Besley-Incumbent Behavior (1995) PDF

Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition Author(s): Timothy Besley and Anne Case Source: The American Economic Review , Mar., 1995, Vol. 85, No. 1 (Mar., 1995), pp. 25- 45 Published by: American Economic Association Stable URL: https://www.jstor.org/stable/2117994 J...

Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition Author(s): Timothy Besley and Anne Case Source: The American Economic Review , Mar., 1995, Vol. 85, No. 1 (Mar., 1995), pp. 25- 45 Published by: American Economic Association Stable URL: https://www.jstor.org/stable/2117994 JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms is collaborating with JSTOR to digitize, preserve and extend access to The American Economic Review This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms  Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition By TIMOTHY BESLEY AND ANNE CASE * This paper develops a model of the political economy of tax-setting in a multijurisdictional world, where voters' choices and incumbent behavior are determined simultaneously. Voters are assumed to make comparisons between jurisdictions to overcome political agency problems. This forces incumbents into a (yardstick) competition in which they care about what other incumbents are doing. We provide a theoretical framework and empirical evidence using U.S. state data from 1960 to 1988. The results are encouraging to the view that vote-seeking and tax-setting are tied together through the nexus of yardstick competition. (JEL D72, H20, H71) The electoral cost of raising taxes is a taxing behavior before changing taxes at stock political anecdote. However, while folk home. This would give rise to a kind of wisdom suggests that incumbents raise taxes (yardstick) competition between jurisdic- at their peril, proper treatment of the issue tions, each caring about what the other is recognizes that voters' choices and incum- doing. This paper builds a model of such tax bent behavior are determined simultane- competition, where voters choose whether ously, and that the political consequences of or not to reelect officials based on their a tax increase may vary by circumstance. If performance while in office, using neighbor- voters are skeptical about the need for addi- ing jurisdictions to evaluate the perfor- tional taxes, even a small increase may force mance of their incumbents. We provide a the governor to look elsewhere for work. theoretical framework and an empirical However, if taxes are rising everywhere, vot- analysis that uses data from U.S. states from ers may be convinced that a tax increase is 1960 to 1988. necessary. In this case, even a large increase Our starting point is a world with asym- may be politically acceptable. In a world in metric information between voters and which voters make comparisons between politicians; the latter are assumed to know states, incumbents may look to other states' more about the cost of providing public services than the former. Politicians also differ in their type. Good ones do no rent- seeking, whereas bad ones finance their * Woodrow Wilson School, Princeton University, whims at taxpayers' expense. The problem Bendheim Hall, Princeton, NJ 08544-1022. We are for voters is to distinguish between the grateful to our anonymous referees, and to Kristin Butcher, David Card, Steve Coate, Dan Feenberg, two. Consonant with the large literature on Kai-Uwe Kuhn, John Londregan, Gib Metcalf, Jim multiagent incentive schemes (see e.g., Poterba, Tom Romer, Michael Rothschild, Dan Sasaki, Bengt R. Holmstrom, 1982), we show that it Richard Zeckhauser, seminar participants at Chicago, makes sense for voters to appraise their Columbia, the Federal Reserve Bank of Philadelphia, the National Bureau of Economic Research, the Uni- incumbent's relative performance, if neigh- versity of Pennsylvania, SUNY-Albany, Virginia, and boring states face correlated shocks. VPI for helpful comments, discussions, and encourage- A theoretical model of this kind predicts ment. We also thank Gena Estes for excellent research that the reelection performance of one ju- assistance and the John M. Olin Program for the Study risdiction will depend both upon the juris- of Economic Organization and Public Policy at Prince- ton University for financial support. None of the above diction's own tax policy and upon that of its is responsible for the product. neighbors. In particular, if a state has high 25 This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 26 THE AMERICAN ECONOMIC REVIEW MARCH 1995 tax increases relative to its neighbors, citi- to be augmented by a political framework. zens interpret this as evidence that their This paper has spawned a heated discussion official is bad and unseat him at the next (see e.g., J. Vernon Henderson, 1985).3 election. Our empirical evidence is consis- Whatever the merits of these arguments, it tent with this view. seems reasonable to suggest that resource A second theoretical prediction is that flows can only be a long-run solution to tax-setting behavior is affected by electoral differences in the tax policies of states. In competition. In particular, states may trim the short run, the ballot box may serve an tax rate increases that put them out of line important function and even in the long run with their neighbors. Thus, we have a kind may be a less costly alternative than migra- of yardstick competition, studied previously tion.4 We shall therefore make this the fo- by Andrei Shleifer (1985) among others, in cus of our investigation here. Our results which agents use the performance of others support the view that electoral competition as a benchmark. This too is consistent with affects state tax-setting. our empirical results. That a governor's chance of reelection The importance of asymmetric informa- might in part depend on his track record on tion in local spending decisions has been taxes has long been noted in the political- recognized, inter alia, by David F. Bradford science literature. Thad L. Beyle (1983 p. et al. (1969) who argue that it is difficult for 215), for example, suggests that taxes were a voters to infer the level of services that will "key issue" in the defeat of 30 percent of be delivered for a given expenditure level, the governors who were not reelected in the making efficiency in provision difficult to 1960's and in the defeat of 20 percent of assess. Recent work in political economy, such governors in the 1970's. In a similar such as Jeffrey S. Banks and Rangarajan K. vein, Susan B. Hansen (1983 p. 177) cites Sundaram (1991), David Austen-Smith and evidence that tax issues began to "figure Banks (1989), and Kenneth Rogoff (1990), prominently in decisions to vote for or has also emphasized the importance of against a particular party or candidate" in asymmetric information and has studied the the mid 1970's in determining the outcome resulting political agency problem. We ex- of congressional and presidential races. tend this type of model by considering rela- Moreover, taxes were mentioned directly by tive performance evaluation in voting deci- 15 percent of those surveyed in 1980 as a sions. factor in their ballot choices (Hansen, 1983 The predominant analytical framework p. 177). The political-science literature has for tax competition is the Tiebout model,' also taken the idea of comparisons across which in its purest form argues that re- states seriously, beginning with the analysis source flows between jurisdictions obviate of Jack L. Walker (1969). the need for political competition.2 There The remainder of the paper is structured has however been much debate about the as follows. Section I introduces our data extent to which resource flows alone will and presents a preliminary look at the evi- work. For example, Dennis Epple and Allan dence. Section II presents a simple theoreti- Zelenitz (1981) have argued that, even in cal analysis that solidifies the ideas behind the long run, allowing individuals to sort the empirical work. Section III extends the into jurisdictions will not eliminate rent ex- traction by states and that the model needs 3Epple and Thomas Romer (1989) argue that the key assumptions concern the extent to which jurisdic- 'Other models include that of Ravi Kanbur and tion boundaries are flexible. Michael Keen (1993), which examines implications of 4The view that the time frame of the analysis is cross-border shopping, rather than individual reloca- important is borne out in Epple et al. (1988). In addi- tion. This is, of course, most appropriate for indirect tion, when key industrialists were surveyed by The New taxes. York Times (1991), taxes were cited as being only the 2See Daniel Rubinfeld (1987) for a survey of Tiebout 12th most important factor determining firm location models, empirical and theoretical. decisions. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 27 TABLE 1-REELECTION HISTORIES OF U.S. GOVERNORS 1960-1988 INCUMBENT OUTCOMES Defeated Did not run runot Reelected __________________ ~~~run; - R e l c d Number of Ran for reached Year elections Election Primary Retired Congress limit Reelected Percentage 1960 28 6 0 2 5 8 7 25 1961 2 0 0 0 0 2 0 0 1962 33 9 2 0 3 6 13 39 1963 2 0 0 0 0 2 0 0 1964 26 3 0 3 2 7 11 42 1965 2 0 0 0 0 1 1 50 1966 33 6 2 1 2 9 13 39 1967 2 0 0 0 0 2 0 0 1968 22 3 0 2 3 4 10 45 1969 2 0 0 0 0 2 0 0 1970 33 6 0 8 1 5 13 39 1971 2 0 0 0 0 2 0 0 1972 19 2 2 3 1 4 7 37 1973 2 0 1 0 0 1 0 0 1974 33 2 1 6 1 7 16 48 1975 3 0 0 0 0 1 2 67 1976 14 2 1 3 1 2 5 36 1977 2 0 0 0 0 1 1 50 1978 34 4 2 3 2 11 12 35 1979 3 0 0 0 0 3 0 0 1980 13 3 2 0 0 1 7 54 1981 2 0 0 0 0 2 0 0 1982 34 5 1 6 1 4 17 50 1983 3 1 0 0 0 2 0 0 1984 13 2 0 3 1 3 4 31 1985 2 0 0 0 0 1 1 50 1986 34 2 0 4 2 11 15 44 1987 3 0 0 1 0 2 0 0 1988 12 1 0 2 0 1 8 67 model to give a workable empirical specifi- governors who chose not to run for office cation. Section IV presents the results, and again. Section V discusses some extensions and Table 1 suggests that a nontrivial propor- alternative models. Section VI contains tion of governors eligible for reelection ei- some concluding remarks. ther chose not to run or were defeated at the polls.6 During this 30-year period, there I. Preliminary Data Analysis Our data are centered on the reelection bids of governors in the continental United 5Repeating our analysis excluding the "retired" States from 1960 through 1988. Table 1 group just results in an increase in the standard errors. 6In many states, governors face a term limit. That is, shows the reelection histories of governors by law they may be ineligible to succeed themselves in during this period. We will assume below office. Elections in which the term limit binds are not that eligible governors who did not run for included in our voting analysis. However, term limits reelection and who did not run instead for will be used in tax-setting analysis as a natural means another office chose to step down because of separating those governors who should care about neighbors' taxes (i.e., those eligible to run for reelec- they assumed they would lose or were pres- tion) from those who should not (lame-duck governors). sured to do so by dissatisfied party officials. For further analysis of gubernatorial term limits and The empirical analysis controls for age of policy-making, see Besley and Case (1993). This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 28 THE AMERICAN ECONOMIC REVIEW MARCH 1995 are only two years in which more than half were $4,700 in New York State in that year. of all incumbents are reelected. In a major- In large part these differences reflect diver- ity of the even-year elections, between 15 sity across states in the division of taxing percent and 40 percent of governors eligible authority between state and local levels of for reelection lost either in the primary or government. In addition, states differ in their in the general election. provision of public services, which is also Our analysis makes use of two tax data reflected in tax liabilities. It thus makes sets. The first contains data on the effective greater sense to focus on states' changes in income-tax liabilities of joint filers in each tax liabilities, rather than on states' levels. of the 48 continental states. These data, We also maintain that a model based on generated at the National Bureau of Eco- agency problems due to asymmetric infor- nomic Research (Cambridge, MA) using the mation about shocks to the cost of providing TAXSIM program, accurately capture the public services naturally gives way to a spec- income-tax liabilities that governors and leg- ification in which changes in taxes matter. islatures envisioned for taxpayers in differ- We are interested in the possibility that ent income categories. These liabilities are voters compare their own tax changes with quite appropriate for the analysis at hand: those in neighboring states before heading the effective-tax calculations control for the to the polls.8 Incumbents would then be effects of federal taxes and local property more likely to face defeat if they increased taxes paid when calculating the taxes owed taxes and less likely to lose, ceteris paribus, to state governments, and they reflect the if their neighbors increased taxes. If this will of the elected officials. However, be- were true, elected officials would be sensi- cause TAXSIM estimates are available only tive to their comparative performance on for the period 1977-1988, and since the taxation. Thus, we would expect to find two estimates are available only for income patterns in the raw data. First, electoral taxes, we make use of a second data series defeat would be positively correlated with a constructed from data published annually in tax increase in the incumbent's own state the Statistical Abstract of the United States. and negatively correlated with tax increases These tax data are real per capita income, in neighboring states. In addition, tax sales, and corporate taxes, collected by state, changes in neighboring states would tend to for the period 1960-1986. Jointly, these be positively correlated. taxes account for 90 percent of state tax Table 2 presents correlations between collections in 1980.7 While having the ad- states' effective income-tax changes (t - vantage of being more comprehensive in [t - 2]) and those of their geographic neigh- terms of state taxes covered, such tax data bors for the 10-year period 1979-1988, us- may be a less accurate reflection of elected ing the TAXSIM data. We define "neigh- officials' intentions, as taxes paid also reflect bors' tax change" as the average change in economic conditions within the state. As the tax liability or real tax revenues (depending reader will see, our results are robust to the on the data set) of geographically neighbor- choice of data set. Both data sets reveal tax liabilities that vary markedly between states in a given income category. For example, effective income-tax liabilities for $60,000 joint filers were $108 in Tennessee in 1980, while they 8We assume throughout that voters are most inter- ested in how they personally would fare in a neighbor- ing state. Thus $40,000 earners look at the taxes faced by $40,000 earners in nearby states and so on. Incum- bents may be more sensitive to a particular income group, if this group is either better informed or more apt to vote in the next election, or both. Our analysis 7Source: State Government Finances (1980), pub- allows voter and incumbent sensitivity to tax changes to lished by the Bureau of the Census. vary by income level. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 29 TABLE 2-CORRELATION BETWEEN CHANGES IN TAx LIABILITY AND THE UNSEATING OF INCUMBENTS, 1979-1988 (TAXSIM DATA) A. Correlation in Neighboring States' Tax Liability Changes (t - [t - 2]) Income groups $25,000 $40,000 $60,000 $100,000 Pearson product- moment correlations: 0.18 0.24 0.29 0.30 B. Correlation Between Changes in Effective Income-Tax Liability and Governor Defeat at the Polls Primary + general- General-election defeat election defeat Defeated or retireda Income groups Income groups Income groups Tax change-- - (t - [t -21) $25,000 $40,000 $100,000 $25,000 $40,000 $100,000 $25,000 $40,000 $100,000 Own 0.25 0.17 0.07 0.21 0.14 0.07 0.22 0.17 0.18 Neighbors' -0.12 - 0.09 -0.11 -0.10 -0.09 -0.11 -0.07 -0.05 -0.08 Number of observations: 66 66 66 69 69 69 85 85 85 a,"Retired" governors are those eligible for reelection wh ing states.9 Table 2 reveals that there is a second part of Table 2 reveals, changes in a significant amount of correlation between state's income-tax liability are positively and neighbors' tax changes and a given state's significantly correlated with unseating an in- tax changes, with the Pearson correlation cumbent governor, with a correlation coef- coefficient ranging from 0.18 for the $25,000 ficient of roughly 0.20. At the same time, income group to 0.30 for the $100,000 in- changes in neighbors' tax liabilities are neg- come group. For all groups, this correlation atively correlated with defeat of an incum- is significant. This could, of course, be ex- bent in a given state, with a correlation plained by a number of factors. Below, we coefficient of roughly -0.10. Thus while will control for year effects and for the neighbors' tax changes are positively corre- possibility that neighbors face common lated with a given state's tax change, they shocks. are negatively correlated with the defeat of Correlations between increases in effec- that state's incumbent. tive income-tax liabilities and incumbent de- feat are also present in the raw data. As the II. A Theoretical Example Our empirical specification, developed below, allows for an informational external- ity between neighboring jurisdictions, which affects both voting behavior and incentives for incumbents to increase taxes. There are 9We choose a geographical definition of neighborli- ness for two main reasons. First, geographic neighbors three premises behind this: are quite likely to experience similar shocks to their tax bases and, for this reason, provide information on the Premise 1: Agency problems due to asym- size of the innovation to neighboring states' voters. metric information are a feature of political Second, geographic neighbors capture as nearly as pos- competition. Specifically, incumbents know sible the idea that states belong to the same media market, having good information about what is going more about the short-term evolution of some on close by. key variables than do voters. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 30 THE AMERICAN ECONOMIC REVIEW MARCH 1995 Premise 2: Voting is the main incentive Each jurisdiction is run by elected offi- mechanism used to discipline incumbents. cials, who are potentially of two kinds: This is a special feature of a political analy- "good" or "bad." The former do no rent- sis. The literature on incentive schemes un- seeking, providing public services at cost, der asymmetric information in general al- while the latter engage in rent-seeking, lows the principal(s) wider-ranging incentive charging more than the cost of services mechanisms than just deciding whether or in taxes. Thus the bad official adds either not to reelect an incumbent.'0 A or 2A to the tax burden. Incumbents' (pure) strategies are denoted T(Oi, i) (i E Premise 3: Voters are able to appraise in- {L, M, H}; j E {G, B}, where G stands for cumbents' relative performance. From the good and B stands for bad).14 The behavior media or other sources, voters can gain ac- of the good incumbents is always T(OL, G) = cess to information about what other in- T17 i(Om, G) = T2 and T(OH, G) = T3. cumbents are doing, which serves as a The example has two time periods, and benchmark for their own jurisdiction. the voters' and politicians' discount factor is 8, which we suppose satisfies 1 > 8 > 2. The To fix ideas, it is useful to present a latter part of this guarantees the willingness simple example. While this is specific, the of an incumbent to give up A to be re- ideas are quite general." Consider a elected. A newly elected official is good with "jurisdiction" whose government provides probability y. Voters observe his tax-setting one unit of a public service of a given qual- decision and then choose whether or not to ity,'2 financed entirely by taxes. The cost of reelect him. We assume that voters care providing public services is initially Oi, which about minimizing their expected period-2 is stochastic and observed only by incum- taxes and use period-1 taxes to update their bents. The shock can take on one of three beliefs (using Bayes' rule) that the incum- values: low, medium, or high (denoted bent is good. We denote voters' strategies L,M,H), which are evenly spaced with dif- by A(Ti) E [0, 1] (i E {1, 2, 3, 4, 5}), which de- ference, A. The probabilities of the three notes the probability that they will reelect outcomes are: (qL, qM, qH). The incumbent an incumbent who sets a tax of Ti. If re- can charge rent on top of the cost of provi- elected, bad incumbents face no period-2 sion of either A or 2 A, giving five possible reelection discipline and hence set a tax tax levels, denoted by {i7, T2 73, 74, 75}- equal to Oi + 2A (i E {L, M, H}). Good incum- {OL, OM, OH OH + A ,OH + 2A}.13 bents provide services at cost in period 2. We find perfect Bayesian equilibria of the tax-setting game.15 First, nature selects an incumbent type and a cost shock. Bad in- cumbents then choose taxes to maximize '0There are other mechanisms of political disci- their discounted utility. Voters observe taxes pline, such as party structures. These may be important in affecting the behavior of incumbent officials (e.g., if and update their beliefs using Bayes' rule. a governor might like to be selected to run for Congress Their choice of whether or not to reelect or President in the future). However, these schemes the incumbent is based on minimizing ex- are not at the discretion of most voters. pected period-2 taxes. In equilibrium, voters " A more detailed presentation of some theory is available in our discussion paper (see Besley and Case, and incumbents have rational expectations. 1992). 12The assumption that quality is fixed is extreme, particularly so in an empirical context. More generally, one might imagine the government choosing tax/qual- ity pairs. The absence of any measure of quality in our example, by concerns about being prosecuted if found empirical work is apt to mean that we understate theout. It could represent the limits placed by migration costs. If taxes are too high then it will be worthwhile sensitivity of voting to taxes, since some tax increases may be reflecting increases in quality and should not for voters to leave. 14The Appendix exhibits some mixed-strategy equi- therefore result in taxpayer hostility. We explore alter- native measures of fiscal performance in Section V. libria. 13Thus we are assuming that the governor cannot 15Readers unfamiliar with this should see Drew claim unbounded rents. This could be justified, for Fudenberg and Jean Tirole (1991). This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 31 Different equilibria (with both pure and other's type17 and that qH ?1- y. What mixed strategies) are possible at different happens if voters have access to information parameter values. A full characterization is about taxes in both jurisdictions? To deter- given in Appendix A. Voters will always mine the implications of this, there are three believe that an incumbent who sets T4 cases to consider: or r5 is bad with probability 1, so that A(TO = A(7) = O. With 8 < 1, there(A) Both incumbents are good. This is also cannot be an equilibrium where r(OL,B) = straightforward: each sets taxes equal T1' i(OM, B)= T2, or T(OH, B) =T3, and to Oi, i E {L, M, H}, as we had above. hence A,ir1) = 1. The following proposition (B) Both incumbents are bad. In this case, illustrates the full strategies of voters and the equilibrium described in the Propo- incumbents in an interesting case. sition is a perfect Bayesian equilibrium for the two incumbents.18 Thus both PROPOSITION 1: If qH > 2, then the fol- incumbents decide to reduce their lowing constitute an equilibrium: rent-seeking when the cost shock is OM. (a) good incumbents set (OL, G) = 1 (C) One incumbent is good and the other is i-(OM, G) = 2' T(H(, G) =73; bad. In this case, the bad incumbent (b) bad incumbents set r(OL, B) = 3 knows that he will be found out by T(OM, B)=T73, T(OH, B)= 75; setting a tax above his neighbor's, and (c) voters set ,(Tl1)=1, (TO2)=1, (T(3)= he can no longer sustain the strategy 1, L(T4) = 0, (T=5) =O. described above: playing T3 when the shock is 0M will result in his being The bad incumbent takes a reduction in unseated. Thus we would now have rent when the cost is OM in order to be T(OM' B) =T4 for a bad incumbent. Thus reelected. He is willing to do so since 8 > 12 period-1 taxes are higher under yard- Note that both voters and incumbents are stick competition. However, since bad behaving rationally. Voters find it worth- incumbents are "found out" in this case, while to reelect any incumbent who sets 73, the voters have lower expected period-2 since the probability that the incumbent is taxes.19 The good incumbent inflicts an good given a choice of T3 iS YqH/{yqH + ( -y)(qL+ qM)} 2 y if qH ?24.16 Intuitively, a high enough value of qH is needed for it to be sufficiently 17This is a bit too strong. While it is probably likely that an incumbent who chooses T3 iS reasonable to suppose that neighboring incumbents good, so that voters are willing to reelect know more about each other than voters do, full infor- mation may be an exaggeration. incumbents if they see T3. 1 The voter in one jurisdiction now decides to re- To get an informational externality, imag- elect the incumbent if he sees T3 in both jurisdictions if ine now that there are two jurisdictions with y2qH /[y2qH +(1 - Y)2(qL + qM)] > y, which reduces identical environments and costs shocks, but to qH> 1- y. This condition differs from that in Proposition 1 because voters allow for the possibility which may elect officials of different types. that both incumbents are good or bad in their updating We suppose also that incumbents know each after seeing T3 in both jurisdictions. The condition for this pure-strategy equilibrium is weaker if y > 2; seeing both incumbents choosing the same strategy gives the voter more confidence that both are good when the 16In a multiperiod model, with a sequence of two- probability that any given incumbent is good is high period terms, voters have an extra incentive to switch enough. to a fresh incumbent who, if bad, has better period-1 19Yardstick competition does not necessarily domi- incentives to curtail rent-seeking. This leads to a stiffer nate ex ante in terms of expected taxes. It does so if hurdle being faced by an incumbent; the critical value y > 1/27. This makes sense intuitively; the value of of qH needed for the strategies described in Proposi- yardstick competition is in identifying bad incumbents, tion 1 to be an equilibrium is concomitantly higher. while the cost is in terms of higher current taxes. If y Notes on this extension of the model are available from is high enough, then the selection advantage domi- the authors upon request. nates. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 32 THE AMERICAN ECONOMIC REVIEW MARCH 1995 externality on the bad one, reducing results are obtained with differences be- the latter's reelection chances. tween t and t -3.) A representative voter in state i on This illustrates two key features of a the eve of an incumbent's reelection will yardstick-competition mo-del. First, it pro- desire to reelect the incumbent if the ex- vides a rationale for tax-setting to affect pected value of future tax increases under incumbent reelection chances. Second, it the incumbent is less than that with the suggests that incumbents' tax-setting behav- challenger; that is, At1(ATit; AT-it) < ior may be affected by voters looking at AT'(A-Tt; AT_jt), where At1(ATt; AT-it) is neighboring jurisdictions. It is these ideas the expected value of future tax changes that we exploit in our empirical analysis. with the incumbent and AT1(ATit; AT-it) is that expected of the challenger.21 The III. Empirical Specification probability-of-reelection function is esti- mated in a random utilities framework; a The essence of our approach can be cap- shock to preferences denoted by Ei affects tured in a two-equation empirical model. the election outcome.22 The shock is as- The first equation examines the determi- sumed to be normally distributed with mean nants of gubernatorial reelection, and the zero and standard deviation a. second examines the determinants of tax- For simplicity, we take a linear approxi- setting. The interdependence between these mation to the gain from reelecting the in- two is captured by cross-equation restric- cumbent: tions. Our empirical specification will use Qil( ATit, A T-it) ATi( rit; 57-it) changes in taxes as the main tax-setting de- cision. Such changes are most likely to rep- -\ Ati( Tit; AT-it).- resent responses to shocks about which there is asymmetric information.20 Innovations to Thus the probability of reelection is costs can be thought of as fiscal crises due, for example, to increased medicaid ex- penses, increased infrastructure expenses, (1) Pr(fI(ATit; AT-it) > - Eil or recession-driven revenue shortfalls. It is after such events that citizens must deter- =-(((Pxit + y,+iAr ATit + Y2+ r -it)lo) mine whether the change in taxes is "ap- propriate." We use ArTi to denote the = Ri(T it; AT-it) change in taxes in state i and Ar-it to denote the change in state i's neighbors, where F(.) is the cumulative distribution both at time t. Since an incoming governor function of the standard normal distribution may take more than a year to implement his tax program, we use changes in taxes paid in year t relative to year t -2. (Similar 21We expect the voter to care about the whole future sequence of tax increases rather than just the next term's for the reason discussed in footnote 16. 22We interpret this as follows. Voters are heteroge- neous, since they care differentially about higher taxes, are differentially informed, and may have different 20This is strictly speaking inaccurate. Last period's priors about the incumbent's type. Assume that the taxes also reflect whether the incumbent was good or probability that a particular voter votes depends upon bad. In a more general model, all past tax rates would how strongly he or she feels, and that this is monotonic provide information about an incumbent's type, and in Ar and Ar-i, in the way that our model suggests, incumbents would strategically manipulate the se- for all voters. However, shocks on election day, such as quence of tax rates to optimally influence voters' be- inclement weather, make some types of voters more or liefs. Incorporating these elements into a structural less likely to turn out. The actual election outcome is model would require a considerably more complicated therefore random even though it is affected by tax- analysis, which we leave for future work. setting in the way that our model suggested. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 33 and xit denotes a vector isof other character- istics thought to influence the representa- tive voter. (3) AiV=-t Shocks to voting behavior may be corre- lated with shocks to tax changes (ArTi). For example, the discovery of a toxic waste dump (Yl /o)4((txit + mY1 may require additional tax revenues for clean-up purposes. It may also have an in- t Ett+ l(Oit+ dependent effect on the governor's popular- ity: voters perceive that the governor has where +() is the density of the standard put them at risk. Thus we present estimates normal distribution. This becomes of the reelection equation in which tax changes in state i (Arit) are instrumented (4) m1 1Tit = - Xit-2 6T-it on state demographic and economic vari- ables. + r-1[-Ajo8r/(-y1 E{JVti1})]. As above, we allow incumbents to differ in the value they place on rent-seeking and If we use the linear approximation index the latter, in the case of the ith in- cumbent, by Ai. (Above we had only good y- E(_AVt /C1 8E{J7+1})] ~ Olzit + nit incumbents with A = 0 and bad ones with A = 1.) The cost of providing public services for some vector of state- and incumbent- is denoted Oit and is assumed to consist ofspecific a characteristics zit and a residual hit, known component ci and an independently we then have the following equation for the and identically distributed shock 8it. We tax change in state i: assume that an incumbent who can run for another term faces the following optimiza- (5) ATit =-( /yl)xXit + (a /yl)zit tion decision over taxes at date t: -(Y2 /yl) AT-it + n7it /yl (2) Vti(oit) where we have made a standard identifying assumption that o-e = 1. We allow both for idiosyncratic shocks = max -(it- it) and for year effects (Y) when estimating (5). rrit The latter may enter if, for example, busi- ness cycles or changes in federal fiscal pol- icy move states' taxes in a synchronous + R'(Arjt; Ar_jj5E t +1(Oit+1)}ltit ? oit} way.24 Equation (5) then becomes where we have normalized the payoff from (5') ATit = -( /yl)xit + (a /yl)zit not being reelected to zero and E{ I de- 23 -(Y2 / YO 5r-it + 1Y + Pit notes expectations. Equation (2) embodies the dynamic trade-off that the incumbent = Xit + a*zit + 5Ar-it + *Y+ Vit. faces; higher taxes today mean more rent but a lower probability of reelection. The Because of the potential interaction be- first-order condition associated with (2), as- tween neighboring states' tax increases due suming an interior solution where rit 2 Oit 24 We also allowed for spatial correlation in the shocks received by neighboring states. However, in 23This formulation does not allow incumbents to estimation we found no spatial correlation in the er- strategically influence voters' beliefs in the future by rors, and we removed reference to it here to simplify changing voter tax rates. the presentation. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 34 THE AMERICAN ECONOMIC REVIEW MARCH 1995 to strategic behavior, Ar-it on the right- and (v)] presents the probability of incum- hand side of (5') may be endogenous. To get bent defeat as a function of the change in consistent estimates of the coefficients taxes (t - [t -2]) in state i and changes in (13/y1), (a /y1), and (Y2/Y1) in this case, for state i's neighbors. For each in- taxes we may use either an instrumental-variables come group, increases in a state's own taxes approach or a maximum-likelihood estima- increase the probability of incumbent de- tion scheme. Instrumental-variable estima- feat. However, if neighboring states raise tion provides a check that correlation in taxes at the same time, then neighbors' tax taxes is not due to a common exogenous increases offset the effect of tax changes at shock experienced by neighbors: once in- home. strumented, correlation in taxes is due only These results are consistent with our to those parts of neighbors' tax changes that model of incumbent behavior.25 However, it are attributable to the state economic and is important to consider the possibility that demographic variables used as instruments. shocks in the incumbent-defeat equation An alternative to instrumenting for tax may be correlated with shocks in the tax- changes in the reelection equation (1) is to change equations. Thus, we present esti- estimate the equations (1) and (5') jointly. mates from models in which state tax Details of the joint estimation are presented changes are instrumented with state demo- in Appendix B. graphic variables and year effects. Specifi- There are two sets of overidentifying re- cally, the second set of estimates [columns strictions to test. The ratio of (- Y2 / Y1)'(ii) and (vi)] presents results in which tax identified from the election equation (1), changes have been instrumented on year should equal the spatial correlation coeffi- effects alone, and the third set [columns (iii) cient p identified from the tax-setting com- and (vii)] presents two-stage least-squares ponents of equation (5'). In addition, vari- estimates in which tax changes have been ables thought to influence a governor's instrumented on changes in the proportion reelection odds (elements of x) that are not of elderly individuals (greater than age 65) thought to determine the incumbent's ex- in the state's population, the proportion pected payoff from reelection (elements of young (ages 5-17), and year indicator vari- z) provide a second set of overidentifying ables.26 While it is possible for changes in restrictions: the ratio of (- 0 / y1), identi- the proportion of the population who are fied from the election equation (1), should elderly or young to have independent ef- equal corresponding elements of *, identi- fects on the reelection odds of incumbents, fied from the tax-setting equation (5'). overidentification tests fail to reject the hy- pothesis that the instruments can be ex- IV. Results cluded from the second-stage equation. The instrumental-variable results are virtually We estimated a number of specifications identical with and without the use of demo- of our equations for gubernatorial defeat and changes in taxes. Table 3 presents esti- mates of incumbent-governor defeat and re- tirement as a function of the tax change 25It is also consonant with some findings in Sam observed in the official's own state and that Peltzman (1992) who shows that incumbents are pun- observed in neighboring jurisdictions, using ished for spending growth. However, Peltzman does the TAXSIM data. We present estimates not consider the possibility of yardstick competition. for two different income categories: joint 26To increase the precision of the estimates, the first-stage regression for the instrumental-variables es- filers with no dependents earning $40,000 timation in columns (ii) and (vi) included all 48 states and $100,000 in 1977. For each income cat- for all years. Overidentification tests for the two-stage egory, we present four sets of estimates, least-squares results are F tests of the joint signifi- each based on different assumptions about cance of the instruments in regressions of the differ- ence between the election outcome and that predicted the underlying model. The first set of esti- (from the two-stage least-squares estimation) by the mates for each income group [columns (i) tax-change variable. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 35 TABLE 3-ESTIMATION OF INCUMBENT DEFEAT BASED ON LINEAR PROBABILITY MODELS USING TAXSIM DATA ON CHANGES IN INCOME-TAX LIABILITY, 1977-1988 (DEPENDENT VARIABLE: GOVERNOR DEFEATED OR RETIRED) Income = $40,000 Income = $100,000 Variable (i) (ii) (iii) (iv) (v) (vi) (vii) (viii) Own tax change 0.0004 0.0001 (1.44) (1.84) Own tax change (IV)a 0.0022 0.0006 (1.56) (1.67) Own tax change (2SLS)b 0.0015 0.0005 (1.57) (1.80) Neighbors' tax change - 0.0012 - 0.0014 - 0.0013 - 0.0005 - 0.0007 - 0.0007 (1.94) (1.80) (1.94) (2.85) (2.71) (2.82) Unanticipated own tax 0.0004 0.0001 changeC (1.35) (1.58) Unanticipated neighbors' -0.0008 -0.0004 tax changed (1.43) (2.31) A State income per -0.123 -0.005 -0.052 -0.144 -0.214 -0.286 -0.280 -0.216 capita ($1,000's) (0.79) (0.02) (0.29) (0.93) (1.42) (1.55) (1.56) (1.40) A Neighboring states' incomes per - 0.089 - 0.104 - 0.098 - 0.048 - 0.003 0.137 0.124 0.008 capita ($1,000's) (0.52) (0.47) (0.51) (0.28) (0.02) (0.61) (0.58) (0.05) A State's unemployment rate 0.082 0.088 0.085 0.088 0.069 0.043 0.046 0.083 (1.76) (1.48) (1.65) (1.87) (1.50) (0.76) (0.83) (1.79) A Neighboring states' -0.067 - 0.059 - 0.062 - 0.078 -0.045 - 0.011 - 0.014 - 0.073 unemployment rate (1.17) (0.80) (0.97) (1.35) (0.79) (0.16) (0.21) (1.28) A Total state debt -0.236 - 6.77 -0.502 -0.249 -0.317 -0.739 -0.700 -0.317 ($1,000's) (0.69) (1.24) (1.15) (0.73) (0.95) (1.45) (1.47) (0.93) A Total neighboring state debt 0.701 1.354 1.095 0.790 0.724 1.087 0.001 0.821 ($1,000's) (1.48) (1.74) (1.77) (1.48) (1.58) (1.80) (1.82) (1.76) Governor's age 0.024 0.022 0.023 0.023 0.025 0.022 0.023 0.023 (3.44) (2.48) (2.94) (3.25) (3.61) (2.76) (2.85) (3.56) Number of observations: 85 85 85 85 85 85 85 85 Overidentification test:e 0.706 0.640 (P value for F statistic): (0.716) (0.774) Notes: Numbers in parentheses are t statistics. "Re to run and do not run for Congress. "Unanticipa and that predicted by an ordinary least-squares regression that includes changes in state income per capita, unemployment, proportion elderly, and proportion young as explanatory variables. aInstruments = year indicators. bInstruments = year indicators and changes in the proportions of elderly and young. CAT E(Ari I xi, zi,Y). A-i -E(ATjIlx-,z-i,Y). eTest of exclusion of year effects and changes in proportions elderly and young in a residual regression. See text for details. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 36 THE AMERICAN ECONOMIC REVIEW MARCH 1995 graphic variables as instruments. The re- find that older governors are significantly sults from instrumental-variable estimation less likely to be reelected. are consistent with those presented in the The results presented in Table 3 suggest first set of columns: own tax changes in- that voters are sensitive to the tax changes crease the probability of incumbent defeat, they face, relative to those observed in and neighbors' tax changes reduce the prob- neighboring states, and that this sensitivity ability. translates into votes against an incumbent All of our estimations in Table 3 allow whose tax changes are high by regional stan- gubernatorial defeat to depend upon the dards. The impact of such comparisons on state's relative economic performance by in- gubernatorial behavior can be seen in Table cluding changes in state income per capita 4, which presents results from tax-setting and neighboring states' changes in income equations. We model tax change as a func- per capita as explanatory variables in the tion of state economic variables (including defeat equation. Changes in unemployment change in real state income per capita and rates both at home and in neighboring states state unemployment) and state demo- are also allowed to affect the governor's graphic variables (including change in the reelection odds. While we find that a state's proportion elderly and in the proportion indicators of economic well-being signifi- young in the population). We also include cantly affect the probability of the governor's state and year effects. The latter will absorb reelection, we find little evidence that voters the impact of changes in national economic measure a governor's relative performance climate and changes in federal fiscal behav- in this way. Increases in state unemploy- ior that may have similar effects on all states. ment significantly increase the probability of That governors often face binding term a governor's defeat under most specifica- limits, under which they are not allowed to tions tested. However, neighbors' unem- run for reelection, gives us a simple (some- ployment rates have insignificant effects on what less structural) test of our yardstick- the odds of reelection. Increases in state competition model.28 If neighboring states' income per capita increase the probability tax rates are interdependent because of of reelection for $100,000 filers.27 while yardstick competition, then tax rates among changes in income per capita in neighboring neighbors should be uncorrelated in those states do not appear to influence reelection years in which a state is run by a governor probabilities. Thus, while it is possible for who cannot run for reelection. Sensitivity to citizens to give the governor a relative grade neighbors' taxing behavior should be mani- based on these criteria, it does not appear fest only during those years when the gover- that voters are judging governors in this nor is eligible to run again. This is consis- way. This may be because such measures tent with our findings in both tax data sets: are regarded as a less good barometer of a in years in which a state is governed by a governor's performance than taxes. lame duck, there is no sensitivity to neigh- We include retirements in our election- bors' tax behavior. However, in states where outcome measure to capture retirements the governor is eligible to run again, we find taken by governors who anticipate defeat. in both data sets that when a neighboring We add the incumbent governor's age as an state increases/decreases taxes by one dol- explanatory variable in our reelection re- lar, the home state will increase/decrease gression to control for retirement due to taxes by roughly 20 cents. We take this as physical, rather than political, reasons. We 28We owe the idea of splitting the sample to an 27Here, collinearity between state income per capita anonymous referee. The raw correlations between tax and state unemployment rates reduces the significance changes in neighboring states, presented in Table 2, of state income estimates. Both appear to be picking are also larger when the sample is restricted to states up the same effect. in which governors are eligible to run for reelection. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms  VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 37 TABLE 4-ESTIMATION OF STATE TAX CHANGES Dependent variables Change in sales, income, Change in income-tax liability, and corporate taxes $40,000 joint filers 1979-1988 per capita, 1962-1988 Governor Governor cannot Governor can cannot Governor can run for run for run for run for reelection reelection reelection reelection Explanatory variable OLS OLS 2SLSa OLS OLS 2SLS Neighbors' tax change - 0.006 0.305 0.746 0.086 0.216 0.538 (t - [t - 2]) (0.05) (2.49) (1.81) (1.01) (3.23) (1.96) State income per capita -0.011 -0.068 -0.073 0.023 0.016 0.014 (t - [t -2]) (0.34) (2.09) (2.16) (3.70) (3.84) (2.80) State unemployment rate 9.13 17.35 18.52 -0.665 -3.17 -2.19 (t - [t - 2]) (1.58) (1.71) (1.77) (0.45) (2.07) (1.25) Proportion young (aged 5-17) -3,381.30 -3,680.97 -356.80 631.96 618.10 545.51 (t - [t - 21) (0.74) (0.80) (0.92) (2.17) (2.63) (2.17) Proportion elderly (aged 65 +) 4,315.03 15,791.35 12,813.98 1,287.50 512.57 697.88 (t -[t -2]) (1.05) (2.33) (1.72) (1.75) (1.18) (1.46) Governor's age - 7.75 -0.126 0.027 0.323 -0.118 - 0.096 (2.12) (0.06) (0.01) (1.12) (0.50) (0.38) Number of observations: 113 302 302 354 846 813 Overidentification test:b 1.26 0.20 (P value): (0.287) (0.820) Notes: Numbers in parentheses are t statistics. All regressions include state and year indicator variables. OLS denotes ordinary least-squares analysis; 2SLS denotes two-stage least-squares analysis. aFirst-stage regression using TAXSIM data: Change in Neighbors' Tax Liability = Constant + 14.70 (Neighbors' Change in Unemployment Rate) [t = 1.66] -3.99 (Neighbors' Change in Unemployment Rate Lagged) [t = 0.43] -0.092 (Neighbor's Change in Income per Capita Lagged) [t = 3.79] + 5,551.04 (Neighbor's Change in Proportion Young Lagged) [t = 2.50] + state and year indicators and own state covariates (those that appear in table above) (number of observations = 302, R2 = 0.4413; observations for 1987 and 1988 restricted to states with information available on whether incumbent governor can run in next election). First-stage regression using sales, income, and corporate tax data: Change in Neighbors' Taxes = Constant + 0.027 (Neighbor's Change in Income per Capita Lagged) [t = 6.97] + 4.28 (Neighbors' Change in Unemployment Rate Lagged) [t = 3.23] +year and state indicators and own state covariates (those that appear in table above) (number of observations = 813, R2 = 0.7889). bF test of significance of instruments in regression: [Art - bAr_'] on own state covariates and state and year indicators, where b is the estimated coefficient from the two-stage least-squares regression. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 38 THE AMERICAN ECONOMIC REVIEW MARCH 1995 fairly strong evidence that political calcula- to reduce the government's tax revenues. tions are influencing governors' behavior. If This is consistent with results in columns 5 sensitivity toward neighbors' taxes was due, and 6: ceteris paribus, an increase in unem- say, to states sharing common shocks, this ployment reduces the taxes collected by the sensitivity should be as apparent in the years state. The same reasoning suggests that in- in which governors were bound by term come growth may be negatively related to limits as in those years in which they are the income-tax liability of $40,000 filers, and eligible to run again. positively related to the sales, income, and If tax-setting behavior is strategic, we ex- corporate taxes collected. This is also con- pect state tax changes to respond to tax sistent with results presented in Table 4. changes in neighboring states (and vice Governor's age has been added to the versa). To cope with the potential endo- tax-setting equation because of its potential geneity problem, we present estimates in effect on the governor's reelection odds. Table 4 of the impact of neighbors' taxes on Using the notation of Section III, this vari- tax changes at home, using two-stage least- able belongs to x (determining reelection) squares estimation, for the governors who but not to z (payoff from reelection). This is can run for reelection.29 These results are a variable that may be used in overidentifi- the last column of estimates for each data cation tests; we will discuss these tests for series. Although imprecisely estimated, the the maximum-likelihood estimates below. results from both data sets suggest that The two-stage least-squares estimates in neighbors' tax changes are positively corre- Table 4 are consistent in the presence of lated with a state's own tax change. Since correlation between shocks to the voting neighbors' tax changes are instrumented, and tax-setting equations. They are also this correlation is not attributable to com- consistent if there is spatial correlation in mon unobservable shocks that may have hit the errors of the tax-setting equation, be- neighboring states; the correlation is in the cause we have instrumented for neighbors' component of neighbors' tax increases that tax changes. However, these estimates are is attributable to neighbors' observable vari- not efficient if there is correlation in the ables, used here as instruments. shocks to the tax-setting and voting equa- The two tax variables are different mea- tions. For this reason, we have estimated sures of state taxes, and for this reason, we these equations jointly, using data on per expect them to respond differently to capita sales, income, and corporate taxes. changes in economic and demographic vari- We present these results in Table 5.3? ables. For example, if unemployment in- The results of joint estimation for coef- creases in the state, this is apt to place a ficients on tax-setting variables are almost fiscal strain on the state and result in an identical to those found in Table 4. With increase in the income-tax liability of respect to the tax-setting equation, neigh- $40,000 filers. This is consistent with the bors' tax changes continue to have a posi- results presented in columns 1-3: ceteris tive and significant effect on a given state's paribus, an increase in the unemployment tax changes; a one-dollar increase in neigh- rate has a positive and significant effect on bors' taxes results in roughly a 20-cent in- the tax liability of $40,000 filers. However, crease in a given state's taxes. Increases in using instead the per capita taxes collected by the state as a tax measure, we might expect increases in the unemployment rate 30We attempted to estimate a joint likelihood using our data on changes in income-tax liabilities of $40,000 filers. However, when year indicators were included in 29The neighbors' instrument list includes neighbors' the model, the program would not converge. A GAUSS demographic and economic variables and neighbors' program to estimate the joint likelihood is available demographic variables lagged. from the authors upon request. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 39 TABLE 5-MAxIMuM-LIKELIHooD ESTIMATION We can formally test whether the sensitiv- OF VOTING AND TAx-SETTING BEHAVIOR ity to neighbors' tax changes is of a size Coefficient using data on consistent with the yardstick-competition changes in sales, model, by testing whether p = - )2 / 'y1i The income, and corporate likelihood-ratio test statistic associated with tax per capita, constraining this relationship to hold is 4.48. Variable 1962-1986 Although the rejection holds in a 90-percent Tax change coefficients: confidence interval, it is not a strong rejec- Neighbors' tax change 0.177 tion. We find the results to be broadly con- (t -[t -2]) (3.92) sistent with the model presented in Sections State income (t -[t -2]) 0.017 II and 111.31 (5.68) State unemployment rate -3.313 V. Extensions and Alternative Models (t -[t -2]) (2.79) Proportion young 6.563 A. Consistency of the Results with the (t - [t - 2]) (1.84) Tiebout Model Proportion elderly 6.988 (t - [t - 2]) (4.77) It is interesting to speculate whether our Governor's age 0.135 results are consistent with Tiebout-style tax (0.73) competition based on factor mobility. At first sight, a negative effect of own taxes on Year effects yes reelection is hard to justify in a Tiebout Incumbent-defeat coefficients: framework: individuals should move if they Own tax change 0.015 are dissatisfied with the tax change. This (t - [t - 2]) (1.14) would leave only contented voters in the Neighbor's tax change - 0.033 state and thus enhance the probability that (t -[t -2]) (1.99) the incumbent is reelected. Likewise, in- State income - 0.317 creases in taxes in a neighboring state would (t - [t - 2]) (1.38) lead to an influx of voters into a state that State unemployment 0.044 disliked high taxes, thus lowering the aver- (t - [t - 2]) (0.79) age tolerance to taxes at home.32 Thus in- Governor's age 0.023 creases in neighbors' taxes tend to decrease (2.25) the probability that an incumbent will sur- vive. At face value, therefore, both of the Numbers of observations: predictions of the Tiebout model would be Tax-setting 846 Election 266 contrary to what we find in our empirical results. Note: Numbers in parentheses are t statistics. It is important to acknowledge that some stories based on factor mobility could be consistent with our results. Suppose that higher taxes lead businesses to relocate and unemployment continue to reduce the per that this reduces property values, which capita taxes collected, while increases in state income per capita add to per capita taxes collected. In addition, taxes increase with an increase in the proportion of elderly 31Note that we cannot reject the null of equality in people and young people in the population. our second set of overidentification tests: P1* = Consonant with the theory presented (1- ,I/ y1). However, this is only because the standard above, the probability of incumbent defeat errors on the coefficients in the tax-setting equation are large. is increased by an increase in state taxes. 32However, to the extent that taxes are capitalized However, this effect is offset if neighbors into property values, the incentive to move would be increase their taxes simultaneously. weakened. This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 40 THE AMERICAN ECONOMIC REVIEW MARCH 1995 makes voters unhappy. This is a hybrid of neighboring states. In addition, this ap- our model and a Tiebout approach. Other proach would give way to yardstick competi- explanations that emphasize factor mobility tion between states in unanticipated tax might also be possible. All of this notwith- changes. We would be unable, in this world, standing, we find the simpler and more di- to distinguish correlation between neigh- rect explanation of the relationship between bors' taxes that is due to common shocks vote-seeking and tax-setting spelled out from that which is due to strategic behavior. above to be a reasonable working hypothe- Given that voters appear to respond to both sis based on the evidence. However, future anticipated and unanticipated tax changes, work might be able to suggest ways of dis- we are comfortable with the assumption that tinguishing between alternative explana- voters condition on neighbors' tax changes, tions. without regard to whether it is a change that could have been anticipated with B. Alternative Voter Information Sets enough information. Nonetheless, the ques- tion of what information voters have and In our model, voters react to tax changes. use to evaluate their incumbents is worthy However, if voters understand the way in of further investigation. which changes in demographic and eco- nomic variables influence tax changes, then C. A More General Model they should penalize incumbents only for that part of any tax change which is unantic- Our model of incumbent and voters' be- ipated, given economic and demographic havior is somewhat special in focusing only changes, and not matched by neighboring on tax-setting. In reality there is a whole states. Thus imagine that each voter re- array of incumbent actions about which vot- gresses taxes on state characteristics and ers care, directly or indirectly, and which estimates a predicted tax change based on they might use to decide whether or not to changes in the right-hand-side variables of reelect an incumbent. A more general ap- this regression. The cost "shock" is then the proach to the issues treated here would residual of such a regression. This view sug- involve studying the links between all as- gests that it is the residual in this regres- pects of incumbent behavior and electoral sion, relative to the residual in such a re- performance. This exercise requires consid- gression for neighboring states, that indi- erably more effort in data collection and cates whether a tax increase is justified. analysis and must be left for the future. To test whether this is the case, refer to Nonetheless, we have some preliminary columns (iv) and (viii) of Table 3, which findings to report. present our estimates of the effect of unan- Tax and debt are substitute ways of fi- ticipated tax changes on gubernatorial re- nancing expenditures; some would even take election using the TAXSIM data. Here, we the Ricardian view that they are equivalent. see a pattern consistent with that observed It is thus interesting to know whether they in the other columns of the same table: have the same effect on gubernatorial re- unanticipated own tax increases reduce the election chances. The literature suggests odds of reelection, while unanticipated in- possible cross-cutting reasons for asymme- creases in neighbors' taxes increase the tries between the political effects of debt probability of reelection. and taxes. The public-choice tradition (e.g., In spite of this finding, a word of caution James M. Buchanan and Richard E. Wag- seems necessary. One might question the ner, 1977) has often argued that voters do plausibility of assuming that voters are do- not perceive the true cost of debt finance. ing regression-based evaluations of incum- On the other hand, current regulation of bents in their heads. In forming estimates of bond finance through referenda serves to unanticipated neighbors' tax increases, vot- increase debt's visibility and cost. Of course, ers must be versed in the demographic and such hurdles could be put in place by voters economic conditions in both their own and who fear excessive deficits because debt is This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms VOL. 85 NO. 1 BESLEYAND CASE: INCUMBENT BEHAVIOR 41 less visible. In either case, there is no rea- a significant determinant of who is elected, son to think that taxes and debt would have rationalizing effort put into curbing tax in- the same effect on reelection odds. To in- creases that are out of line with neighbors. vestigate the effect, we included changes in Much scope remains, however, to study the level of state debt (t - [t - 2]) and contexts in which policy choices and politi- changes in neighbors' debt as right-hand- cal fortunes are jointly determined. There side variables in the reelection equations of may be additional implications of yardstick Table 3.33 Own state debt levels do not competition for governmental behavior that appear to affect significantly the odds of can be explored. Thus our paper only con- being reelected. This finding could also be stitutes a beginning. Data on the U.S. states explained by our use of total long-term are a rich source for exploring these issues. debt, while in reality only certain kinds of It is also a natural situation in which to debt finance may be politically sensitive. consider yardstick competition, given that A more complete model would also allow there is a significant common component to voters to evaluate governors with respect to incumbents' environments. expenditures (both level and' composition). Recent work by Case et al. (1993) suggests APPENDIX A that state spending may respond to spend- ing decisions made in neighboring states. PROPOSITION Al: In any equilibrium of While a complete expenditure model lies the tax-setting game, good incumbents behave beyond the scope of this paper, preliminary as {r(OL, G) = r1, r(OM, G) =T2' T(OH, G) = investigation found no effect of changes in 73}, and voters use {,(LLG) = 1, p(r4) = O, expenditure on the probability of ,A(75) = 0}. Bad incumbents set taxes as fol- reelection.34 lows. VI. Concluding Remarks (i) If qH 2 12, then: r(OL,B)=T3 T(OM,B)=T3 r(OH,B)=T5 The main achievements of this paper are twofold. First, we have demonstrated the /(LC2)=l 1L(3)=l. importance of jointly estimating incum- (ii) If qH < 2, then there are three c bents' policy choices and their likelihood of reelection. If some policies yield electoral (a) if qL 2 12 then: success, then we would expect to see more (72 with probability q m / of them. Second, we have shown why one might expect a kind of yardstick competi- 'r(OL,B) = 3with probability tion in the political sphere. q- ( qm /qL We have studied yardstick competition in states' tax-setting decisions between 1960 r(OM,B) =74 r(OH,B) =75 and 1988. The results are encouraging to the view that vote-seeking and tax-setting ('r2) = 1/(28) 4(73) = 0. are tied together through the nexus of yard- (b) if qL < qH < 2, then: stick competition. Tax changes appear to be r(OL,B) = 73 (73 with probability 33These are changes in total state debt outstanding at the end of the fiscal year (Source: State Gouernment rT(Om,B)= (H-q)q Finances, various years). ) 74 with probability 34Using data on total state expenditure from 1950 to 1990, we found no significant effect of changes in (qm + qL -qH)/qm expenditure on the probability of incumbent defeat. This is true whether or not one controls simultaneously 7(0H,B) = 75 for changes in state income per capita, state unemploy- ment rates, and taxes collected. (u72) =1 A(r3)=1/(28). This content downloaded from 87.146.199.24 on Sun, 21 Apr 2024 15:27:29 +00:00 All use subject to https://about.jstor.org/terms 42 THE AMERICAN ECONOMIC REVIEW MARCH 1995 need to check that the voter is indifferent (c) if q H < q