for Managing an Energy MarketPresenter: Vincenzina Vitale, Università Roma TreCo-author(s): Valentina Guizzi, Flaminia Musella, Paola VicardAbstract: Energy markets are typically characterized by high complexity due to severalrea-

sons such as the large number of occurring variables, different in nature, and theirassociative structure. Estimating a statistical model that properly represents thedependencies among the variables is crucial for managing the complexity. In thispaper the Colombian energy market is studied. Since the variables of interest arequantitative but non Gaussian, non parametric Bayesian networks are used to inferthe Colombian energy market association structure.

Contributed Session CON-18 • Friday 10, 17:30-18:45 • Room: V. Foa

BUSINESS STATISTICSChair: Alessandra Righi

Title: How do firms perceive their competitiveness? Measurement and determinantsPresenter: Eleonora Bartoloni, IstatAbstract: This contribution provides a novel approach for analysing firm-level competitive-

ness, namely, that of a firm’s subjective perception. By using a large, integrateddatabase and an econometric strategy based on a generalized order logit model,our results indicate the presence of sectoral specificities and group heterogene-ity in the way firms perceive their competitiveness. Industrial and services firmsperceive differently those factors of competitiveness such as profitability, techno-logical innovation, knowledge capital, complex ownership structure and interna-tionalization patterns. In addition, firms’ top performers tend to score more posi-tively a number of competitive factors indicating technological input and output,knowledge capital and managerial abilities. We suggest that the use of a per-ceived competitiveness indicator could provide useful insights for more focusedcompetitiveness policies.

Title: An evaluation of export promotion programmes with repeated multiple treat-ments

Presenter: Chiara Bocci, IRPET Istituto Regionale per la Programmazione Economica dellaToscana

Co-author(s): Marco MarianiAbstract: Export promotion programmes represent a usual tool of enterprise policy world-

wide. They usually consist of the provision to firms of a vast array of servicesand aids, including specialised consultancy, participation in trade missions andinternational fairs, organisation of business-to-business meetings and the set-upof temporary selling outlets. Despite their diffusion, these programmes have met,so far, only limited interest in the applied programme evaluation literature. Apeculiar issue with export promotion programmes is that aspiring exporters cantake advantage of suites of integrated services and aids, simultaneously or in asequence of moments in time. This constitutes an unusually complex setting forprogramme evaluation, since the analysis has to be carried out in a context withsequences of multiple treatments, which is unfeasible with ready-to-use econo-metric and thus requires to build on the existing literature and to design an ad hoc

82

evaluation strategy. Relying on assumptions of sequential ignorability extendedto the multiple-treatment framework, and exploiting the programme participationdata of an Italian region (Tuscany), we estimate the treatment effect of differ-ent services and aids on multiple aspects of the firms’ export performance usinga marginal structural model that adjusts for dynamic confounding by means ofinverse-probability-of-treatment weights.

Title: The inter-enterprise relations in ItalyPresenter: Alessandra Righi, IstatCo-author(s): Alessandra Nuccitelli, Giovanni Alfredo BarbieriAbstract: The paper analyses the inter-enterprise relations in Italy showing that they are

concentrated in enterprises with a high relational attitude, while more than a thirdof firms has no relations. The propensity to the inter-enterprise relations is stud-ied using official data sources and a multinomial logistic model. The results showthat the sector of economic activity, having made innovations, and the level of rev-enues from sales of goods and/or services are the most relevant factors in settingthe relational behaviour of the enterprises.

Contributed Session CON-19 • Friday 10, 17:30-18:45 • Room: SP/1

CLUSTERING AND CLASSIFICATIONChair: Giovanna Menardi

Title: Dendrograms Stability Analysis of Sub-periods Time Series ClusteringPresenter: Carlo Drago, Università di Roma Niccolò CusanoCo-author(s): Roberto RicciutiAbstract: An important problem on the analysis of time series is understanding the impact

that different unknown shocks may have on the stability of the clusters obtainedfrom a time series clustering. In this work we consider an approach based onthe comparison and the visualization of dendrograms on different sub-periods inorder to evaluate the stability of the results. The analysis allows to understandthe structural changes occurring on the different sub-periods by considering thedifferent clusterings.

Title: Stability-based model selection in nonparametric clusteringPresenter: Giovanna Menardi, Università di PadovaAbstract: In the recent years stability has been often employed as a criterion to evaluate

the quality of a clustering or to select the optimal partition among a number ofclustering alternatives, e.g. based on a different number of groups. In this work,the use of stability is discussed in the context of clustering based on nonparametricdensity estimation to select the optimal smoothing parameter, and its effectivenessshown via simulations and a real data example.

Title: Topological signatures for classificationPresenter: Tullia Padellini, Sapienza, Università di RomaCo-author(s): Pierpaolo BruttiAbstract: In recent years the statistical analysis of topological signatures computable from

a possibly high-dimensional point cloud has attracted increasing attention from atheoretical as well as applied point of view. In this work we briefly review the

83

basic techniques and main inferential results available in the TDA (TopologicalData Analysis) toolkit, showing them in action on two classification problemsrelated to activity tracking tasks.

Contributed Session CON-20 • Friday 10, 17:30-18:45 • Room: D

DEMOGRAPHICS AND SOCIAL STATISTICS (2)Chair: Bruno Arpino

Title: Ecolabels: informin or confusing customers? Evidences form the agrifood sec-tor

Presenter: Margaret Antonicelli, Libera Università Mediterranea “LUM Jean Monnet"Co-author(s): Donato Calace, Domenico Morrone, Angeloantonio Russo, Vincenzo VastolaAbstract: In the last decades, the proliferation of labels signalling the “green” features of

the products – the so-called ecolabels- has been evident. Ecolabels are a volun-tary cue that private and public organizations, in particular retail companies, useto give expression to their social, environmental, and ethical attitude, exceedingacquiescence to law. Ecolabels fall under the wider umbrella of Corporate SocialResponsibility (CSR) policies. However, scepticism taints the full accreditation ofecolabels, as the menace of greenwashing raises consumers’ distrust. Analysingthe roots of such scepticism, a central element to take into account is the confu-sion generated by ecolabels proliferation. The high number of these hampers aclear comprehension and trust building in the consumers’ mind. In particular, it isnot clear whether ecolabels convey valuable information or not, that is informa-tion able to change the perceived features of a product. The aim of this work is toinvestigate (1) to what extent consumers pay attention to ecolabels, consideringthem in their purchasing process, and (2) whether ecolabels add incremental in-formational value to popular brands, altering customers’ perception of a productmain features. We focus the attention on the agri-food sector, as it feats a highnumber of ecolabels thus offering an ideal setting their influence on customers’purchasing behaviours. The method of analysis is based on a survey, directly ad-ministered to a random sample of customers. A factor analysis is run to buildhomogeneous clusters of customers, according to the impact that ecolabels haveon their purchasing habits.

Title: What makes you feeling old? An analysis of the factors influencing perceptionsof ageing

Presenter: Bruno Arpino, Universitat Pompeu FabraCo-author(s): Valeria Bordone, Alessandro RosinaAbstract: Using a representative sample of people aged 65-74 in Italy, we studied gender

and educational differences in who feels old and the reasons to feel old. Womenare more likely to feel old and to think that the society considers them to be oldthan men. While men feel old mainly when they retire, women mainly associatethis feeling with loosing physical autonomy, widowhood, and absence of projects.However, both men and women report having felt old when turning 65. Havinggrandchildren reduced the likelihood to report boredom as a reason to feel oldamong both men and women. We also found that high educated are less likelyto associate ageing with loneliness and boredom, but more likely to link “feeling

84

old” with absence of projects as compared to their lower educated counterparts.

Title: A (partial) solution to the intractability of APC modelsPresenter: Gustavo De Santis, Università di FirenzeCo-author(s): Massimo MucciardiAbstract: APC models, which try to detect the influence on social behaviour of age, period

and cohort, while intellectually stimulating, suffer from the intrinsic collinearityproblem (period-age=birth cohort), which has greatly limited their use. However,in line with a recent approach to the study of cultural distances (known as DBS-EM, or Distance Between Strata estimated with the Expectation Maximizationalgorithm) differences in terms of age, period and cohort can be treated as dis-tances, i.e., as absolute values. This eliminates their collinearity and, albeit witha few limitations, permits researchers to introduce all the three (APC) as predic-tors in regression models. An example on Italian cultural data (1993-2013) isprovided.

Title: Partner reunification of first generation immigrants in LombardyPresenter: Giuseppe Gabrielli, Università di Napoli Federico IICo-author(s): Anna Paterno, Laura TerzeraAbstract: Lombardy is the Italian region where the foreign presence is more numerous than

elsewhere, showing a continuous process of stabilization. Keeping in mind therelevant literature, we apply a “life histories” approach to data coming from 2010ORIM Survey to describe, in a gender perspective, how the migration event affectsthe couples’ reunification process. The results show different patterns and timingfor men and women and that the differences are mostly linked to the key roleplayed by the citizenship and by the religion.

Contributed Session CON-21 • Friday 10, 17:30-18:45 • Room: A

STATISTICAL INFERENCEChair: Antonello Maruotti

Title: Median bias reduction of maximum likelihood estimates in binary regressionmodels

Presenter: Euloge Clovis Kenne Pagui, Università di PadovaCo-author(s): Alessandra Salvan, Nicola SartoriAbstract: With small sample sizes or sparse data, binary regression models may encounter

the problem of complete or quasi-complete separation. This means that one ormore of a model’s covariates are highly predictive for a binary outcome. As aconsequence, all or some components of the maximum likelihood estimates areinfinite. Firth’s (1993) bias prevention method provides a solution to this problemby modifying the score equation. The correction, however, depends on the cho-sen parameterization. Here, we consider an alternative modification of the scoreequation whose solution is an approximately median unbiased estimator for eachparameter component. The proposed method respects equivariance under repa-rameterizations and also solves the infinite estimates problem. We apply the newapproach to a dataset from U.S.presidential elections and illustrate a simulationstudy comparing the new estimator with maximum likelihood and Firth’s (1993)

85

bias reduced estimators.

Title: On penalized likelihood and bias reductionPresenter: Nicola Lunardon, Independent researcherCo-author(s): Gianfranco AdimariAbstract: We propose a likelihood function endowed with a penalization that reduces the

bias of the maximum likelihood estimator in regular parametric models. Thepenalization hinges on the first two derivatives of the log likelihood and can becomputed numerically. The behavior and the sensitivity to nuisance parametersof the penalized likelihood and derived quantities are addressed. An illustrationis provided in survival models for stratified censored data.

Title: Population size estimation and heterogeneity in capture-recapture count dataPresenter: Antonello Maruotti, Libera Università Maria Ss. AssuntaCo-author(s): Orasa Anan, Dankmar BohningAbstract: The purpose of the study is to estimate the population size under a truncated

count model that accounts for heterogeneity. The proposed estimator is basedon the Conway-Maxwell-Poisson distribution. The benefit of using the Conway-Maxwell-Poisson distribution is that it includes the Bernoulli, the Geometric andthe Poisson distributions as special cases and, furthermore, allows for heterogene-ity. Parameter estimates can be obtained by exploiting the ratios of successivefrequency counts in a weighted linear regression framework. We further providean empirical comparison of bootstrap methods and exact estimation for produc-ing standard errors and confidence intervals for the population size, under theConway-Maxwell-Poisson distribution. These methods are illustrated and con-trasted in a simulation study and in benchmark real data examples.

Contributed Session CON-22 • Friday 10, 17:30-18:45 • Room: B

SURVEY METHODSChair: Renato Salvatore

Title: Italian consumers’ food risks perception: an approach based on the correspon-dence analysis

Presenter: Anna Pinto, Istituto Zooprofilattico Sperimentale delle VenezieCo-author(s): Erlis Ruli, Stefania Crovato, Laura Ventura, Licia RavarottoAbstract: The Italian consumers’ perception towards food risks is a topic widely studied

and analyzed in the recent decades, both in the national and international context.In view of the recent increase in the flow of foreign populations in Italy, newfoods appeared on the Italian food market. A national survey was conductedin order to investigate the consumers’ perception towards these new foods. Thecorrespondence analysis allowed us to explore and summarize the relationshipsamong the variables investigated. Two components that explain the 69.77% ofthe total variance were identified and detailed in the present study.

86

Title: Spatial-temporal multivariate small area estimationPresenter: Renato Salvatore, Università di Cassino e del Lazio MeridionaleCo-author(s): Fabio CappuccioAbstract: A small area estimation area-level spatio-temporal multivariate model is intro-

duced. The residual maximum likelihood estimation procedure is adopted, inorder to apply the model and assessing by simulations the property of the model.Diagnostics and influence analysis issues are also analyzed. The focus is on therelationships between multivariate survey population characteristics in the smallareas, and the model spatial correlation and time autocorrelation parameters.

Title: Is the Smartphone Participation Affecting the Web Survey Experience?Presenter: Daniele Toninelli, Università of BergamoCo-author(s): Melanie RevillaAbstract: The last years’ worldwide spread of mobile devices (smartphones and tablets)

considerably encouraged the mobile participation to web surveys. These devicesare different from PCs, e.g. in terms of screen size and portability. In particular,we expect that the higher portability makes respondents more likely to participatefrom public spaces and/or in the presence of other people. This could affect thesurvey answers, mostly when sensitive topics are asked. This paper focuses onthe comparison of PCs and smartphones and is based on a two-wave experimentthat involved 1,800 panellists for Spain of the Netquest opt-in panel. We studiedto what extent the locations for the PC and the mobile participation are differentand how this factor can affect how respondents felt about the participation itself.

87

Authors index

Aassve, A., 6Accordini, S., 5Adelfio, G., 39, 77Adimari, G., 86Agasisti, T., 77Alfo’, M., 74Alibrandi, A., 76Ambrogi, F., 66Amendola, A., 52Amenta, P., 56Anan, O., 86Andreano, M., 80Andreano, M. S., 80Andreis, F., 1, 9Angelini, C., 4Angotti, R., 36Angrisani, M., 72Antolini, L., 66Antonelli, B., 47Antonicelli, M., 84Arbel, J., 16Arbia, G., 38Ardu, S., 21Arezzo, M., 72Argiento, R., 12Arpino, B., 6, 43, 44, 84Arsenis, S., 57Atella, V., 72Attanasio, M, 69Aversa, M., 35Avner, B., 60

Bacaro, G., 14Bacci, S., 70Bachelet, M., 8Baldacci, E., 1Baldi Antognini, A., 79Balia, S., 67Balzanella, A., 40Bar-Hen, A., 60Barbara, C., 7

Barbati, G., 21, 75Barbiano di Belgiojoso, E., 64Barbieri, G., 83Barcaroli, G., 27Bartoloni, E., 82Bartolucci, F., 70Battiston, M., 12Beamonte, M., 81Beaumont, J., 2Beh, E., 57Bellani, D., 18Bellio, R., 13Benassi, F., 54Benedetti, R., 80Bennani, Y., 60Berger, Y., 10Bernardi, M., 41, 42, 48, 52, 55, 73Bernasconi, D., 66Bertaccini, B., 30Bia, M., 37Bianchi, A., 23Bianchi, G., 27Biffignandi, S., 23Biggeri, L., 69Bignozzi, V., 73Billio, M., 2Bini, M., 42Bisaglia, L., 42Bocchetti, D., 47Bocci, C., 48, 82Bohning, D., 86Bondonio, D., 37Bono, F., 20Boracchi, P., 66Bordone, V., 84Borgi, M., 2Boscaino, G., 77Bossoli, D., 43Boswijk, P., 55Brombin, C., 76Brunetti, M., 72

89

Bruni, R., 27Brutti, P., 81, 83Buondonno, A., 65Busetta, G., 23

Cafarelli, B., 20Calì, C., 43Calabrese, R., 54Calace, D., 84Calciano, L., 5Calculli, C., 20Caloffi, A., 11Calzaroni, M., 65Cameletti, M., 53Camminatiello, I., 70Campolo, M., 19, 23Canale, A., 41, 81Candila, V., 52, 80Cannas, M., 43, 44Capasso, A., 53Capezza, C., 47Capogna, S., 47Cappelli, C., 79Cappuccio, F., 87Capursi, V., 77Carbonara, M., 77Cardone, P., 35Carfagna, E., 1, 26Carpita, M., 56, 72Casacchia, O., 54, 63Casarin, R., 2, 55Cascini, E., 78Castellanos, M. E., 58Cerchiello, P., 2Cers, E., 71Cerulli, G., 37Chiodi, M., 39Chiodini, P., 67Chirico, P., 79Ciavolino, E., 72Cindolo, L., 67Cocchi, D., 20Colangelo, L., 19Coli, A., 44, 69Colombi, R., 34Commowick, O., 75Conti, C., 63Conti, P., 1Contini, D., 67

Conversano, C., 44Conzo, P., 6Corrado, P., 53Corsi, M., 7Cosma, A., 44Costa, M., 79Costola, M., 2Cozzi, S., 38Crippa, F., 23Crosato, L., 31Crovato, S., 86Cruz, D., 37Cugliari, J., 61Cugmas, M., 10Cugnata, F., 76Cusatelli, C., 65Czellar, V., 56

D’Agostino, A., 51, 68D’Agostino, L., 35D’Ambra, A., 47, 56D’Ambra, L., 56D’Ambrosio, D., 50D’Epifanio, G., 72D’Esposito, M., 11D’Esposito, M. R., 10Dai, H., 50Danesi, I., 45, 51Davino, C., 74Dawid, A., 24De Angelis, L., 3De Angelis, M., 47De Blasi, P., 26De Candia, G., 77De Cantis, S., 19De Castris, M., 38De Chiara, G., 47De Iorio, M., 4De Luca, G., 80De Rosa, D., 32De Rose, A., 35, 69De Santis, G., 78, 85De Stefano, D., 11Decarli, A., 4Del Re, F., 53Delucchi, L., 14di Bella, E., 7, 21Di Bella, G., 64Di Consiglio, L., 29

90

Di Lenarda, A., 75Di Marzio, M., 58Di Nardo, E., 34Di Palma, M., 31Di Pino, A., 18, 19Di Salvo, F., 21, 39Di Serio, C., 4, 76Di Stefano, C., 24Dickson, M., 38, 39Diday, E., 29Dobewall, H., 7Drago, C., 83Dunson, D., 3, 41Durante, D., 3Durante, F., 20, 21

Egidi, L., 16Enea, M., 68Erto, P., 46, 47Espa, G., 38Esping-Andersen, G., 18

Falorsi, P., 1Falorsi, S., 27Faroughi, P., 17Fattorini, L., 13Faustini, L., 8Favaro, S., 12, 16Fedele, F., 46Feng, Q., 59Fensore, S., 58Ferligoj, A., 10Fernandes, T., 37Ferrante, M., 19, 75, 76Ferrario, M., 76Ferraro, M., 30Ferretti, F., 13Ferretti, M., 14Filipponi, D., 38Fiorillo, F., 23Franceschini, C., 74Francesco, G., 15Francis, B., 17Fuochi, G., 6Furfaro, E., 9Fusco, D., 29

Gabrielli, G., 55, 85Gabrielli, L., 45

Gallo, G., 2, 15, 55García-Escudero, L., 24Gargallo, P., 81Garofalo, G., 64Gasperoni, F., 21Gelfand, A., 81Genna, S., 47Gerlach, R., 63Gerolimetto, M., 42Ghellini, G., 68Ghiglietti, A., 25Giacalone, M., 65, 76Giacomarra, M., 20Giaimo, R., 20Giambalvo, O., 70Giambona, F., 45, 68Giannerini, S., 15, 62Giordani, P., 30Giordano, G., 49Giordano, S., 34Giorno, V., 52Giovannelli, A., 15Giudici, P., 1, 2Giuliani, D., 38, 39Giummolè, F., 16Giusti, A., 30Giusti, C., 8Gong, M., 40Goracci, G., 62Gottard, A., 34Goude, Y., 61Grassini, L., 30Grazian, C., 58Greco, L., 18, 71Greselin, F., 24Grilli, L., 70Grimaccia, E., 33Grossetti, F., 22Grossi, L., 73Grund, T., 10Guadagnini, A., 31Guagnano, G., 72Gubbiotti, S., 81Guglielmi, A., 4Guindani, M., 2Guizzi, V., 82

Hannig, J., 59Haziza, D., 2, 9

91

Heins, F., 54Hoti, V., 17Hsu, H., 15

Iannario, M., 18, 34Ichino, M., 30Ieva, F., 21, 22, 25, 75, 77Iezzi, D., 24Ing, C., 15Ismail, N., 17Iyer, H., 59

Kabzinska, E., 10Kenett, R., 31Kenne Pagui, E., 85Kirschstein, T., 25Koskinen, J., 10Kosmidis, I., 13Krejci, I., 21Kroonenberg, P., 17, 57

La Sala, G., 53Lagona, F., 57Laloe, T., 61Lanza, M., 53Lanzotti, A., 53Lariccia, F., 19Laureti, T., 8Leckie, G., 70Leone, C., 47Leporatti, L., 21Lepore, A., 46, 47Lepore, M., 46Ley, C., 57Li, F., 37Liccardo, F., 53Liebscher, S., 25Lipizzi, F., 54Liseo, B., 12, 28Liu, X., 59Lombardo, R., 57Longobardi, M., 43Longobardi, S., 68Loni, A., 13Loperfido, N., 74Lorenti, A., 35Loriga, S., 27Lovisolo, F., 31Lucadamo, A., 18

Lunardon, N., 86Luppi, F., 6Luta, G., 71

Maggino, F., 8Magrini, A., 46Maillet, B., 55Mamede, R., 37Mameli, V., 16Mancini, P., 19Manisera, M., 34Mao, X., 56Marcaletti, F., 36Marchese, G., 81Marchetti, S., 8Marella, D., 1, 26Mariani, M., 82Marino, M., 73Marrone, R., 47Martín-Fernández, J., 31Martelli, C., 64, 65, 70Martoni, R., 76Maruotti, A., 57, 85, 86Masci, C., 77Masserini, L., 42Mastrantonio, G., 46, 58Mastrovita, M., 53Mayo-Iscar, A., 24Mazza, A., 55Mazziotta, M., 7McLachlan, G., 24Mecatti, F., 1, 9Menafoglio, A., 31Menardi, G., 83Menardi, M., 83Mencarini, L., 6Mercatanti, A., 37Micheletti, E., 69Milito, A., 76Miller, C., 40Mingo, I., 33Mira, A., 13Missiroli, S., 26Montanari, G., 22Montanari, G. E., 21Monti, A., 71Morrone, D., 84Mucciardi, M., 85Mueller, T., 10

92

Muliere, P., 13Musella, F., 47, 82Musio, M., 24Mussini, M., 73

Naccarato, A., 27Natale, L., 55Nedoluzhko, L., 18Nicola, G., 2Nicolardi, V., 69Niglio, M., 15Nipoti, B., 3, 16Nowok, B., 6Nuccitelli, A., 83Nurra, A., 27

O’Donnell, R., 40Oneto, G., 70

Paccagnella, O., 18Paci, L., 81Pacini, B., 44, 69Padellini, T., 83Paganoni, A., 22, 25, 77Paliotta, A., 24Pallotti, F., 13Palma, R., 70Palumbo, B., 46, 47, 53Palumbo, F., 25, 56Paluzzi, E., 54Panarello, D., 23Pandolfi, S., 22Pandolfo, G., 25Pankaj, Y., 5Panzera, A., 58Pappada’, R., 16Pappalardo, L., 45Parroco, A., 19, 76Paterno, A., 63, 85Pauli, F., 16, 59Paulon, G., 4Pavan, S., 18Pelagatti, M., 62, 78Pelle, E., 26Pellegrini, G., 38Peluso, S., 13Pennoni, F., 34Percuoco, C., 53Perri, P., 26

Petrella, L., 48, 52, 55, 73Petrucci, A., 48Petrucci, P., 54Piacentino, D., 39Piacenza, F., 51Piccolo, D., 34Pierini, A., 27Pignotti, E., 20Pini, A., 78Pinnelli, A., 19Pinto, A., 86Pirralha, A., 7Plaia, A., 21Poggi, J., 61Poggioni, F., 48Polli, C., 36Pollice, A., 46Pons, F., 45Porciani, L., 8Porcu, M., 45Portas, L., 5Porzio, G., 24, 25Postiglione, P., 80Pramov, A., 48, 49Pratesi, M., 42Primerano, I., 49Proietti, T., 15, 62

Racioppi, F., 36Racugno, W, 58Ragozini, G., 11, 25, 50Rahbek, A., 56Rampichini, C., 70Ranalli, M., 74Ranalli, M. G., 1Ravarotto, L., 86Rea, C., 45Redko, I., 60Restaino, M., 50Revilla, M., 87Reynaud, C., 63Riani, M., 57, 71Riccardi, G., 45, 51Riccardini, F., 8, 32Ricciuti, R., 83Righi, A., 82, 83Righi, P., 1, 27Rivellini, G., 36Rivieccio, G., 80

93

Rocchini, D., 14Rocco, E., 48Rodriguez, A., 4Román-Román, P., 52Romano, A., 65Romano, E., 39, 40Romano, R., 25Rondinella, T., 33Rosina, A., 84Rossetti, S., 32Rossi, F., 11Rotolone, D., 24Ruggieri, M., 21Ruggiero, M., 25, 26Ruiz, E., 56Ruiz-Gazen, A., 2Ruli, E., 41, 51, 71, 86Russo, A., 84Russo, M., 11

Salamone, S., 27Salaris, L., 64Salinari, G., 78Salvador, M., 81Salvan, A., 13, 85Salvati, N., 51, 74Salvatore, R., 86, 87Samaritani, A., 65Sangalli, L., 40Santarcangelo, V., 65Santi, F., 39Sarnacchiaro, P., 70Sartori, N., 13, 71, 85Scagnetto, A., 75Scalvini, S., 22Scarnò, M., 27Schirripa Spagnolo, F., 51Scognamillo, A., 80Scott, M., 40Scricciolo, C., 17Scrucca, L., 74Secchi, P., 31Secchi, P. C., 41Secondi, L., 8Serino, M., 50Sforzi, A., 13Shoval, N., 19Simonacci, V., 31Simone, R., 34

Sinagra, G., 75Soscia, M., 52Spina, S., 52Spreafico, L., 78Squittieri, B., 66Stamm, A., 75Storti, G., 52, 63Strozza, S., 55, 63, 69Sulis, I., 45, 70

Tagliaferri, F., 47Talucci, V., 33Tancredi, A., 28Taylor, C., 58Tedesco, N., 64Teh, Y., 12, 16Terzera, L., 64, 85Todorov, V., 31Tong, H., 15Toninelli, D., 53, 87Torelli, N., 16, 62Torres-Ruiz, F., 52Tucci, E., 63Tuoto, T., 29

Ungaro, P., 33

Vagheggini, A., 79Valeinis, J., 71Valentini, A., 44, 77Valsecchi, M. G., 66Vanacore, A., 53Vannini, I., 24Vantini, S., 75, 78, 81Vastola, V., 84Veiga, H., 56Velina, M., 71Ventura, L., 16, 51, 71, 86Venturi, S., 44Verde, R., 29, 40Vicard, P., 26, 82Vietti, A., 78Viroli, C., 3Vistocco, D., 73, 74Vitale, C., 15Vitale, V., 82Vitiello, L., 47Vitolo, B., 53Vives-Mestres, M., 31

94

Viviani, A., 1

Walker, S., 26Wang, C., 59Wang, H., 50Warfield, S., 75

Xifara, T., 4

Zaccarin, S., 11Zagoraiou, M., 79Zanin, L., 54Zavanella, B., 31Zenga, M., 23Zirilli, A., 76Zuccolotto, P., 34

95