Self-Government and Public Goods: An Experiment

Transcription

1 Self-Government and Public Goods: An Experiment Kenju Kamei and Louis Putterman Brown University Jean-Robert Tyran* University of Copenhagen * No blame for this draft.

2 Centralized vs. Decentralized Sanctions and Rewards We know punishment of free riding is sometimes an effective way to generate more sustained collective action (contributions to a public good) We ve seen that rewards are usually less effective than punishments, although the combination of the two* can be more effective than punishment alone. In the lab, we ve studied the problem mainly with decentralized punishment or rewards i.e., it s up to individuals to decide how much costly punishing to do and who to punish. * Page, Putterman & Garcia 2008, Getting Punishment Right: Do Costly Monitoring or Redistributive Punishment Help? Also: Sausgruber-Tyran, 2007, Economics Letters paper. 2

3 Alternatively, punishment (or punishment and rewards) could be implemented in a centralized fashion as, for example, in the 2000 AER paper by Falkinger, Fehr, Gächter and Winter- Ebmer. By centralized scheme, I mean one in which the punishments or taxes (and/or rewards or transfers) are automatically implemented, not a matter of individual choice. (An intermediate possibility is centralized choice of rules, e.g. who can be punished, as in Ertan, Page & Putterman, Who to Punish?, or of other rules about punishing, but decentralized decisions on costly punishment.) The present paper is about endogenous choice of a centralized scheme that is automatic once selected. 3

4 In the scheme of Falkinger, those who contribute less than the group average pay a penalty that increases in the size of their contribution shortfall, and those who contribute more than the group average receive a bonus that increases in the size of their overshooting of the average. The scheme has a balanced budget. Its presence leads to almost 100% efficiency in the linear public good case. (It also works well in their non-linear example with interior equilibrium.) 4

5 The objective of our research is to consider schemes like Falkinger s from the standpoint of self-government (endogenous choice). We ask: how efficiently will groups construct schemes of this sort on their own through majority voting. Rather than asking groups to vote up or down on the Falkinger scheme, we give them the more difficult task of devising a scheme from crude raw materials: possible fines, called taxes, for contributing too little (or if the subjects so choose, too much) to the public good. 5

6 Subjects vote on (a) whether to penalize those who contribute too much or to penalize those who contribute too little; (b) what level of contribution is exempt from taxation; and (c) the tax rate (how much per dollar above or below the exemption level). Unlike the (in some ways richer) experiments of Ostrom, Walker and Gardiner, we allow no communication among the subjects. 6

7 We want to pay attention to both - potential difficulties of understanding what choices would be efficient (leading us to include a cognitive ability test) - potential differences in voting due to differences of preference type (leading us to conduct a test of contribution strategy as an indicator of type) 7

8 The standard theory prediction for a public goods game with binding voted rules is very different from that for a public goods game with voluntary costly punishment. It s rational to select mandatory maximum contributions (linear PG case) out of selfinterest, whereas with universal selfinterest there will be no punishing and no contributions in the decentralized punishment case. 8

9 Question: might differences in preference types that are manifested in different punishing and different contributing behaviors in the (decentralized) VCM-withpunishment affect voting over centralized rules? Example: those inclined to free ride and punish high contributors in the VCM might be less likely to vote for efficient schemes in the centralized case. 9

10 Design Core of experiment: subjects play linear public goods game in partner groups of 5 for 6 sets of 4 periods (24 periods total). In treatments with possible taxes, subjects vote on the three parameters at the start of each set of 4 periods except the first set, which is a baseline PGG (VCM) in all treatments. We use the Fischbacher-Gächter-Fehr strategy method procedure to elicit pref. types at the beginning and an IQ test to elicit 10 cognitive skill at the end of the session.

11 Schematic design: Baseline treatment Strategy game schedule completed and instructions IQ test and questionnaire about political preference 4 periods with no rule 4 periods with no rule 4 periods with no rule 4 periods with no rule 4 periods with no rule 4 periods with no rule Periods 11

12 Schematic design: Tax treatments Strategy game schedule completed and 1 st instructions 2 nd instructions and 1 st Vote IQ test and questionnaire about political preference 2 nd Vote 3 rd Vote 4 th Vote 5 th Vote 4 periods with no rule 4 periods with chosen rule 4 periods with chosen rule 4 periods with chosen rule 4 periods with chosen rule 4 periods with chosen rule Periods 12

13 incentivizing strategy choices Contribution decision 1 of the main part of the experiment was also used to determine earnings of a second contribution round round 1½ Each individual s conditional contribution was applied to the average unconditional round 1 decision to determine her contribution and payout in round 1½ 13

14 2 treatments with possible voted taxes Tax-only treatment: any taxes collected are burned Tax-with-redistribution treatment: any taxes collected are equally divided among those not required to pay a tax 14

15 the public goods game n = 5, endowment = 20 Period earnings of subject i equal ( 20 G ) G t (2) i all i in a group i i 15

16 tax system choice System is defined by 3 parameters: (a) which kind of contribution is taxed; (b) what level of withholding or contributing is exempt from taxation; (c) what is the tax rate. Graphically, the 3 parameters are determined by 3 characteristics of a linear tax curve: (a) whether it is upward or downward sloping, (b) its y intercept, a, and (c) its x intercept, b. 16

17 (a) Private account case (b) Public account case 17

18 Every four periods, each group determines anew the 3 parameters, by voting: a) First, to tax contributions to the private accounts (for an downward sloping tax line) or to tax contributions to the public account (for an upward sloping tax line); b) Second, on the value of the y-intercept a, equaling the maximum possible tax, constrained to be 20, and an integer. c) Last, on the value of the x-intercept b, or tax exemption, 0 b 20, and an integer. 18

19 No abstentions were allowed. Which contributions were taxed (public, or private) was determined by simple majority. The values of a and b were set equal to the median values selected in the group. Subjects learned what was taxed before stating their choices of a, then learned the group a before stating their choices of b. Subjects in each group then interacted for 4 periods under the selected values. 19

20 Subjects had no information about how individuals voted, how many voted for a particular value, etc. Subjects had no information about what was happening in any group other than their own. There were 3 or 4 groups in a session, and subjects never learned which others present were in their own group. 20

21 Predictions I: Neoclassical Tax-only Treatment If all subjects care only about maximizing their earnings and believe that others are the same, all prefer universal full contributions (earning =30) to universal zero contributions (earning 20), and the only way to achieve this is to tax contributions to the private account and set b = 20 and a

22 Tax-with-Redistribution Treatment Although the Falkinger scheme s property of rewarding high contributors also increases contributing, the simple redistribution scheme we make available (dividing any tax revenue equally among the non-payers) treats all contributors of more than b equally, thus providing no incentive to contribute more than b. This means that for uniformly self-interested agents, b = 20 and a 14 is optimal in the Tax-with-Redistribution treatment too. 22

23 There is no individual incentive to deviate from the socially optimal values in voting, assuming uniform self-interest. 23

24 A Slight Variant: Heterogenous Preferences for Cooperation; Gaining from Redistributed Taxes If there s no tax, the individual sacrifices 0.7 for every 1 token s/he contribute to the group account. Suppose that some individuals value cooperating to some degree. - If you receive a subjective reward of 0.5 for each token you place in the group account, you ll put your whole endowment in if the tax rate on tokens put in your private account is >

25 Slight Variant, cont. Suppose 3 group members value each token contributed at 0.5, and 2 group members place no value on contributing. The 2 won t contribute unless the tax rate exceeds 0.7, but the 3 will contribute as long as the tax rate > 0.2. The 3 may vote for a tax just below 0.7, contribute their whole endowments, and share the maximum tax collected from the other 2. 25

26 Slight variant, concluded The possibility of such an equilibrium being more beneficial to the 3 non-taxpayers than is an equilibrium of full contributions may lead to inefficiency due to incomplete information i.e., subjects have difficulty inferring the actual distribution of preferences, hence miscalculations are likely. 26

27 Major Variant: Voting by Type Possibly, conditional cooperators may be more inclined than are free riders to promote cooperation by a high tax and high tax exemption level. Free riders may vote to tax contributing to the public account, not the private account! Free riders may likewise vote for a = 0 and/or b = 0, if the private account is chosen. 27

28 or Possibly, free riders may be the ones most likely to grasp the benefits of forcing contributions to the public good by taxes, since they recognize the rationality of free riding in the taxes absence better than others do (or since being less cooperative types makes them less reticent about imposing punishments). 28

29 Role of Cognitive Ability? Possibly smarter subjects are quicker to appreciate the benefit of taxing contributions to the private accounts and of a high tax and high exemption level. Groups with less smart subjects may make more mistakes in choosing their tax regime. 29

30 Results (preliminary analysis) 30

31 Sessions, Groups & Subjects 31

32 Results: I Types by Strategy Method 66% Cond. Cooperators, 16% Hump-Shaped, 10% Free Riders, 8% other. 32

33 33

34 34

35 Result of the IQ-test: cumulative percent of correct answers Average was 6.2 correct answers 3 correct answers or below is consider bad 9 answers or above is considered good. Anything in between 4 and 8 is normal 35

36 Contributions over time, Baseline treatment Contribution Period 36

37 Contributions over time, all treatments 37

38 Contributions begin around the typical 50% of endowment and have the usual declining trend over the first four periods, played without taxes in all treatments. Contributions are significantly higher in the Tax-only and Tax-with-redistribution treatments than in the Baseline treatment, suggesting that subjects were able to come up with rules that effectively increased contributions. 38

39 Earnings are also significantly higher in the two Tax treatments than in the Baseline treatment, suggesting that the rules subjects chose also tended to increase efficiency. 39

40 Earnings over time, all treatments Earning 30 baseline tax-only tax-with-redistribution Period 40

41 Contributions and earnings are a little lower in the Tax-with-redistribution than in the Tax-only treatment, suggesting that subjects found it easier to select good rules in the former. 41

42 Voting outcomes In each treatment, there was only one instance in which a group chose to tax contributions to the public account. This occurred in a last vote of the Tax-only treatment and in a first vote of the Taxwith-redistribution treatment. The proportion of voting outcomes of this kind is 1-in-35 (Tax-only) and 1-in-30 (Taxwith redistribution). 42

43 Since there s little variation between the two treatments with regard to choosing the right account to tax, the differences accounting for different contribution levels must be in the choices of the a and b parameters. As expected, b = 19 or 20 and a > 15 substantially more often in the Tax-only treatment than in the Tax-withredistribution treatment. This condition holds for 23 of 35 vote outcomes in Taxonly versus only 4 of 30 vote outcomes in Tax-with-redistribution. 43

44 Groups select more optimal values of a and b in later than in earlier votes, especially in the Tax-only treatment. In 6 of the 7 Tax-only groups, optimal values were chosen in each of the last 3 (out of 5) votes. This is true for only 1 of the 6 Tax-withredistribution groups. 44

45 (i) Tax-only treatment Table : Results of voting choice (account to be taxed, a, b and a/b) (ii) Tax-with-redistribution treatment 45

46 We are exploring strategies to understand why some individuals and groups selected more efficient parameters than others. On average, there are no systematic differences in the distributions of types and IQ between the two voting treatments, so the poorer choices in the Tax-withredistribution treatment must be due to either (a) the extra difficulty of understanding the choices, or (b) attempts to profit from the collected taxes (as in the differentiated preferences story). 46

47 regarding (b) Were more taxes collected in the Taxwith-Redistribution Treatment, reflecting a successful strategy of gaining from the redistributed taxes?

48 Answer: no; there s no evidence of this. 48

49 But this doesn t rule out the possibility that some subjects were attempting to do this and thereby decreasing overall efficiency.

50 Individual Level Analysis We look at 1. Contribution decisions 2. Voting on which account to tax 3. Voting on a and b

51 type helps predict contributions in periods without taxes (pd. 1 4) 51

52 contributions responded rationally to tax rates 52

53 voting on what to tax between 74% and 86% of individual votes were to tax private account contributions, in both treatments Figure 4: The proportion of individuals voting for the private account to be taxed 53

54 there s a trend of learning to tax the right account in the Tax-w/-Redis. but not in the Tax-only Treatment Figure 4: The proportion of individuals voting for the private account to be taxed 54

55 In the Tax-with-redistribution treatment, there s an indication that conditional cooperators were less likely to vote for taxing contributions to the public account than were free riders ( consistent with the voting by type hypothesis). We define a type index = 1 for conditional cooperators, -1 for free riders, and 0 for hump-shaped and other contributors. 55

56 In probit, random effects probit, and logit regressions to explain voting to tax contributions to the private account, the Type Index has a significant positive coefficient for the Tax-with-redistribution Treatment data, but the coefficient for the Tax-only Treatment data isn t significant. (perhaps there s a stronger type effect on voting due to trying to gain from redistributed taxes?) 56

57 57

58 individual voting on a and b still at exploratory stage IQ has a positive effect on preferred values, sometimes significant, in some specifications being a conditional cooperator (or having higher type index) often has a positive significant effect on preferred values, sometimes significant 58

59 Tobit regressions, initial individual votes 59

60 same, using type index (tobit regressions, initial individual votes) 60

61 Tobit regressions, individual votes, 2 nd through 5 th votes, Tax-w/-Redis. 61

62 same, using types index (Tobit regressions, individual votes, 2 nd through 5 th votes) 62

63 neither specification is significant for votes in the Tax-only Treatment 63

64 (types index version) 64

65 very preliminary conclusion The regressions support the idea of a mild positive impact of intelligence on choosing a good tax scheme, and a types effect whereby the more cooperatively-inclined individuals also tend to vote for better tax schemes. The latter contradicts the idea that more cooperative types reduced efficiency by trying to milk others for tax revenue that would be distributed among them.

66 Main Conclusions Subjects are generally able to choose tax schemes that lead to considerably better outcomes than VCMs without taxes. Making the scheme more complicated seems to reduce the efficiency of tax choices, which leaves it unclear whether subjects would come up with something like the Falkinger scheme on their own.