Measuring Preferences for Redistribution and Fairness: An Experimental Approach

1
Measuring Preferences for Redistribution and
Fairness An Experimental Approach
Ruben Durante Louis Putterman Brown
University
2
Stylized Facts

• Luxembourg income survey data for OECD countries
  shows post-tax-and-transfer income share of
  bottom 40 is 14.7 percentage points above their
  pre-tax-and-transfer share (Milanovic, 2000)
• Even in the U.S. the post-tax income shares of
  the bottom 40 is 7.8 higher than the pre-tax.
• There are considerable differences across
  countries.

3
Big Questions

• Why do we observe redistribution in democracies?
• Why dont we observe more of it?
• Why does the amount of redistribution vary among
  countries?
• These questions motivate our study, but we try to
  address more specific subsidiary questions.

4

• To know what determines the amount of
  redistribution, we need to know what accounts for
  redistribution in the first place.

5
Possible reasons for redistribution

• Simple self-interest (the poor and middle class
  vote to redistribute from the rich).
• Insurance motive (all risk-averse people may
  favor a system of social insurance)
• Desire to reduce inequality
• a. out of fear of social instability
• b. out of social preference (inequality
• aversion)

6
Possible limits on redistribution

• Concern for negative impact on incentives, other
  dead weight losses.
• Belief that inequalities are justly earned.
• Disproportionate political influence of the rich.

7
main goal of our experiment is to investigate
relative importance of

• Simple self-interest (the poor and middle class
  vote to redistribute from the rich).
• Insurance motive (all risk-averse people favor a
  system of social insurance)
• Desire to reduce inequality
• (a. out of fear of social instability)
• b. out of social preference (inequality
• aversion, fairness preference)

8
we also obtain evidence on two of the limiting
factors

• 1. Concern for negative impact on incentives,
  other dead weight losses (although theres no
  production or investment in the experiment, we
  can study concerns over exogenous changes in size
  of the pie)
• 2. Belief that inequalities are justly earned.
• (3. Disproportionate political influence of the
  rich not considered)

9
the experimental literature

• There are many experiments bearing on fairness
  and other distributive preferences (e.g.,
  dictator games, ultimatum games, trust games,
  moonlighting games, Charness Rabins simple
  games, etc.) but usually these involve only two
  or three subjects interacting and have no
  suggestion or reference to the macro/political
  issue of redistribution in society.

10
experiments to measurepreferences for societal
income equalization

• At least half of these (e.g., Amiel and Cowell)
  elicit preferences survey style, rather than in a
  manner that determines actual payoffs.
• Others (Beck Beckman, Formby Smith Ackert,
  Martinez-Vasquez Rider) use real payoffs but
  lack one or more elements of our experiment, such
  as

11

• - the group of affected individuals is large (we
  use group size 20 a dictator)
• - decisions are made both as a disinterested
  observer, as an interested party under
  uncertainty, and as an interested party after
  resolution of uncertainty about own income
• - decisions are made with respect to initial
  incomes determined both arbitrarily and in
  earned income conditions
• - the variation in pre-tax incomes is very large
  (from max of 100 to min of 0.11)

12

• - decisions are made with both different dead
  weight losses and with different private costs to
  the dictator, to study demand (willingness to
  pay)
• - Unlike Frohlich and Oppenheimer and others, we
  dont explicitly ask subjects to think about
  their fairness or redistributive beliefs (until
  the debriefing), but we provide one mild framing
  cue (our pre-tax income distribution mimics that
  of income 5-percentiles in the U.S., and we tell
  our subjects that this is so).

13
tax scheme

• The pre-tax-and-transfer distribution of income
  can be modified by a proportionate tax of 0, 0.1,
  0.2, 0.9, 1.0, with the tax proceeds (or the
  proceeds minus an exogenously imposed dead weight
  loss) redistributed equally among the twenty
  affected subjects.

14

• We use a strategy-type method in which every
  subject potentially dictates the degree of
  redistributive taxation under each condition, so
  as
• - to maximize the amount of data generated
• - to avoid strategic voting issues

15
Why care what determines desires for
redistribution?

• Normative relevance
• If widespread support for a more equal income
  distribution exists, then redistributive taxation
  at reasonable cost unambiguously improves social
  welfare. If redistribution serves the selfish
  interest of some but not others, we need
  interpersonal comparability of welfare to make
  social welfare evaluations.
• Positive relevance
• Understanding to what extent redistribution is
  due to preferences for equality as opposed to
  other factors may help identifying the
  determinants of support for redistributive
  programs in society, and to understand why the
  level of redistribution differs significantly
  among countries.

16
Experimental Design
Two or three sets of decisions are made by each
subject in each session

• Part 1 Disinterested Observer Condition
• Each participant is asked to choose a tax rate to
  be applied to the pre-tax earnings distribution
  among the other 20 participants. At the end of
  the session one person, the decisive
  individual, is randomly selected to determine
  the actual degree of redistribution.

The earning of the D.I. is randomly selected from
the interval (19.80 21.80).
Purpose Evaluate participants redistributive
choices when self-interest is absent (or largely
so). Keep D.I.s identity secret.
17
Experiment Design (Cont.)

• Part 2. Choice with Personal Stake Uncertainty
• The setup is the same as in Exp. 1.
• However, in this case the base earning of the
  decisive individual is determined in the same way
  as those of the other participants. The tax rate
  chosen by the decisive individual applies to
  everyones earning, including the D.I.s.

Purpose Analyze participants redistributive
choices when self-interest but also uncertainty
are present (the usual voting situation).
18
Experiment Design (Cont.)

• Part 3. Choice with Personal Stake, Certainty
• Subjects get to reconsider their tax choices in
  Part 2 after learning their exact income rankings.

Purpose Analyze participants redistributive
choices when self-interest is present and
uncertainty is resolved.
19
Pre-tax Earning Distribution
The pre-tax earnings of the twenty participants
are distributed proportionally to the pre-tax
income in the United States in 2000, by
5-percentiles.
20
Pre-tax Earnings Determination

• How is assignment of subjects pre-tax earnings
  ranks determined? Four possibilities
• Randomly
• Based on home area average income, from ZIP code
  or country (Where From WF)
• Based on performance in a game of skill (adapted
  version of Tetris)
• Based on performance on a general knowledge Quiz

Purpose Analyze differences in agents attitude
toward redistribution for different determinants
of income inequality (i.e. luck, parental wealth,
ability, effort).
21
Taxes and Redistribution

• Twenty participants pre-tax earning (xi) are
  taxed at a rate t and the proceedings are equally
  distributed among all agents. The post-tax
  pay-off for an affected subject is
• yi (1 - t ) xi 1/20 (1 - e) (?i t xi) , ? i ?
  iD

Efficiency Loss (e) Proportion of the tax
revenue lost to redistribution (possible values
0, 12.5 25).
22
Tax cost versus efficiency loss

• Tax cost is charged to the D.I. (dictator) only
  it is 0, 25 cents, 50 cents, and 1 dollar per 10
  worth of tax, varying across experiment sessions.
  
• Tax cost affects the D.I. directly and thus lets
  us gauge her WTP for more equality.
• Efficiency loss doesnt affect the D.I. in Part 1
  unless she cares about the sum of others
  earnings (as in Charness-Rabin 02).

23
Sequence of the Session

• Part 1 - Instructions
• Part 1 - Comprehension test
• Part 1 - Decision stage
• Part 2 - Instructions
• Part 2 - Comprehension test
• Part 2 - Pre decision stage (forecast)
• Part 2 - Decision stage
• Quiz
• Tetris training
• Tetris game
• Selection of part paid-off on (Part 1 or Part 2)

24
Sequence of the Session (Cont.)

• if Part 2 is selected
• Provisional earning summary to each subject
• Part 3 Tax Revisions possible
• Method Determination (Random, WF, Quiz, or
  Tetris)
• Risk aversion test
• Decisive individual drawing
• Final earnings summary
• Exit survey
• Background questions
• Political inclination / ideology
• Socio-economic status
• General attitude toward equality and
  redistribution

25
Subject Pool(s)

• We focus on 16 sessions with 336 Brown University
  undergraduates (21 per session) studying in every
  subject area.
• 55 adult non-student subjects were recruited to 4
  other sessions not discussed, but generating
  similar results.

26
Sessions by Tax Cost and Efficiency Loss
of participants in parenthesis
27
Part 1Disinterested Observer
28
Demand for Equality

• Although self-interested dictators should always
  select t 0 when tax cost gt 0, more than 75
  prefer a positive tax (prefer some
  redistribution), and even when tax cost 1 per
  10, 70 prefer some redistribution.

29
Preferences for Equality

• Prediction When tax cost gt 0 we would expect t
  0

Mean Tax Rate 41.4
Mean Tax Rate 33.7
30
Responsiveness to Cost

• Cross subject comparison shows some
  responsiveness to tax cost the higher the cost
  to the D.I., the less redistribution is demanded.

31
Tax Cost and Willingness to Pay
32
Responsiveness to dead-weight loss

• Cross subject comparison shows that the larger
  the fraction of taxed output that is lost to
  inefficiency/leakage, the smaller the demand for
  redistribution.

33
Equality-Efficiency Trade-off
34
Demand for redistributionand the source of
inequality

• Within-subject comparisons show that many
  subjects demand more redistribution when
  inequalities are arbitrary (random and where
  from methods) than when inequalities are earned
  (quiz and Tetris methods).

35
Tax choices differ significantlyacross methods.
36

• However, the tendency to distinguish between
  inequalities of different origin
• is linked to gender
• males distinguish strongly,
• females almost not at all.
• Most of the difference between male and female
  tax preferences is due to different preferences
  under the earned methods.

37
(No Transcript)
38
These and other findingsare supported by
multivariate regressions, including Tobit.

• Demand for redistribution varies
• negatively with cost to the D.I.
• negatively with dead weight loss
• negatively with earned sources of inequality
  (quiz and Tetris)
• positively with female gender
• 5. positively with more liberal political
  philosophy

39

• 6. positively with professed preference for
  equality.
• 7. negatively with number of economics courses
  taken.
• 8. positively with measured risk aversion
• 9. negatively with family income

40
(No Transcript)
41

• The impact of the source of income inequality,
  and the fact that gender differences are mainly
  due to different reactions to this, are shown in
  the following set of regressions.

42
(No Transcript)
43
PART 2Interested Decision-Maker
44

• Recall that in Part 2,
• - the D.I.s tax choice determines the degree of
  redistribution of the pre-tax earnings of 20
  subjects including him/herself.
• - self-interest dictates choosing high (low) tax
  if you think youll rank near bottom (top)
• - D.I. has not yet played Tetris or taken the
  quiz . . . decision-making under uncertainty
• - decision in random condition is like
  Harsanyis choice behind veil of ignorance

45

• - D.I.s earnings are directly affected by both
  tax cost and efficiency loss

46
Results in brief

• Overall distribution of tax choices is very
  similar to that in Part 1
• But self-interest contributes importantly to
  individuals choices where they have guesses
  about their pre-tax income ranks.
• For Tetris, Quiz, and Where From
• ? factors used in Part 1 regressions explain 13
  of variance in tax choices
• ? own expected rank increases explained variance
  to 31, hence adds 18 to explained variance
  (R-Squared).

47
(No Transcript)
48

• Interestingly, subjects expected own-income
  ranks for Where From, Quiz, and Tetris incomes
  also help to predict their Part 1 tax
  choicesalthough the addition to explained
  variance is lower than in Part 2.
• Since subjects did not even know what the Part 2
  treatment would be when making Part 1 choices,
  the explanation is probably that subjects
  identify with people like themselves when
  choosing t (e.g., if I think Id do badly on
  Tetris, Id want the D.I. to select a high tax,
  so this is what I do as D.I., though it doesnt
  affect me personally).

49

• In Part 2 like Part 1, males prefer more
  redistribution under Where From than under Quiz
  and Tetris, while females prefer similarly high
  redistribution under all three methods.

50
females
males
51
Random Pre-tax Incomes

• Tax choices under random incomes (veil of
  ignorance) in Part 2 are significantly higher
  than in Part 1average 54.6 versus
  49.3consistent with self-interest and
  risk-aversion.
• The next regression studies tax choice with
  random pre-tax incomes in both Part 1 and Part 2.

52
(No Transcript)
53

• The regressions show that being more risk averse
  and being involving (being in Part 2) raise
  chosen t, but counter-intuitively the interaction
  between these two variables is insignificant and
  doesnt raise explained variance.

54
Average Part 2 Tax for All 4 Methods
55
Part 3 involved tax choice with certainty
about own rank

• Included only for the 7 of 16 sessions in which
  Part 2 (involved) payoffs were chosen by coin
  toss
• This helps keep D.I.s identity secret ex post.
• Thus made by 147 subjects, 588 decisions (t for
  each of 4 methods x 147).

56
most decisions were fullyconsistent with
self-interest

• After resolution of uncertainty about own rank,
  payoff maximization requires selecting t 0 or t
  100 (depending also on tax cost and dead
  weight loss)
• About 66 of decisions were for own payoff
  maximizing tax rate.
• There are slightly more deviations from a
  payoff-maximizing tax of 0 (toward more equality)
  than from a payoff-maximizing tax of 100 (toward
  less equality), but the difference is slight.

57

• The cost of deviating varies depending on rank
• For example, for subject with highest pre-tax
  income, each 10 of redistribution represents a
  personal sacrifice of about 8 not counting the
  regular tax cost and any dead weight loss. For
  subjects near cut-off rank, personal sacrifice is
  well under 1 per 10 when tax cost is small.
• ? Calculating the individual cost of deviating
  by rank, factoring in tax cost and dead weight
  loss, we find deviations negatively related to
  this cost. (We plan to carry out an integrated
  analysis of response to tax cost using Part 3 and
  Part 1 together, hence covering wider range of
  costs.)

58

• We analyze separately decisions of subjects
  whose income-maximizing tax choice is 0, then
  those whose income-maximizing tax choice is 100.
• (Observations are by subject-and-method.)

59
Deviations from income-maximizing tax of 0

• A quadratic specification with cost of deviation
  alone can explain 20 of the size of deviations
  (including 0 deviations).
• Adding Part 1 tax choice (reflecting pure
  preference for equality) or explanatory variables
  including political philosophy can explain an
  addition 4 to 6 of the variance.

60
(No Transcript)
61
Deviations from income-maximizing tax of 100

• This allows expression of preference for
  inequality under some conditions.
• The quadratic personal cost specification
  explains only 0.3 of these deviations.
• Adding Part 1 tax choice raises R-squared to 4.
  Part 1 tax choice is significant at the 1 level
  in this regression people who choose lower taxes
  for the income-determination method in question
  in Part 1 also choose lower taxes in Part
  3suggesting that they forego some earnings gain
  in support of their belief about just
  inequalities (e.g., right to earn more if did
  better on Quiz).

62

63
Provisional Conclusions

• Most subjects are willing to incur some cost to
  reduce inequality among others
• Willingness to pay for equality is sensitive to
  cost theres much less demand for equalization
  at very high prices
• Uncertainty and risk-aversion is a major driver
  of demand for equality
• Demand for equality is significantly affected by
  the source of the inequalityearned vs. arbitrary.

64

• There are substantial gender effects consistent
  with the political gender gap in the U.S.
• The external validity of the experiment is
  supported by correlations with reported political
  preferences, and by performance of the
  non-student sample

65
To be done

• More integrated analysis of relative weights on
  self-interest, risk-aversion, and social
  preferences (values, fairness preferences) in
  determining tax choices.
• Calculation of implied voting outcomes and
  comparison with observed degrees of
  redistribution in U.S. and other countries.
• Using an explicit Social Welfare Function to
  calculate aggregate utility maximizing tax
  choices.