Spite, Stigma, and Development Karla Hoff Presentation for Trento Festival of Economics May 29-June 1, 2009 1.

Spite, Stigma, and Development

Karla Hoff

Presentation for Trento Festival of Economics

May 29-June 1, 2009

1

Distributional preferences can be ‘learned’

• A famous paper identified disparate distributional (i.e. other-regarding) preferences across groups:

– “Economists free ride, does anyone else?” --Marwell and Ames 1981

– But that paper couldn’t distinguish between• The self-selection into a discipline (i.e. selfish people choose economics) & • The learning that the discipline provides (i.e. learning economics induces greater

selfishness)

• A natural experiment at Yale Law School shows that exposure to economics shapes preferences

– First year Yale law students are randomly assigned to instructors in required courses. Some instructors have PhDs in economics.

– Students exposed to economists as instructors display greater selfishness and less concern with distribution in experimental games one year later

2

The questions I address:

- How does social structure shape social identities (in particular, distributional preferences)?

- Why do they matter in practice for development?

• To address these questions, we need to find a case of exogenous (i.e., fixed, not earned or chosen) social identities

3

The caste system of rural North India is a good setting for studying this

• Stark historical differences in rights between the high status (H) and low status (L) castes, who were “untouchables”

• Caste membership is inherited

• Caste rank at the extreme ends of the hierarchy has been rigid for hundreds of years

• YET, overlap in wealth and education distribution between individuals in high and low castes today

– thus we can compare effect of caste status controlling for wealth & education

4

The location

• Uttar Pradesh, one of the poorest states in India

5

Experiment I: Social identity & coordination• 10 periods. In each period, a player has an endowment of 6 and chooses:

– To cooperate by investing all in a joint project

– To defect (i.e. keeping 4 tokens for himself and contributing 2)

• After each period, he learns his partner’s decision.

– If a player cooperates, he gets a high payoff if his partner cooperates but a low payoff (the ‘loser’s payoff’) if his partner defects

– If a player defects, he gets the same moderate payoff no matter what his partner does

• Mutual cooperation is an equilibrium; and so is mutual defection

• Matching condition: for 5 periods play with person of same caste status, for 5 periods play with person of different caste status. Anonymity always maintained, but player knows his partner lives in the player’s own village and whether he has high or low caste status

• Experiment mimics situation of coordination on drainage, on timing of sowing, on forming a cooperative or a new convention or a new institution.…

6

7

Player sees full history of the game in his game box

Period 1 Period 2 Period 3 Period 4 Period 5

Partner’s action

Player’s own payoff from the joint project

Player’s action

8

Most low caste pairs achieve success in coordination

Period 5 Period 10

0.8 0.8

LL

0.8 0.8Proportion 0 Proportion Failure Success

Proportion 0 Proportion Failure Success

Coordination success

Collective coordination failure

Individual coordination failure

9

Does anyone else?

10

High caste pairs do not.

HH

Failure 0 Success Failure 0 Success

Period 5 Period 10

11

Mixed status pairs are in-between.

LL

HL

HH

Failure 0 Success Failure 0 Success

Period 5 Period 10

12

How in theory do people coordinate?

• The challenge is to establish beliefs that will support cooperation

• The challenge is met in dynamic play if players can establish a reputation for cooperation – Fudenberg and Levine 1989 (“Reputation and equilibrium

selection…”)

13

Actual period-by-period outcomes (cohort 1)

LL

1

2

3

4

5

HHPeriod

Failure 0 SuccessFailure 0 Success14

Key finding

High- and low-caste players differ only slightly (not stat. significantly) in probability of cooperation in initial period, but respond very differently to the ‘loser’s payoff.’

A high caste person who obtained the “loser’s” payoff in the first period of a pairing is 35 percent less likely than a low caste person to continue to cooperate in the next period. Thus he cannot establish a reputation for cooperation

15

Probability of Cooperation in Periods 2 & 7

Preceding period outcome was: All (D,D) (D,C) (C,D) (C,C) High caste -0.162 -0.0982 -0.363 -0.349** 0.115 (0.116) (0.157) (0.239) (0.162) (0.0981) Mixed status treatment 0.0977 0.0686 -0.206s 0.341* 0.0392 (0.0831) (0.291) (0.250) (0.198) (0.0685) High caste*mixed -0.0153 0.0964 0.205 -0.201 -0.157 (0.156) (0.379) (0.254) (0.457) (0.156) Secondstagebase -0.126 0.137 0.0701 -0.0802* (0.0960) (0.248) (0.117) (0.0454) (D,C) in period 5 0.134 0.0494 0.0144 -0.00973 (0.106) (0.310) (0.247) (0.116) (C,D) in period 5 0.204* 0.0749 0.141 0.0665 (0.106) (0.387) (0.0992) (0.0882) (C,C) in period 5 0.268*** 0.170 0.127 0.112*** (0.0500) (0.346) (0.137) (0.0359) Cooperated in period 1 0.187*** -0.000504 0.165 0.159*** (0.0609) (0.372) (0.162) (0.0465) Observations 242 361 54 54 92 p-value of 2 test 8.53e-11 0.188 0.0324 0.0000 0.000279

16

Is high caste behavior after earning the ‘loser’s payoff’ a response to ‘loss of face’?

Earn ‘loser’s payoff ‘ in initial period

Perceive an insult Perceive a misunderstanding

Defect in next period Cooperate in next period

as retribution to establish trust

17

To provide more direct evidence that we are measuring differences in preferences rather than differences in trust, we conduct another experiment a year later in the same set of villages.

18

Experiment II. Social identity and spite

• Binary choice dictator games to estimate distributional preferences of H and L individuals

• Anonymous pairings; subjects know only that their pattern is a high caste person from their village, or a low caste person from their village

• Double blind

• Volunteer monitor who checks that the experiment really is double blind

19

Trays in binary choice dictator games

20

Which do you want?

90 for partner, 90 for self, or70 for partner, 90 for self (spiteful option)

N = 7621

Which do you want?

180 for partner, 110 for self or100 for partner, 100 for self (spiteful option)

N = 9022

150 for partner, 120 for self, or

100 for partner, 130 for self (less charitable option)

150 for partner, 150 for self, or100 for partner, 160 for self

(less charitable option)

N = 60N = 74

Charity: Which do you want? Charity: Which do you want?

23

Conclusion so far

For low caste individuals, social preferences we estimate from these games are similar to those estimated for university students in the US and Barcelona (Charness and Rabin 2002)

For high caste individuals, social preferences are not the same—they are more spiteful.

24

May make sense in a feudal environment:

“In traditional societies, he who has political power soon

acquires wealth as a kind of consequence. ..[But ] in

commercial and industrial societies, …the best way to make

money is to make money. It is quite possible to do this

without acquiring … power.”

Gellner, Conditions of Liberty 1994

25

Experiment III. Stigma & altruistic punishment of norm violations

• Two individuals (“a trustor” and “a trustee”) play a sequential exchange game

• A disinterested third party has the opportunity to punish the trustee.

• The three players than interact come from distant villages and play the game in their own villages.

• Experiment mimics informal sanctions of opportunists, cheaters, & free-riders. Informal sanctions play a major role in the restraint of opportunism in markets and in collective action. People face these situations every day

26

Treatment One

Trustor and third party punisher are from the same specific caste (endogamous group), which we convey by using last names

27

Mean punishment for defection

4.82

2.87

2.07 2.22

0

1

2

3

4

5

6

HHH LLL HHHBC LLLBC

Me

an

re

lati

ve p

un

ish

me

nt 4.82

2.87

2.07 2.22

0

1

2

3

4

5

6

HHH LLL HHHBC LLLBC

Me

an

re

lati

ve p

un

ish

me

nt

28

Do richer individuals punish more?(house type) .

4.47 4.37

2.66

1.89

0

1

2

3

4

5

6

Lives in a mud house Lives in a brick house or amixed mud and brick house

Me

an

re

lati

ve p

un

ish

me

nt

Punisher is high caste Punisher is low caste

4.47 4.37

2.66

1.89

0

1

2

3

4

5

6

Lives in a mud house Lives in a brick house or amixed mud and brick house

Me

an

re

lati

ve p

un

ish

me

nt

Punisher is high caste Punisher is low caste

29

Do richer individuals punish more?(land ownership)

• .5.05

4.12

2.35 2.24

0

1

2

3

4

5

6

Owns below median land Owns at least median land

Me

an r

ela

tive

pu

nis

hm

en

t

Punisher is high caste Punisher is low caste

5.05

4.12

2.35 2.24

0

1

2

3

4

5

6

Owns below median land Owns at least median land

Me

an r

ela

tive

pu

nis

hm

en

t

Punisher is high caste Punisher is low caste

30

Do income differences explain the punishment pattern?

-10

-50

510

15

-.4 -.2 0 .2 .4 .6predicted consumption

95% CI Fitted valuespd

If anything, punishment levels are lower by richer than poorer individuals

. (Consumption is predicted from LSMS by land owned, house quality, caste, & education)

31

Hypothesis: In-group affiliation differs between high and low castes

Vast experimental literature shows that people favor their own community (“in-group”)

Now, in order to dominate a group thoroughly, the group has be pulverized and atomized (Gellner1994). The relentless stigmatization, and the social restrictions historically imposed on the low castes – exclusions from public celebrations, bans on marriage ceremonies, etc – make sense because they hindered the low castes from developing positive and cohesive group identities 32

Treatment Two

Trustor and third party punisher are from different specific castes

33

Mean punishment for defection

4.82

2.87

2.07 2.22

0

1

2

3

4

5

6

HHH LLL HHHBC LLLBC

Me

an

re

lati

ve p

un

ish

me

nt 4.82

2.87

2.07 2.22

0

1

2

3

4

5

6

HHH LLL HHHBC LLLBC

Me

an

re

lati

ve p

un

ish

me

nt

Treatment One--Punisher and victim from the same specific caste

Treatment Two--Punisher and victim from different specific castes

34

Contributions of this work: -Identifies new influences on distributional preferences, & -New reasons why they matterAffiliation to a high versus a low caste shapes distributional preferences; these are ‘learned’

Concern of high caste with “loss of face” engenders spite and can block coordination on new arrangements that could make everyone better- off

Social exclusion of low caste hurts in-group affiliation and the capability to punish opportunism that hurts one’s group

Possible consequence: Difficulty in enforcing contracts and sustaining collective action. Perpetuation of social hierarchy even if all other obstacles to the low caste are removed

35

36

Thanks to my coauthors

Ernst Fehr

Mayuresh Kshetramade

Priyanka Pandey

Papers

Ernst Fehr, Karla Hoff, and Mayuresh Kshetramade, “Spite and Development,” American Economic Review, Papers and Proceedings, May 2008.

Karla Hoff, Mayuresh Kshetramade, and Ernst Fehr, “Caste and Punishment: The Legacy of Caste Culture in Norm Enforcement,” manuscript, 2009

Karla Hoff and Priyanka Pandey, “Loss of Face as an Obstacle to Coordination,” in progress.

37