1 surveys: using lsms, hbs, lfs and silc for poverty analysis rachel smith-govoni april 4, 2008
TRANSCRIPT
1
Surveys: Using LSMS, HBS, LFS Surveys: Using LSMS, HBS, LFS and SILC for Poverty Analysisand SILC for Poverty Analysis
Rachel Smith-Govoni
April 4, 2008
2
Goals and NeedsGoals and Needs
Goals:
• Measure the poverty impact of economic policy
• Measure the distributional impact of economic policy
Needs:
• Rely heavily on household survey data
3
Household Surveys - typesHousehold Surveys - types
• Single Topic
• Labour Force Surveys( LFS) (ILO)
Census – national, 10 years – Serbia 2002
• In-between
• Multi-topic
4
Household SurveysHousehold Surveys
• Single Topic
• In-between
• Agricultural Surveys (FAO)
• Demographic and Health (DHS)
• Household Budget Surveys (HBS)
• Multi-topic
5
Household SurveysHousehold Surveys• Single Topic
• In-between
• Multi-topic• Multiple Indicator Cluster Survey
UNICEF
• Living Standards Measurement Study
• Survey on Income and Living Conditions (SILC, EU)
6
CensusCensus
• Accurate measure of the population of a country
• Geographic distribution of the population
• Basic demographic information
Purpose
7
CensusCensus
• Not a sample
• Universal coverage
• No sampling errors in estimates
• Some corrections for non-response may be needed
• Not many items
8
CensusCensus
• Demographic information: age, sex, race/ethnicity, family and household composition
• Housing information
• Others: basic education, labour, disability
Content
9
CensusCensus
Limited monitoring
•Albania: 2001 (1989)•BiH 1991 (1981)•Montenegro 2003 (1991)•Serbia 2002•Kosovo 1981
Limited use if looking at impact of policies affecting taxes, tariffs or pricing
10
CensusCensus
• Sample frame
• Link with household surveys for small area estimation (data mapping)
Uses
11
Two types of errors:Two types of errors:
Sampling and non-sampling
• Cost •Time
•Non-response• Training
DECRG: May 7 2004 Sample size
Sampling errorNon-sampling error
Sampling vs. non-sampling errorsSampling vs. non-sampling errors
Total error
13
14
15
Labour Force Survey (Anketa Labour Force Survey (Anketa o radnoj snazi – ARS) o radnoj snazi – ARS)
• Direct measurement of unemployment
• General characteristics of the labour force
Purpose
16
Labour Force SurveyLabour Force Survey
• Relatively large samples
Desire to disaggregate to different geographic areas
• Individuals of working age
Sample
17
Labour Force SurveyLabour Force Survey
• Characteristics of the labour force
– Demographics
– Education
• Sectoral distribution of employment
• Degree of formality
• Seasonal
• Income
Content
18
Labour Force SurveyLabour Force SurveyLimitations:
• LFS typically capture partial, not total, income, under-estimate welfare
• Measurement Error - Labour income measurement error at both ends of the distribution
19
LFS in Latin AmericaLFS in Latin AmericaItem non-responseItem non-response
Salaried Self-employed
Employer
Mean non- response rate
3.9% 10.2% 12.0
Source: Feres, 1998
20
Household Budget Survey Household Budget Survey (Anketa o potrosnji (Anketa o potrosnji domacinstava – APD, domacinstava – APD,
• Inputs to National Accounts on consumer expenditures
• Track changes in expenditures over time
• Weights for the Consumer Price Index
(Indeks Potrosackih Cijena)
21
• Usually medium size sample
• High non-response rates
Sample
Non response rates (Non response rates (Eurostat Eurostat
Household Budget SurveysHousehold Budget Surveys, 2003), 2003)•Bulgaria: 39.7%•Estonia, 44%•Hungary, 58.8% before replacement•Romania, 21.6 %
22
Household Budget SurveysHousehold Budget Surveys
• Total Income
• Total Consumption - diary
• Short Demographics
• Central Europe: agriculture
• Limited health and education
Content
23
Household Budget SurveysHousehold Budget Surveys
• Consumption based welfare measure
• Purpose of an HBS survey is NOT to measure welfare but to precisely measure mean expenditures on specific goods and services
• These are conflicting goals
Poverty Measurement
24
Household Budget SurveysHousehold Budget Surveys
• Shortest possible reference periods
• Minimize number of omitted expenditures
• Good for precise measurement of regional or national means
• Because of lumpy nature of purchases, not good for comparisons among households
Poverty Measurement
25
Multi-topic Household SurveysMulti-topic Household Surveys
Those with a focus on measuring poverty
• Survey on Income and Living Conditions (SILC)
• Living Standards Measurement Study Surveys (LSMS)
26
Multi-topic Household SurveysMulti-topic Household Surveys
• Analysis of welfare levels and distribution
• Study links between welfare levels and individual and household characteristics, economic, human and social capital
• Social exclusion
• Levels of access to, and use of, social services, government programs and spending
Purpose
27
Multi-topic Household SurveysMulti-topic Household Surveys
• Small sample sizes
• Trade-off issue: Quality and cost considerations
• Limits ability to assess programs or policies that affect small groups or small areas (over-sample)
• Infrequent in many countries
Sample
28
LSMS 2002, 2003, 2007LSMS 2002, 2003, 2007
Content1 household composition
2 housing
3 individual demographics
4 health
5 labour
6 work history
7 social programs
8 migration
9 values and opinions
10 consumption
11 agriculture
29
Multi-topic Household SurveysMulti-topic Household Surveys
• Total consumption
– Longer reference periods
– Able to calculate use value of durables and housing
• Total income
– Suffers from standard measurement errors
Poverty Measurement
30
Designs for surveys across time Designs for surveys across time
Repeated cross sectional surveys
(e.g. Household Budget Survey, Labour Force Survey)
• Common design for large government surveys
• New sample drawn for each survey
• Carry similar questions each year
• Used for trend analysis at aggregate level
31
Designs for surveys across time Designs for surveys across time
Cohort Studies
• Sample often based on an age group
• Follow up same sample members at fairly long intervals
• Developmental data as well as social and economic data
• Data from parents, teachers associated with cohort member
32
Designs for surveys across timeDesigns for surveys across time
e.g. Panel Study of Income Dynamics, USA – since 1968!
Living in BiH 2001-2004, LSMS Albania 2002-2004,
LSMS Serbia 2002-2003
• Draw a sample at one point in time and follow those sample members indefinitely (or as long as the funding continues)
• Collect individual level data in household context
• Repeated measures at fixed intervals (annual data collection)
33
Advantages of Panel DataAdvantages of Panel Data• Comparison of same individual over time - outcomes
• Track of aspects of social change
• Facilitates study of change and causal inference
• Minimise the problem of inaccurate recall
• Compare a person’s expectations with real change
• Look at how changes in individuals’ behaviour affects their households
Identifies the co-variates of change and the relative risks of particular events for different types of people
34
Changes in Employment Changes in Employment StatusStatus
A: CROSS-SECTIONAL INFORMATION
Unemployed
Employed
2001 2007
Net change - 0.1% unemployed
35
Changes in Employment Changes in Employment StatusStatus
B: PANEL INFORMATION
Still Unemployed
Still Employed
Unemployed
Employed
2001 2007
Net change - 0.1% unemployed Actual change is 10.1
continuouslyemployed
86.7%
employed 2001but unemployed 2007
5%
continuouslyunemployed
3.2%
unemployed 2001 but employed 2007
5.1%
36
Balkan ExamplesBalkan Examples
Albania - 15% of the unemployed in 2002 had made the transition to formal sector employment by 2004
BiH - About half who were poor in 2001 remained poor in 2004. Many individuals moved out of poverty.
(Cross section headcount 18% for both years)
37
Employment and the labour market Unemployment duration and exit rates
Do the unemployed find stable employment?
The effect of non-standard employment on mental health
Temporary jobs: who gets them, what are they worth, and do they lead anywhere?
Family and Household Patterns of household formation and dissolution
Breaking up - finances and well-being following divorce or split
The effect of parents’ employment on children's educational attainment
38
A SampleA Sample• Concept of ‘longitudinal household’
problematic for a panel - households change in composition over time or disappear altogether
• Individual level sample
39
Following rulesFollowing rules• All members of households interviewed at
Wave One
• Children born to these original sample members
• Original members are followed as they move house, and any new individuals who join with them are eligible to be interviewed
• New sample members are followed if they split from the original member
40
Questionnaire designQuestionnaire design
• Core content carried every wave
• Rotating core questions
• One-off variable components – lifetime job history
– marital and fertility history
• Variable questions to respond to new research and policy agendas
41
Attrition in panel surveysAttrition in panel surveys• Inevitable to some extent but can be
minimised
• Multiple sources of attrition in a panel– refusal to take part
– respondents move and cannot be traced
– non-contacts
• Worry is potential bias if people who drop out differ significantly from those who stay in
42
70
87.7 90.394.9 94.8 97.5 97.2 97
0
10
20
30
40
50
60
70
80
90
100
Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 Wave 7 Wave 8
UK Panel Wave 1 RespondentsUK Panel Wave 1 RespondentsWave-on wave re-interview ratesWave-on wave re-interview rates
43
FieldworkFieldwork• respondent incentives as a ‘thank-you’
• extended fieldwork period for ‘tail-enders’
• refusal conversion programme
• tracking procedures during fieldwork
• panel maintenance between waves– Change of Address cards to update addresses
– mailing of Respondent Report
– details of contacts with respondents between waves
44
The user databaseThe user database
• Longitudinal data is complex
• Provide users with database structure which enhances usability
• Consistent record structure over time
• Key variables for matching and linking data cross wave
• Consistent variable naming conventions
45
ConclusionsConclusions• Longitudinal panel data allows us to answer
research questions that cannot be answered with with cross-sectional data
• Provides a different view of the world - see process through the life-course not just a static picture
• Is complex (but so is the real world) - so needs to be well designed and conducted with sufficient resources to be successful
46
FinalFinal pointspoints
• Welfare: household surveys- always missing the homeless, street children, institutionalized population
• No one survey can meet all needs, review its purpose, coverage, content and quality before using
• Need a system of surveys that meets the needs of data users