icse 2014 research track - does latitude hurt while longitude kills?

27
IBM Watson Life ©2014 IBM Corporation 1 IBM Watson Group - Watson Life, 2 IBM Research, 3 Singapore University of Technology and Design June 4, 2014 Patrick Wagstrom 1,2 , Subhajit Datta 3 Does Latitude Hurt while Longitude Kills? @pridkett http://wagstrom.net/ [email protected] @datta_subhajit http://www.dattas.net/ [email protected]

Upload: patrick-wagstrom

Post on 13-Jul-2015

154 views

Category:

Technology


5 download

TRANSCRIPT

IBM Watson Life ©2014 IBM Corporation

1IBM Watson Group - Watson Life, 2IBM Research, 3Singapore University of Technology and Design June 4, 2014

Patrick Wagstrom1,2, Subhajit Datta3

Does Latitude Hurt while Longitude Kills?

@pridkett http://wagstrom.net/ [email protected] @datta_subhajit http://www.dattas.net/ [email protected]

IBM Watson Life :: ©2014 IBM Corporation

Overview

• Origins of Latitude Hurts, Longitude Kills

• Work Item Based Data Collection

• Relationship Between Geographic Distance, Temporal Offset and Response Time

• Relationship Between Geographic Distance, Temporal Offset and Link Formation

• Implications

2

http://comments.gmane.org/gmane.comp.programming.extreme-programming/111068

5 May 2011 from Hennah Sugumaran !Hello guys, I wanted to know the Challenges that the agile developers face in the dispersed environment (i.e. when the developers in a team is spread across the various places). The real time challenges. Hennah.

6 May 2011 from Tim Ottinger !Five rules: 1) Don't 2) Dont' treat remotes as locals. 3) Don't treat locals as remotes4) Latitude hurts, longitude kills 5) Don't always be remote !…. !Fourth one is dead serious. Time zones suck. It's almost nice if you are centrally located to have peers one or two time zones earlier and some one or two time zones later, but a lack of "common hours" really screws up the pairing work.

IBM Watson Life :: ©2014 IBM Corporation 4

7636km geographic distance 1 hour temporal offset 63° Latitude ∆ 27° Longitude ∆

7501km geographic distance 8 hours temporal offset 15° Latitude ∆ 110° Longitude ∆

IBM Watson Life :: ©2014 IBM Corporation

Understanding Work Items

5

IBM Watson Life :: ©2014 IBM Corporation

Work Items - A Mechanism for Conversations

6

IBM Watson Life :: ©2014 IBM Corporation

Work Items - Building Networks

7

Patrick Wagstrom

Sandeep Somavarapu

IBM Watson Life :: ©2014 IBM Corporation 8

Patrick Wagstrom

Sandeep Somavarapu

6326km 6 hours ∆ 63 hours 11 minutes for response

IBM Watson Life :: ©2014 IBM Corporation

65222 Comments from 13399 Work Items

9

IBM Watson Life :: ©2014 IBM Corporation

Most Discussions Have Few Comments

10

IBM Watson Life :: ©2014 IBM Corporation

Many Repeated Interactions

11

IBM Watson Life :: ©2014 IBM Corporation

438 Rational Employees

12

12 Locations 3 Timezones

IBM Watson Life :: ©2014 IBM Corporation

848 Non-Rational IBMers

13

116 Locations 17 Timezones

IBM Watson Life :: ©2014 IBM Corporation

367 Non-IBMers

14

204 Locations 16 Timezones

IBM Watson Life :: ©2014 IBM Corporation

No Clear Relationship Between Geographic Distance and Response Time

15

IBM Watson Life :: ©2014 IBM Corporation

No Clear Relationship Between Temporal Offset and Response Time

16

IBM Watson Life :: ©2014 IBM Corporation

Relationship to Response Time

• Previous Delay - Some conversations just move more slowly

• Repetition - Have the participants corresponded before

• Geographic Distance - How many kilometers between the pair

• Temporal Offset - How many hours between the pair

17

IBM Watson Life :: ©2014 IBM Corporation

Relationship to Response Time

!

!

• Previous delay - if previous communication took a long time, future communications will take a long time

• Repetition - conversations between a pair of developers will speed up future conversations

• Geographic distance - oddly enough the further away two developers are, the faster the response

• Temporal offset - However, time separation quickly swamps this effect in most cases

18

IBM Watson Life :: ©2014 IBM Corporation

Obligatory Sociogram

19

IBMer

Rational

Non-IBMer

IBM Watson Life :: ©2014 IBM Corporation

ERGMs in a Nutshell

• Model that predicts the probability of the presence of an edge in a network

• Generates random networks that try to match network provided

• Prediction factors: • Network topology (general density, mutuality, triad closure, stars, etc)

• Nodal properties (works for Rational, etc)

• Dyadic properties (temporal distance, geographic distance, repeat interactions)

• Still an evolving area of research

• A little like logistic regression for networks

20

IBM Watson Life :: ©2014 IBM Corporation

Obligatory Math Slide

21

IBM Watson Life :: ©2014 IBM Corporation

Very Basic ERGM

!

!

• “Edges” parameter is the density of the network

!

!

!

• 99.75% of edges are expected to be reciprocated

22

IBM Watson Life :: ©2014 IBM Corporation

Geographic Distance Has Little Effect on Edge Creation

!

!

!

• Slight preference to connect to employees of Rational (that’s good!)

• Geographic distance make it slightly less likely that a dyad will be connected

• Not statistically significant

23

IBM Watson Life :: ©2014 IBM Corporation

Temporal Offset Has Little Effect on Edge Creation

!

!

!

• Temporal offset has little effect on the probability of edge creation

• Still a preference to connect to Rational employees

• This says nothing about the time it will take for a response, however

24

IBM Watson Life :: ©2014 IBM Corporation

Strange Interactions Between Temporal Offset and Geographic Distance

!

!

!

• Temporal offset makes two people slightly *more* likely to interact

• Geographic distance makes two people slightly *less* likely to interact

• This is VERY strange...no idea what’s going on

25

IBM Watson Life :: ©2014 IBM Corporation

So, Does Latitude Hurt and Longitude Kill?

• Maybe?

• Network connections show a very strange picture: • Geographic distance makes people less likely to collaborate (marginally signification)

• Temporal distance makes people more likely to collaborate

• Effect is still weak

26

IBM Watson Life :: ©2014 IBM Corporation

Thoughts On Our Findings

• Maybe high quality tools can overcome some of the pain of distributed work

• Maybe the changes in process and flexible working hours really help out

• Team practices may have limited the pain

• Maybe our data just stinks...

27

@pridkett http://wagstrom.net/ [email protected] @datta_subhajit http://www.dattas.net/ [email protected]