Electoral agency in the lab: preliminary findings

1
Electoral agency in the lab preliminary findings

• Leif Helland
• Lars Monkerud
• Rune Sørensen

2
Motivation

• Liberal concept of democracy
• Roots in liberal political philosophy (Mill,
  Locke)
• Democracy negatively defined citizens should be
  free from restraint and exploitation by state
  power a) constitutional constraints, b)
  accountable leadership
• Accountability achieved by retrospective voting
  for competing alternatives (Schumpeter 1942
  Popper 1982, Riker 1984).
• A contemporary economists formulation (Myerson
  1999)
• When elected leaders use political power for
  their own profit, we call it corruption or abuse
  of power, or in its most extreme form, tyranny.
  One of the basic motivations of democracy is the
  hope that electoral competition should reduce
  such political abuse of power, below what would
  occur under an authoritarian system, just as
  market competition reduces oligopolistic profits
  below monopoly levels
• What we want is
• An explicit model of electoral agency a) based
  on parsimonious motivational assumptions, b)
  allowing for moral hazard on behalf of elected
  representatives, and c) allowing voters to select
  representatives through elections (Austen-Smith
  Banks 1989, Banks Sundaram 1993, Besley 2006)
• Data on how real voters behave in situations
  resembling such a model.

3
The electoral agency model (1)

• Two periods t(1,2) with an election in between.
• Incumbents are of two possible types i ?H,D.
• Let ??s,1 signify a productivity parameter with
  0ltsltdlt1, where d is a discount factor.
• At the beginning of t1 random draws determine
  the type of the first period incumbent and the
  nature of the productivity shock, with commonly
  known probabilities Pr(iH)p and Pr(?s)q.
• Type and productivity is private information to
  the incumbent.
• Let q gt ½ by assumption.
• The t1 incumbent selects public production for
  t1, and production is publicly announced.
• There is an election at the end of t1.
• If the challenger wins, challenger type is
  randomly determined at the beginning of t2, with
  commonly known probability Pr(iH)p.
• If the challenger looses, the t1 incumbent
  continues as incumbent in t2.
• The t2 incumbent selects public production for
  t2. Production is publicly announced, payoffs
  are distributed and the game ends.

4
The electoral agency model (2)

• Let payoff to an incumbent of type D be
  VDr1dr2, where rt is rents (diversion of public
  funds for private ends) in period t
• Let the type H incumbent set rt0 for t(1,2)
• Let voter payoff be Ut(1-t)yaxt, where t is the
  tax rate, y is income before tax, xt is public
  production in period t, and agt1
• Let the budget restriction of the incumbent be
  ?(ty-rt)xt.
• Finally, let maximal rents equal rtty.
• It follows readily that a D-type incumbent always
  extracts maximal rents in t2.
• Note also that r1ty dominates r10 for a D-type
  incumbent

5
A pooling equilibrium

• It may still be in the interest of a D-type
  incumbent to mimic a H-type incumbent (by setting
  0ltr1ltty) in order to be reelected (and take
  r2ty)
• The central question becomes what is the
  rationally updated voter belief after observing
  production x1sty?
• Let 0lt?lt1 be the probability that a D-type
  incumbent facing ?1 mimics a H-type incumbent
  facing ?s
• Bayes rule now provides an answer

Now, assuming voters use pure cutoff strategies
in which they reelect only if Pgtp, we appreciate
that as long as q gt ½ all incumbents that
produces either x1sty or x1ty will be reelected
unanimously while no incumbent that produces x10
will get any votes. As long as dgts mimicking is
profitable for a D-type incumbent.
6
Experimental design (1)

• Endowment 100 Schillings per period (1
  Schilling 0.25 NOK in the money sessions)
• Tax rate 50 per period
• 1 Schilling in public production is worth 1.1
  Schilling
• Productivity parameter ??½,1
• Politician is programmed as follows
• If H-type and ?1 50 Schillings worth of
  production in both periods
• If H-type and ?½ 25 Schillings worth of
  production in both periods
• If D-type and ?1 25 Schillings worth of
  production in t1 and 0 in t2
• If D-type and ?½ 0 Schillings worth of
  production in both periods
• In all treatments p0.20
• TREATMENT q0.55 or q0.85
• This produces a marginal update to P0.23
  (??0.03) and a substantial update to P0.59
  (??0.39) respectively
• We have done 4 sessions with electorates of n3 /
  7 games
• Two of these where in the pilot (lottery-payoff)
  and two where money sessions
• We will do sessions with n1 / 20 games (and
  possibly also with n5 / 4 games)
• If things get interesting we may seek to fund
  cross-cultural studies

7
Experimental design (2)

• Two desiderata in the design is a) to root out
  social preferences, and b) produce statistically
  independent observations
• The politician is an automaton (a preprogrammed
  machine). There is no sense in punishing a
  machine.
• Every electorate is unique (due to an absolute
  stranger design). There is no sense in trying to
  punish or reward other subjects for previous
  play, since this can not possibly have any
  disciplining effects that the subject benefits
  from (he or she does not meet the punished or
  rewarded subject again).
• Majority decision ensures that all subjects in
  the same electorate earn the same amount in a
  specific game. No subject therefore is ahead or
  behind any other subject in a specific game.
  Thus, social preferences based on inequality
  aversion can have no effect.
• Due to 2. we may also be confident that
  observations are statistically independent

8
(No Transcript)
9
(No Transcript)
10
The pilot Right direction

• Master of Science students (PE 1st and 2nd year).
  First session absolute stranger second session 5
  subjects reused. Payoff Points provide winning
  chances in a lottery for two bottles of wine
  (utility will then be linear in points, cf. Roth
  Malouf 1979)
• Main result

Substantial update
Marginal update

• Difference with respect to production 77.5 is
  significantly different from zero in a Wilcoxon
  test (Mann Whitney U test) (p0.0001 z-value
  3.945 with 2 df).

11
The two money sessions

• Questions of interest
• Do we find a difference between treatments that
  goes in the right direction?
• How much of this difference remains if we check
  for adaptive / fictitious play updating?
• Does aggregation by majority rule lead to more
  correct decisions (rational expectations /
  Condorcet jury theorem)
• Can we describe the heterogeneity of the subjects
  with respect to updating rules?

12
Money sessions Right direction

• BA students (various years and majors). Both
  sessions absolute stranger. Payoff Schillings 1
  Schilling 0.25 NOK
• Main result

• Difference with respect to production 77.5 is
  barely significantly different from zero in one
  sided Wilcoxon test (Mann Whitney U test)
  (p0.12 z-value 1.168 with 2 df).

13
Logistical regression (periodgt0)
Combination of rational and adaptive belief
formationFrequency honest observed honest
politicians in past play / (observed honest
observed dishonest in past play)
14
Probability curves
15
Individual- versus majority decisions

• Does majority rule aggregate to behavior more in
  accordance with the model?
• Main result Maybe not

• Difference with respect to production 77.5 is
  barely significantly different from zero in one
  sided Wilcoxon test (Mann Whitney U test)
  (p0.11 z-value 1.204 with 2 df).

16
All sessions
17
Heterogeneity?
Roughly 1/3 is within an absolute deviation of 8
percentage points. We call them roughly
Bayesians. Roughly 1/5 is off mark by more
than the prior probability of drawing a good
politician (20 percent).
18
Consistency of beliefs and actions
6/10 subjects have one or less mistakes over the
nine rounds.No subject has more than 4/9
mistakes
19
Summing up

• Providing a more powerful update assures behavior
  more in line with model
• There are traces of adaptive expectations in the
  subject sample
• Controlling for these increases the effect of the
  update
• No clear results with respect to the aggregation
  properties of majority rule a consequence of the
  distribution of non-Bayesians in the sample?
• Heterogeneity Few Bayesians / few imbeciles.
  High level of consistency between actions and
  given believes.