Inequality, networks and distributive decisions: A field experiment

1
Inequality, networks and distributive decisions
A field experiment
ESA World Meeting Rome, June 30, 2007

• Ben DExelle
• University of Antwerp - IOB
• Maastricht University
• Arno Riedl
• Maastricht University

2
Research questions

• What determines giving in small-scale
  (face-to-face) societies
• Social networks social distance, network
  structure, individual position within networks
• Individual characteristics of community members
  (economic variables, sex, age, education, etc.)
• Correlation between both dimensions?
• Use of a dictator game experiment, complemented
  with survey data on individual characteristics
  and social networks

3
Literature references

• Social networks and dictator game giving
• Leider, Mobius, Rosenblat Do (2006) dictators
  give 50 more to friends than to strangers
• Goeree, McConnell, Mitchell, Tromp Yariv
  (2006) Distance in networks lowers giving
• Brañas-Garza, Cobo-Reyes, Paz Espinosa, Jiménez
  Ponti (2006) Social integration (betweenness)
  increases giving.

4
Experimental design procedures

• What is new about our experiment
• Large heterogeneity rural village where poverty
  and inequality are highly present (economic
  inequality sex education age)
• Multiple relations details on the type of
  relation (15 types)

5
Experimental design procedures

• Dictator game (divide 20 coins of 1 c) with 1
  stranger and 5 different (randomly selected)
  village members
• max. earnings two days income (6.7)
• Minimizing reciprocity by one-way anonymity only
  dictator knows the identity of the recipient
• Decentralized setup individual visits (lower
  self-selection lower public exposure no
  communication)
• Reduce experimenter-effect use of rings (to
  maintain weight of boxes) sealing of boxes
  recording of decisions by supervisor
• First, the networks (trust-building with
  participants) then, the experiment in one day
  (to limit contagion)

6
Experimental design procedures
7
Results
2
1
8
Economic resources (between subject analysis)
9
Procedure to study social relations

• Measuring social networks
• Use of small cards, each representing a household
  (on each card the names of both husband and wife)
  
• Do you know the household? (93.5 of all possible
  dyads)
• Do you have a social relation with one of its
  members?
• Give details on the type of relation (e.g. land,
  labor, mutual support, family, religion,
  neighbor, etc.).
• Completeness of networks 100 out of 123 adults
  (81.3)

10
Procedure to study social relations

• Analyzing social networks
• Only household heads. Other members were rarely
  mentioned.
• Members of the same household are always linked
  whatever the type of relation we are looking at.
• OR-networks we symmetrized the adjacency matrix.
  We took the maximum value and missing values were
  eliminated by the non-missing value in the other
  direction.
• Few missing values remained (3.5) conversion
  into zeros.

11
Mutual support network in the village (N 123)
density 0.0321 (matrix average) mean degree
3.919 (3.754) freemans graph centralization
measure 21.73
12
Social distance (general relation)
13
Dep. var. coins to recipient

Coef. S.E.
Sex of the dictator (1 male 0 female)
Land dictator yes recipient yes (dummy)
Land dictator yes recipient no (dummy)
Land dictator no recipient yes (dummy)
Distance
Size of ego-network (dictator)
Number of ties in ego-network (dictator)
Power (Bonacich neg. beta)
Number of decision (min. 1 max. 5)
Constant
14
Dep. var. coins to recipient
General relation General relation
Coef. S.E.
Sex of the dictator (1 male 0 female) 2.049 0.046
Land dictator yes recipient yes (dummy) 1.719 0.130
Land dictator yes recipient no (dummy) 1.501 0.170
Land dictator no recipient yes (dummy) -0.289 0.503
Distance
Size of ego-network (dictator)
Number of ties in ego-network (dictator)
Power (Bonacich neg. beta)
Number of decision (min. 1 max. 5) -0.182 0.097
Constant 8.760 0.000
R-squared 0.070 0.070
Number of observations 280 280
15
Dep. var. coins to recipient
General relation General relation
Coef. S.E.
Sex of the dictator (1 male 0 female) 3.147 1.028
Land dictator yes recipient yes (dummy) 2.200 1.098
Land dictator yes recipient no (dummy) 2.185 1.056
Land dictator no recipient yes (dummy) -0.240 0.431
Distance -1.000 0.364
Size of ego-network (dictator) -10.309 4.382
Number of ties in ego-network (dictator) 0.001 0.004
Power (Bonacich neg. beta) 10.731 4.584
Number of decision (min. 1 max. 5) -0.173 0.107
Constant 13.820 3.344
R-squared 0.1512 0.1512
Number of observations 280 280
16
Dep. var. coins to recipient
General relation General relation Mutual support Mutual support
Coef. S.E. Coef. S.E.
Sex of the dictator (1 male 0 female) 3.147 1.028 2.427 1.026
Land dictator yes recipient yes (dummy) 2.200 1.098 0.803 1.218
Land dictator yes recipient no (dummy) 2.185 1.056 0.595 1.150
Land dictator no recipient yes (dummy) -0.240 0.431 -0.287 0.435
Distance -1.000 0.364 0.408 0.801
Size of ego-network (dictator) -10.309 4.382 -0.094 0.803
Number of ties in ego-network (dictator) 0.001 0.004 0.295 0.118
Power (Bonacich neg. beta) 10.731 4.584 -0.336 0.853
Number of decision (min. 1 max. 5) -0.173 0.107 -0.184 0.111
Constant 13.820 3.344 9.270 1.058
R-squared 0.1512 0.1512 0.1426 0.1426
Number of observations 280 280 280 280
17
Conclusions

• Networks matter for dictator giving
• But, it depends on the networks you look at
• General relation
• Distance and Bocanich centrality
• Mutual support
• Ties within the ego-network (norm-based
  behaviour)
• Sex and economic variables remain important, when
  controlling for networks

18
(No Transcript)
19
Mutual support networks
Distance
Centrality
20
The influence of individual characteristics

• Do dictators take account of their economic
  situation?
• Do female dictators behave differently?

21
Sex of dictator
22
Multivariate analysis
Regression on coins left to recipient
Random effects (panel model) robust standard
errors
23
Mutual support networks
Distance
Centrality
24
Multivariate analysis
Regression on number of coins left to recipient
25
Correlation between networks and individual
characteristics

• Possible correlation between networks and
  individual characteristics (economic assets, sex,
  age, etc.)
• Most important networks that may be influenced by
  individual characteristics are mutual support
  relations, economic relations (any kind of
  economic transaction) and friendship relations

26
Network formation

• Procedure to analyze individual social relations
• We only took account of the household heads.
  People only very occasionally mentioned other
  household members.
• Second, members of the same households are always
  linked whatever the type of relation we are
  looking at.
• We symmetrized the resulting adjacency matrix.
  For each dyad in each of both directions, we took
  the maximum value and missing values were
  eliminated by the non-missing value in the other
  direction.
• After this only a limited number of missing
  values remained, which we converted into zeros.
• Most important networks that may be influenced by
  individual characteristics are friendship
  relations and mutual support relations

27
Multivariate analysis
Regression on number of coins left to recipient
28
Individual network position

• Centrality degree centrality
• But, the links other people have are important
  too
• The links other people have
• Control of access and benefits (structural holes
  Burt, 1992)
• Leadership position (responsibility)
• Searching costs (Coleman, 1990)
• The links other people have in the ego-network
• Reputation effects (Burt, 1992)
• Norm-based behavior (reputation effects, indirect
  reciprocity) importance of mutual support
  networks
• Distinguish between mutual support relations and
  general (whatever type) relations the first are
  more specific and may be related with social norms

29
Wealth ranking of recipient
30
Economic resources (within subject analysis)
31
Sex of dictator and recipient
32
(No Transcript)
33
Mutual support networks
2. Centrality
34
Mutual support networks
2. Centrality
35
Logit regression on directed relations
36
Research questions

• Our interest comes from the current debate in
  development economics on decentralization
• Decentralization delegation to the local level
  of any type of decision-making
• In our case the task to distribute aid resources
  ? less costly but also less control on
  distributive outcomes
• Two ways for the policymaker to maintain certain
  influence on distribution
• One focus monitoring mechanisms
• Other focus ( our focus) influencing local
  determinants behind local distributive processes

37
Experimental design

• First, mapping of social networks confidence is
  built with local participants support of local
  leaders
• Then, experiment in one-day contagion is limited
• Payments to recipients are made the day after
  recipients are likely to have played as dictator
  too (acceptance of payments!)