The disciplining role of repeated elections: some experimental evidence

Abstract

This paper reports on an experiment that studies the functioning of a political market. The formal model, on which the experiment is based, considers a situation wherein two candidates participate in a series of structurally identical elections. The model also incorporates and emphasizes an important informational asymmetry between voters and candidates: specifically, a candidate's diligence in office (i.e. the amount of effort expended) is unobserved by voters. How do subjects, voters and candidates, behave in the laboratory? My principal findings are twofold. First, given the institutional structure of repeated elections, voters have the ability to extract effort from candidates. However, candidates' effort level choices fall well short of the upper endpoint of the effort choice set. In other words, the electorate's ability to sanction candidates is not sufficient to eliminate the rents of office. Second, candidates' effort level choices are mostly consistent with incentives implicit in the elections set-up.

Keywords:

• elections
• principal-agent
• moral hazard
• experiments

Acknowledgements

The author is grateful to Sanmitra Ghosh, Kenneth C. Williams, Rick Wilson, and seminar participants at the Jawaharlal Nehru University, New Delhi, for very helpful comments. Two anonymous referees suggested numerous changes that vastly improved the paper's structure and quality. Of course, the usual disclaimer applies.

Notes

1. Consider the measurement of legislator ideology in Kau & Rubin (1979) and Kalt & Zupan (1984). Summary indices of legislators' voting records are first regressed on variables that proxy constituent interests. This done, the regression residuals are interpreted as legislators' ideology. Peltzman (1984) correctly points out that the regression residuals may simply be picking up omitted constituency variables. As yet another example of an awkward measure, consider the case of senator shirking in Figlio (2000). The measure of a senator's ideological shirking is taken to be the absolute value of the difference between the senator's ADA score (a special interest index of congressional voting) and the average ADA score of members in the state's House contingent. Figlio (2000, p. 273) justifies this approach by invoking the following thoroughly ad hoc and untested assumption: ‘Even if individual House members are shirking, it is likely that for every House member voting to the left of his constituency, there is another who is voting to the right of hers.'

2. Consider the well-known and innovative paper by Lott (1987). To show that retiring congressmen do not alter their voting behavior relative to non-retiring congressmen, Lott (1987) regresses the change in a congressman's average NTU score (a special interest index of congressional voting) between the 95th and 94th Congresses on a few controls and a dummy variable that equals 1 if the congressman retires in 1978 and is 0 otherwise. However, the retirement decision is not exogenously assigned to politicians. Therefore, there remains a concern that omitted variables correlated with the decision to retire also affect changes in the NTU scores.

3. Agency theory, for example, suggests that legislator shirking diminishes when the value of being in office increases. This intuitive result has, to the best of my knowledge, escaped empirical scrutiny.

4. Since my experimental findings are in accord with received wisdom, have I simply confirmed the obvious? I shall argue that the answer is ‘no.’ Recall that the model is beset with a multiplicity of equilibria. In fact, for all parameter values, it is a symmetric and stationary subgame perfect equilibrium for both candidates to expend zero effort in office (refer to Proposition 1 in Section 3). Thus, the relationships identified by the experiment (e.g. average candidate effort level is increasing in the private benefit of office) do not directly follow from equilibrium considerations alone. Instead, the experimental environment produces behavior that happens to confirm long-standing intuitions.

5. Since E ≡ [0, a] and a <1, the period-t incumbent candidate is not permitted to set e_t equal to one. Why is the restriction ‘a less than one’ imposed? If the set of feasible period-t effort levels is allowed to be [0,1], a standard assumption of the moral hazard framework is violated: the support of the distribution of y_t now changes with e_t (if π _H  = 1 and π _L <π _H , the support of the distribution of y_t is {y_L , y_H } when e_t ∊ (0, 1) and {y_H } when e_t = 1). It turns out that the theoretical properties of the model with π _H  = 1 and E ≡ [0, a], a <1 are markedly different from the case wherein π _H  = 1 and the period-t effort level is chosen from the set [0, 1].

6. In the experiment, one of the treatments used the following parameter values: B = 7425, k = 1, δ = 0.9, π _L  = 0, and π _H  = 1. Now pick any [ebar] in the interval [0, M (7425, 1, 0.9, 0, 1)]. Proposition 1 guarantees the existence of a pair ([rbar]_L, [rbar]_H ) such that ([ebar], [rbar]_L, [rbar]_H ) constitutes a symmetric and stationary subgame perfect equilibrium of the model.

7. For concreteness, suppose candidate A is in office in periods one and two of a trial. Suppose also that the period-one effort level chosen by candidate A results in a cost of c ₁ Franks. Then, c ₁ Franks is required to be weakly less than 5×B Franks. Observe also that candidate A's cumulative earnings at the time that she chooses the period-two effort level is 5×B Franks [the initial endowment] + (B – c ₁) Franks [the earnings in period one]. Candidate A in period two is required to choose an effort level e ₂ whose cost does not exceed 6×B – c ₁ Franks. The parameters in the trial were so chosen that, regardless of the history up to period t, the cumulative earnings of the period-t incumbent candidate were sufficient to make all choices of e_t ∊ [0, 95] (see Step 3) budget feasible. In other words, bankruptcy considerations did not constrain the effort level choices of the period-t incumbent candidate.

8. The instructions read out to subjects avoided use of the word ‘effort’. Instead, e_t was referred to as the policy adopted by the period-t incumbent candidate.

9. In the election model of Section 2, candidates' effort choice set is the interval [0, a], a <1. To allow for greater variation in observed behavior, the experimental set-up expands the effort choice set to the interval [0, 95]. The chosen effort level e_t ∊ [0, 95] is first mapped into the interval [0, 0.95] by the transformation e_t /100. This done, assumptions [A.1]–[A.3] are used to determine the probability of generating y_H and y_L .

10. At the conclusion of a session, subjects' cumulative earnings in Franks were converted into dollars. For candidate subjects, 7500 Franks exchanged for a dollar while, for voters, 40 Franks exchanged for a dollar. Each subject was aware of her own exchange rate before the experiment started.

11. If l(t) equals zero (i.e. candidate –i(t) was never in power between periods 1 and (t - 1)), I set [ebar]^E (t) equal to 50.

12. For concreteness, assume that the history at period t is such that y_L is observed in only periods 1 and 3. Then, T_L = {1, 3} and T_H = {2, 4, 5, … , t – 1}.

13. If in period t, there is no previous realization of y_L (i.e. T_L = ⊘), I arbitrarily set (t) = 0. If in period t, there is no previous realization of y_H (i.e. T_H = ⊘), I arbitrarily set (t) = 1. The model's predictions are insensitive to these initial value choices.

14. To obtain e^C (t), equations (2) –(7) need to be solved simultaneously. Since closed form solutions are unavailable, numerical techniques were employed. I set the value of δ to be 0.90; this is not strictly valid since the trial termination procedure (refer to Step 6 in Section 4.1) operates only after 25 election periods have elapsed.