Legislative Term Limits, Polarization, and Representation

Transcription

1 Legislative Term Limits, Polarization, and Representation Michael Olson 1 and Jon Rogowski 2 1 Graduate Student, Department of Government, Harvard University 2 Assistant Professor, Department of Government, Harvard University November 21, 2017 Abstract How do legislative term limits affect representation? Despite proponents arguments that term limits reduce the level of partisan conflict and improve the quality of representation, these expectations have been subjected to little empirical scrutiny. In this paper, we study the effect of term limits in state legislatures from 1993 to At the level of collective representation, term limits exacerbate partisan polarization by increasing the ideological gap between Democrats and Republicans in term-limited states. The effects are greatest among Republican legislators, states with more professionalized legislatures, and legislators serving in lower chambers. Additional analyses suggest that term limits may improve partisan representation at the dyadic level but contribute to aggregate polarization through leapfrog representation. At best, term limits appear not to have achieved the goals articulated by their proponents; at worst, they may have exacerbated the legislative consequences of contemporary partisanship. Brian Lash was a collaborator on an earlier version of this manuscript and we thank him for his contributions to the project. We are grateful to Corban Ryan, Ricky Rodriguez, and Michael Scherr for excellent research assistance and Steve Rogers for sharing some of the data used in this project. Rogowski thanks the Faculty of Arts and Sciences at Harvard University and the Department of Political Science at Washington University in St. Louis for generous research support.

2 Contemporary frustrations with gridlock in the United States Congress have renewed interest in term limits as a potential solution. Political figures from both sides of the partisan aisle have expressed support for term limits. For instance, in a debate among candidates for the 2012 Republican presidential nomination, former Utah governor Jon Huntsman declared that [w]e need a Washington that works... we have a Congress that can t even figure out how to balance the budget. They need term limits. 1 Similarly, in an interview upon leaving Congress, former Senator Joe Lieberman remarked that Congress might be healthier and less partisan and less rigid if it turned over more often, and term limits are one way to do that. 2 Even sitting members of Congress have expressed support for limiting their own terms in office 3 and have introduced legislation in recent congresses to amend the Constitution and limit the number of years served by members of the House and Senate. 4 As the quotes above suggest, proponents of term limits argue that limiting the number of terms legislators can serve in office would reduce the level of partisan conflict, encourage compromise and cooperation, and improve the quality of representation. Consistent with these claims, scholars have argued that term limits put an end to politics as usual (?) by producing citizen legislators who are more likely to behave in the public interest (???), more responsive to public opinion and constituent needs (???), and less susceptible to sanctions threatened by party leaders (?). Other scholars argue, however, that term limits produce 1 CNN National Security Debate, November 22, 2011; transcript available at com/transcripts/1111/22/se.06.html. 2 Nick Wing, December 17, 2012, Joe Lieberman: Term Limits Would Make Congress Healthier And Less Partisan, The Huffington Post; joe-lieberman-term-limits_n_ html. 3 See, e.g., Rep. Mike Gallagher, A Time for Term Limits, The Hill, April 3, 2017; available at thehill.com/blogs/congress-blog/politics/ a-time-for-congressional-term-limits (accessed August 2, 2017) 4 Rebecca Shabad, February 5, 2014, The Hill, House GOP Bill Pushes Term Limits ; available at house-gop-lawmakers-file-bill-to-impose-12-year-term-limits-on; Jordain Carney, Cruz, DeSantis Push for Term Limits, The Hill, January 3, 2017; available at com/blogs/floor-action/senate/ cruz-desantis-push-for-congressional-term-limits (accessed August 2, 2017). 1

3 more partisan legislatures. For instance, term limits may reduce legislators incentives to moderate their behavior, instead choosing to vote with the party over their constituents (?;?;?; but see also?). Term limits may also increase the role of parties in recruiting legislative candidates (?) and crafting legislation (?), thereby increasing party loyalty (?; but see also?). In this paper, we study the effects of term limits on legislative behavior in the American states. After more than two decades of experience with term limits, strikingly little is known about their effects on political representation. 5 This omission is surprising given the expansive literature on the effects of legislative term limits on other important outcomes, including fiscal policy (????), legislative professionalism (?), policy complexity (?), the partisan composition of state legislatures (?), descriptive representation (?), redistricting procedures (?), electoral competition (??), the incumbency advantage (?), attentiveness to district-level concerns (??), cosponsorship activity (?), the distribution of power within the legislature (?), and the balance of power across chambers (?) and branches of government (???). 6 We report new evidence about the relationship between term limits and representation in state legislatures at the collective and dyadic levels. Using a panel design and data on state legislative roll call voting patterns from 1993 to 2014, we show that term limits produced systematically higher levels of polarization in state legislative voting patterns by increasing the ideological gap between Republicans and Democrats voting records. These results are robust across a wide range of model specifications, identification strategies, and characterizations of key variables. We show that while term limits had roughly equivalent effects on legislative leaders and rank-and-file members and among legislators who represented safe and marginal districts, their effects were substantially greater among Republicans, legislators serving in state lower chambers, and states with more professionalized legislatures. At 5? is one notable exception, who reports null findings. 6? and? provide thorough reviews of this literature. 2

4 the dyadic level, we find no evidence that our aggregate patterns reflect decreased legislator responsiveness to constituents; instead, the data suggest that term limits contributed to aggregate polarization by combining improved partisan representation of district preferences with leapfrog representation. Our findings provide the most comprehensive evidence to date on the consequences of term limits for legislative behavior and representation. At best, term limits appear not to have achieved the goals articulated by their proponents; at worst, they may have exacerbated the legislative consequences of contemporary partisanship. Term Limits and Representation Though American interest in term limits has intensified over the last quarter-century, they are hardly a novel idea. The ancient Greeks favored a view of representatives as citizen leaders rather than career politicians and limited many elected officials to just a single term in office. The American Founders also initially supported legislative term limits. During the Continental Congress, for instance, Thomas Jefferson (?, 373) warned of every danger that might arise to American freedom by continuing too longer in office, 7 and the Articles of Confederation codified this view by stipulating that a delegate could not serve for more than three years out of every six year period. During the twentieth century, term limits for national legislatures were enshrined in new constitutions in countries including Costa Rica, Ecuador, and the Philippines (?). The term limits movement in the United States in the last several decades resulted in their passage in 20 states. 8 Term limits are likely to affect the quality of representation that legislators provide by altering their time horizons. Because term limits affect the rate of legislative turnover, they have implications for the career calculations made by potential legislative candidates and incumbent legislators. Not only do term limits forbid legislators from seeking office 7 Jefferson introduced this resolution to the Second Continental Congress in Term limits were later repealed or struck down in six of these states. 3

5 once they have served the allotted number of terms, but term limits may also encourage state legislators to seek higher office (such as state senate or U.S. House) earlier than they otherwise would (?). The increased availability of open seat contests may appeal to potential officeseekers who would be less inclined to challenge an incumbent, and proponents of term limits have argued that this would facilitate the election of legislators from underrepresented groups, including women and people of color (??). 9 However, term limits also change the value of officeholding to would-be candidates. Individuals drawn to public service or who possess policy expertise and are motivated to craft good public policy may be less inclined to seek election to a state legislative position whose time horizon is relatively short (???). In addition to the direct effects of shortened time horizons, term limits may also indirectly alter the incentives for legislators to represent their constituents. The threat of electoral sanction provides incentives for legislators to represent their constituents while in office (???). Most obviously, therefore, term limits sever the electoral connection (?) for officeholders serving in their final legislative terms. In the context of governors, for instance, this logic may explain why state economies fare worse under incumbent governors who are prevented by term limits from seeking another term in office compared to states with re-election-eligible incumbents (?). In addition, legislators seeking to advance to higher office may perceive that audiences other than their district constituencies hold greater importance for realizing their career ambitions. Thus, legislators representing similar districts may nonetheless represent their constituencies in different ways on the basis of term limits. The mechanisms described above may increase the roles of parties and interest groups for legislative behavior and in the legislative process. For example, if term limits reduce the attractiveness of legislative service due to steeper opportunity costs, career ambitions may not be sufficient to induce potential candidates to seek election to state legislatures. 9 There is little evidence to support this claim, however.? characterize the claim that term limits would usher in a new breed of citizen legislators as the dog that won t bark. 4

6 Instead, state and local parties may play larger roles in recruiting candidates to run for these positions while securing support for party agenda items in exchange for campaign resources (??). Similarly, legislators seeking to move up to higher office may place increased emphasis on securing future electoral support from parties and interest groups and thus vote in ways that reflect the priorities of those groups. In addition, the replacement of experienced incumbent legislators with inexperienced legislators could increase the opportunity for parties and lobbyists to fill the expertise gap and shape the content of the legislative agenda and the information legislators bring to bear on those proposals (???). Evaluating the Effects of Term Limits Despite widespread attentiveness to party polarization and proponents emphasis on term limits potential for reducing legislative gridlock and partisan influences in legislatures, few studies have examined this relationship. Using data from roll call votes in state legislatures in the sessions,? provides the best empirical evidence to date on term limits effects on polarization in state legislatures. In comparing legislative behavior in states with and without term limits, Wright finds no evidence that term limits are associated with greater polarization at the aggregate level, nor does he find that that individual legislators compile more ideologically extreme voting records in states with term limits. While Wright s data collection is impressive, the research design limits strong conclusions about the effect of term limits on polarization. As Wright shows, polarization varies considerably across states due to factors beyond the implementation of term limits, and thus a cross-sectional comparison of chambers and legislators cannot tell us whether polarization increased or decreased in term-limited states compared to what would have been observed in their absence. Other research on term limits provides similarly mixed conclusions about their implications for polarization and representation. For instance, in a study of the term-limited 5

7 California legislature,? found that legislators about to be termed out vote with their party more frequently, but downplay the role of term limits as a major source of polarization.? take advantage of a natural experiment that randomly imposed term limits on legislators in the Arkansas state senate, and similarly find no evidence that term limits increased ideological shirking.? report that term limits have had negligible effects on electoral competition and argue that their impact on legislative polarization may have had more substantial had they not been enacted during an era in which polarization increased across all levels of government. In more recent research on term limits in the Michigan legislature,?, 72-3 conclude that term limits have unfastened the electoral connection between voters and legislators, but it is unclear whether these findings would apply more systematically across states. Previous work suggests a number of potential mechanisms through which term limits may affect representation. For instance, term limits may empower ideologically extreme elites in the form of party organizations, interest groups, or potential legislators themselves, and thereby pull legislators and legislative outcomes away from the preferences of district and statewide median voters. Because term limits reduce the number of times legislators can seek office, they may unfasten the electoral connection (?) and induce legislators to shirk from representing constituent preferences. Alternatively, term limits may strengthen the electoral connection by forcing constituents to choose between fresh new faces every few electoral cycles and thus reduce the availability of cues, such as the incumbency and the personal vote, that may otherwise obscure the relationship between district preferences and legislative behavior. Term limits could also strengthen the linkage between public opinion and legislative behavior because, by reducing the value of holding office, term limits increase voters ability to choose legislators likely to advances their interests (?). 10 Others have argued that term limits induce a Burkean shift (??) in legislative voting records in which 10 This finding would also be consistent with the findings from Lax and Phillips (2012), who show that state policies are more responsive to public opinion in states with legislative term limits. 6

8 legislators behave more as trustees by voting for policies they believe are in the long-term interests of their constituents rather than as delegates who are tethered to constituent opinion and electoral pressures. We proceed by testing consequences of term limits for representation from the aggregate to the dyadic levels. We begin by describing the implications of term limits for the collective representation of state electorates by examining whether term limits affect aggregate party polarization in state legislators. We then explore possible institutional and individual-level moderators before examining the potential microfoundations of these results in the dyadic context. To preview the results that follow, our findings suggest that seemingly contrary arguments that term limits may increase polarization and improve district-level representation are not as mutually exclusive as they may at first appear. Data and Empirical Strategy We begin our analysis by studying the effects of legislative term limits on aggregate levels of polarization between 1993 and In an ideal scenario, we would randomly assign states to treatment and control conditions, where states assigned to the treatment condition would implement term limits, and control states would not. Of course, such a research design is not possible in the context of the U.S. states. Instead, we rely on observational data to examine whether term limits increased polarization in state legislatures. Our measures of polarization are based on annual estimates of legislative ideology developed by?. These measures use roll call data for all state legislative chambers to characterize legislators voting behavior and are constructed such that they are comparable across states and time. We use these estimates to construct a state-level measure of Legislative Polarization that reflects the difference in roll call estimates between the median Democratic 7

9 and Republican members of the state legislature. 11 The values of this variable range from approximately zero to 3, with higher levels indicating states and chambers with more polarized legislatures. As? show, these measures document considerable variation in polarization across states and legislative chambers. Because our focus is on party polarization, we exclude Nebraska and its unicameral, nonpartisan legislature from our analysis. Our primary independent variable is an indicator, Term limits, for whether term limits were in effect in a given state-year. Term limits were implemented in 14 states during the period under study; they took effect first in California and Maine (1996) and most recently in Nevada (2010). 12. We provide a graphical representation of the adoption of term limits in Figure??, which plots the share of states using term limits (for their lower chambers) over the period of study, as well as which states were added to those using term limits for each jump. While most states that have implemented term limits did so in the late 1990s or early 2000s, we have good variation in time of implementation across the period of study. We use a panel design and leverage within-state changes in the presence of term limits to identify their effects on polarization. We assume, therefore, that the adoption of term limits is orthogonal to potential outcomes after we condition on time-invariant state characteristics and, in some models, a battery of time-varying covariates. This assumption of parallel trends 11 We use this aggregate measure to avoid chamber-specific polarization measures that are very sensitive to sample size, as most state legislatures upper chambers have fewer than fifty members. However, as we describe below, we obtain similar patterns to those reported in the main text when distinguishing state upper and lower chambers. 12 One may argue that the date of enactment is a more appropriate indicator of treatment status since legislators may adjust their behavior in anticipation of the impending change (?). Though we simply lack sufficient data on legislative voting records to examine the effects of term limits based on when they were enacted (the vast majority of which occurred in 1990 or 1992, prior to the starting data of the Shor-McCarty data), we are comfortable using the implementation date as the indicator of treatment status for several reasons. First, our theoretical perspective suggests that legislators that are termed-out are replaced by legislators who differ from them in fundamental ways. Thus, the effect of term limits on polarization is posited to result from, at least in part, the replacement of termed-out legislators rather than changes in behavior from existing officeholders. Second, to the extent that changes in polarization were due to the enactment of term limits rather than to their implementation, we are likely to under estimate the magnitude of their effects, thus making our empirical strategy a more difficult test of their effect. And third, research on other consequences of term limits reports similar results whether the date of enactment or implementation is used (see, e.g.,?) 8

10 0.3 NV LA OK Share of State Lower Houses with Term Limits CA ME AR CO MI AZ FL MT OH SD MO Year Figure 1: Adoption of Term Limits maintains that the treated states that adopted term limits reacted to them in the same way that control states would have reacted if they had adopted term limits. Given the circumstances surrounding the passage of term limits, the relatively short duration of our panel, and the subsequent plausibility of accounting for potential confounders through the use of state fixed effects in addition to time-varying controls, our approach provides credible causal estimates of the effect of term limits on state legislative polarization. We employ the difference-in-difference (DID) estimator with the following linear regression model: Y ijt = β 0 + β 1 Term Limits ijt + X ijt Ω + D i + T t + ε ijt, (1) where Y is the level of polarization in chamber j in state i in year t, Term limits indicates whether state i had term limits in effect in year t, Ω is a vector of coefficients for a matrix of time-varying state covariates X it described below, D i is an indicator for each state, T t is 9

11 an indicator for each year, and ε it is a random error term, which we cluster on state. We also collected data on a variety of other state-level covariates that may explain variation in state polarization and could confound the potential effects attributed to term limits. First, to account for structural features of state governance, we include indicators for the presence of Divided government, whether the state has a Democratic governor, Legislative professionalism (based on?), the share of seats held by the chamber majority party (Majority seat share), and the Democratic seat share in the chamber. We also account for attributes of the state population, including Population (logged), Per capita income, the Republican share of the vote in the most recent presidential election (GOP presidential state share), and Unemployment rate. Finally, as? find that congressional polarization is closely correlated with both immigration and income inequality at the national level, we also include annual measures of each state s Percent foreign-born and Gini coefficients. Summary statistics for all of these variables are presented in Table??. As noted above, state and year fixed effects are included in all our models which account for time-invariant attributes of states and secular trends that might affect polarization. Table 1: Summary Statistics of Key Variables Variables Mean Median Min Max St. Dev. Legislative Polarization Term Limits Divided Gov Democratic Governor Leg. Professionalism Absolute Difference Seat Share Democratic Seat Share ln(population) Per Capita Income GOP Presidential State Share Unemployment Rate Percent Foreign Born State Gini Coefficient

12 Panel Evidence We begin by estimating our baseline model, which represents Equation (1) but includes only our indicator for Term limits along with state and year fixed effects. The results are shown below in Table??. The coefficient estimate for Term limits is positive (0.105) and statistically significant, providing strong evidence that term limits increased legislative polarization. In models 2 through 4, we add to the model the various state-level controls discussed above. Across each model, we continue to find that term limits were associated with significant and positive increases in legislative polarization. In addition, the coefficients for Term limits are relatively consistent in magnitude across each specification, ranging from to The results shown in Table?? are substantively meaningful in addition to statistically significant. We evaluated the substantive magnitudes of our estimates by comparing the results from Table?? to the within-state variation in polarization. The average withinstate standard deviation is 0.119; restricting the sample to states without term limits, the comparable value is Our reported point estimates therefore constitute a shift of nearly one standard deviation in within-state polarization levels. We also find that some of our control variables are significantly associated with polarization. Increases in legislative professionalism and population were associated with greater polarization, while increases in the size of legislative majority parties and per capita income are significantly associated with declines in polarization. The coefficient estimates for Divided government are negative yet they are all small in magnitude and none are statistically significant. We also find no significant differences in polarization based on the partisanship of the governor, the party composition of the state legislature, or the partisan composition of the electorate (as measured by state voting patterns in presidential elections). Finally, we do not find support for the claims of? in the context of state-level polarization: both higher 13 The median within-state standard deviation is in both cases slightly lower. 11

13 foreign-born population shares and higher state-level inequality are negatively associated with polarization, though neither is statistically distinguishable from zero. Robustness Checks The results presented in Table?? are consistent across a wide range of additional analyses. We estimated a number of alternative models to highlight the plausibility of our assumptions and the robustness of our results across model specifications. We discuss these additional analyses below and present the results in the Supplementary Materials in the interest of space. First, because our binary indicator for states with term limits ignores qualitative differences in term limits enacted across states, we adopt a continuous measure of termlimitedness reported in? and?. This measure describes the change in turnover due to term limits relative to turnover from earlier years. We replace our binary treatment variable with this measure in the same panel specification as above, 14 and report the results in Table??. Using this measure, we continue to find strong evidence that term limits significantly increased polarization in state legislatures. Second, we employ two alternative modeling strategies to guard against common concerns about the differences-in-differences framework in a panel setting. Because treatment status in period t could plausibly affect the values of covariates in period t+1, we estimate a model that interacts covariates known to be pre-treatment in our case, from 1993 with year fixed effects. This approach allows us to incorporate information about relevant covariates, but restricts their year to year changes so they are the same across states. The results of this alternative specification are shown in Table?? and are nearly identical to our baseline results. We also estimate an entirely different panel model which includes one-period lagged 14 Specifically, we mark every state-year without implemented term limits with a zero, and give each stateyear with implemented term limits that state s term-limitedness score. This is analogous to interacting a binary indicator for implemented term limits with each state s term-limitedness score. 12

14 Table 2: Fixed Effects OLS Estimates: State Legislative Polarization and Term Limits Dependent variable: Legislative Polarization (1) (2) (3) (4) Term Limits (0.044) (0.044) (0.034) (0.033) Divided Gov (0.013) (0.012) (0.012) Democratic Governor (0.016) (0.015) (0.015) Leg. Professionalism (0.366) (0.293) (0.286) Absolute Difference Seat Share (0.001) (0.001) (0.001) Democratic Seat Share (0.001) (0.001) (0.001) ln(population) (0.315) (0.314) Per Capita Income (0.004) (0.004) GOP Presidential State Share (0.311) (0.307) Unemployment Rate (0.008) (0.008) Percent Foreign Born (0.009) State Gini Coefficient (0.240) State Fixed Effects Year Fixed Effects Projected R Observations Note: Entries are linear regression coefficients with standard errors clustered on states in parentheses. p<0.1; p<0.05; p<0.01 (all two-tailed tests). 13

15 dependent variables to account for unobserved state characteristics rather than state fixed effects. 15 The results of this model specification are shown in Table??. The reported effects are substantively smaller than those reported in Table?? yet statistically significant due to increased power from not estimating a large battery of fixed effects. Consistent with the interpretation offered by?, the results from this alternative model specification increase our confidence that term limits cause increases in legislative polarization under two possible data-generating processes. Taken together, our robustness checks provide strong and consistent evidence that our baseline results provide clear evidence of the effect of term limits on state legislative polarization. Because our outcome measure of polarization is a reflection of state legislators policy positions and roll call voting records, this suggests that we can soundly reject term limit advocates claims that term limits would reduce inter-party conflict; rather, our results suggest that they markedly increase it. Moreover, in the aggregate, our findings suggest that term limits are associated with declining levels of collective representation. To the extent most constituents are relatively more moderate than most elected officials (see, e.g.,?) and constituent preferences are relatively stable over short periods of time, the increasing movement of state legislators toward the ideological poles suggest that greater numbers of legislators vote in ways that are less representative of constituent preferences. Rather than enhancing democratic representation, as proponents of term limits argued they would, our evidence so far indicates that they may worsen it. 15 The lagged dependent variable model has a useful bracketing property with the fixed effect model from above: if the fixed effect model is the true model (as we believe it is), then the LDV model results will tend to underestimate the true effect of term limits; contrarily, if the LDV model is true, the fixed effects estimates of the effect of term limits will tend to be biased upwards (?, ). 14

16 Potential Moderators Our theoretical discussion identified several factors that link the presence of term limits with increased partisan polarization. In this section we consider how the effects of term limits may be moderated by at the aggregate level based on partisan and institutional characteristics and at the individual level based on attributes of state legislators. Aggregate-level Moderators Partisan asymmetries. Of substantial interest in the literature on legislative polarization is the question of whether or not polarization is symmetric ; that is, do Republicans move to the ideological right to the same degree that Democrats move to the left? Findings in the literature based on Congress are mixed, but a growing consensus indicates that increasing polarization since the 1980s has been driven by a Republican shift to the right, while Democrats have stayed more or less the same (see, e.g.,?). We investigate whether term limits had asymmetric effects across parties on legislative voting records. So doing, we examine whether the effects identified above were concentrated primarily among legislators identifying with a particular party to evaluate how term limits affect the quality of representation among partisan constituencies. To do so, we reestimate our full model (Column (4) from Table??) separately using the Democratic legislative median and the Republican legislative median as our outcome variables. We report the results in the left panel of Figure??, which displays the estimates of the treatment effects of term limits for both Democrats and Republicans. The results indicate that the adoption of legislative term limits led to a statistically significant shift to the ideological right among Republicans (0.062), while the treatment effect among Democrats (-0.018) indicated that term limits were accompanied by a shift in the liberal direction but is not statistically distinguishable from zero. Note that these results reflect the shift in the 15

17 party median for the parties, and therefore do not indicate that term limits had a negative effect on polarization among Democrats, but rather that the Democratic legislature median became more liberal. For ease of presentation, Figure?? reports the absolute value of this difference so that the effects of term limits can be directly compared across parties. Evaluating the magnitudes of the treatment effects, the estimates suggest that term limits effects were about three times greater among Republicans than among Democrats, though we emphasize that we cannot rule out the null hypothesis that both parties contributed equally to increased polarization from term limits. Full results from this analysis are presented in Table?? in the Appendix. Variation across chambers. We also distinguish the effects of term limits across state legislatures upper and lower chambers. As? report, term limits may have had asymmetric effects across chambers as legislators first learn the craft of legislating in the lower chamber before being termed out and pursuing office in the upper chamber. Consistent with this account, we find stronger evidence for the effects of term limits in states lower chambers. Figure?? shows that term limits had a positive, substantively large, and statistically significant effect on polarization in lower chambers; in upper chambers, we continue to find a positive relationship between term limits and polarization yet the estimates are smaller in magnitude and not statistically significant at conventional levels. We also note, however, that the coefficient estimates are themselves not statistically distinguishable across chambers. Nevertheless, our results provide suggestive evidence that the effects reported above are concentrated disproportionately in state lower chambers. Full model results are again presented in the Appendix in Table??. Legislative Professionalism. Third, we evaluate whether the effects of term limits varied based on states levels of legislative professionalism. Prior scholarship argues that states with more professionalized legislatures produce more comprehensive and complex policy 16

18 Coefficient on 'Term Limits' Coefficient on 'Term Limits' Democrats Party (a) Party Republicans House Chamber (b) Chamber Senate Figure 2: Effect of term limits across parties and chambers The plot on the left displays the estimated effects of term limits on the ideological positions of Republican and Democratic legislators. The plot on the right displays the estimated effects of terms limits on polarization for states upper and lower legislative chambers. and experience greater success when bargaining with the executive (?). Because highly professionalized legislatures contain legislators who generally possess greater institutional knowledge and legislative experience, the imposition of term limits may have more substantial effects on polarization in these states. By dramatically reducing legislators expertise, term limits in these contexts may provide greater opportunities for parties and other outside groups to wield increased influence on the legislative process. This argument suggests that term limits may have stronger effects on polarization in states with more professionalized legislatures. We evaluate legislative professionalism as a potential moderator and distinguish the effects of term limits in states whose professionalism levels in 1993 (the first year of our study) were above and below the median values. The results are shown graphically in Figure??. 16 The points displays the effects of term limits on polarization and the vertical lines show the 95% confidence intervals. The horizontal dashed line at zero represents the null hypothesis of no effect of term limits on polarization. We find fairly strong evidence that the effects 16 The table of coefficients is displayed in Table??. 17

19 0.2 Coefficient on 'Term Limits' Low Professionalism High Professionalism Party Figure 3: Legislative Professionalism and the Effect of Term Limits on Polarization of term limits are concentrated in states with more professionalized legislatures. As the left portion of the plot shows, the effect of term limits in states whose legislatures have low levels of professionalism is virtually zero (.006) with an associated p-value of.88. In states with high levels of professionalism, however, we find strong evidence that term limits increased polarization with an estimated effect of.106 (p =.022). These coefficients are themselves statistically distinguishable at p =.093. The results thus indicate that term limits contributed to polarization in state legislatures mostly in states whose legislatures were more professionalized prior to their introduction. Individual-level Moderators We also investigate whether the effects of term limits are moderated by legislator-specific characteristics, focusing on legislators leadership status and their electoral vulnerability. 18

20 Previous research suggests that the effects of term limits on polarization may occur through factors external to the legislature, such as changes in parties efforts to recruit candidates, or within the legislature, such as changes in the nature of party agenda control associated with the selection of more ideologically extreme leaders. We use an individual-level analysis where each observation is a legislator i serving in state j in year t and include the same covariates as Column (4) in Table??. We perform this analysis in two ways: first, estimating separate regressions by party, as in Figure?? above; second, using the absolute value of Shor-McCarty scores as the outcome variable. These results, which we present in the Appendix in Table??, have a slightly different interpretation than those above: while our baseline results use states as the units of analysis, the individual-level results implicitly weight states with large legislatures more highly, as we continue to use state and year fixed effects with the individual-level data. Party leaders and rank-and-file members. We study whether the effects of term limits differed between legislative leaders and rank-and-file legislators by creating an indicator for legislators who served as presiding officers (including President, President Pro Tempore, Speaker of the House, or Speaker of the House Tempore), majority leaders, minority leaders, and majority and minority whips. These data were obtained partially from? and? and supplemented with original data collection from the state Yellow Books to assemble a complete roster of legislative leaders between 1993 and We estimate a model where we interact this indicator for legislators serving in these positions with Term limits. The results provide no evidence that term limits had differential effects among leaders and other legislators. In no specification are the effects of term limits for leaders statistically distinguishable from rank-and-file members, and the coefficients on leaders are inconsistently signed and substantively small. While term limits may indeed empower parties in the electoral and legislative arenas (???), our findings do not support the hypothesis that legislative leaders, as providers 19

21 of partisan brand names (?), were disproportionately affected by their implementation. Electoral security. We also examine whether the effects of term limits vary based on whether legislators represent marginal or safe districts. Using data on state legislative elections, 17 we distinguished marginal districts as those in which the incumbent legislator won by 10 percentage points or fewer. 18 We then interacted this indicator with Term limits. While the direction of all results suggests that term limits had a slightly smaller effect on polarization among representatives from marginal districts than among their safe-seat counterparts, none of the interactions of marginal with Term limits is significant at p < Term limits therefore had relatively constant effects on the extremism of legislators voting records across electoral environments. Term Limits and Dyadic Representation One potential explanation for the increased polarization documented is that term limits have severed the electoral connection between legislators and their constituents, instead privileging party organizations, interest groups, and legislators own preferences (?). Because political elites are generally more ideologically extreme than non-elites, this could account for increasing party polarization in the legislature even in the absence of mass polarization (?). An alternative explanation, which we explore here, is that normatively desirable consequences of term limits at the dyadic level may contribute to collective polarization in the legislature. 17 These data were obtained from ICPSR study and supplemented with: Klarner, Carl, 2013, State Legislative Election Returns Data, , hdl:1902.1/ We acknowledge that incumbent vote shares are an imperfect measure of district competitiveness because they reflect the incumbent s personal vote as well as the constituency s partisan and ideological composition. In additional analyses (Table??), we characterized marginal districts based on whether the vote totals in state legislative districts in the preceding presidential election were 10 percentage points or closer using data from?, and distinguished marginal districts as those where the previous (legislative or presidential) election was decided by five points or fewer. Each of these alternative measurement strategies provides results that are extremely similar to those reported below. 19 The interaction is significant at the p <.10 level for Republicans. 20

22 In this section, we focus on two potential consequences of how term limits affect the relationship between constituents and their representatives: the partisan fit of a district s representation, measured by the alignment between a district s preferences and the party of the legislator who represents it, and responsiveness, or the rate at which district preferences are translated into legislative behavior. Our expectation is that term limits, by increasing the opportunity for parties to recruit candidates and obviating a number of alternative electoral cues, such as incumbency, seniority, or leadership status, will improve the alignment between district ideology and the party representing that district in the legislature. Because Republicans are the clear conservative party and Democrats the clear liberal party (in the present era), we expect that conservative districts will more regularly elect Republicans and liberal districts will more regularly elect Democrats under term limits. Moreover, because elites are more extreme than the mass public, we expect that the improved partisan fit will enhance responsiveness of legislative behavior to district ideology in term limited states, with a small shift in district ideology corresponding to a larger shift in representative voting behavior in term limited states than in non-term-limited states, on average. Consistent with the argument outlined above, we demonstrate that term limits improve the fit between district preferences and the party of their elected representatives. Measured by the relationship between constituent preferences and legislative voting behavior, we also find that legislators in term limited states exhibit greater responsiveness to their constituencies. We posit that these phenomena are related, as term limits result in the forced retirement of more moderate legislators who are succeeded by more extreme replacements. While we lack the research design to provide dispositive evidence for this series of events, we present evidence below that is consistent with these claims. At the very least, polarization induced by term limits does not simply result from representatives choosing to ignore their constituents, as some opponents of term limits have argued. Instead, term limits appear to strengthen the links between district preferences and partisan representation at the dyadic 21

23 level while contributing to decreased representation of state electorates at the aggregate level. To explore the consequences of term limits at the dyadic level, we require a measure of district preferences; we use estimates of state legislative district ideology from?. 20 Because the district ideology data begin in 2003, we restrict our attention to the 2000 Census districting cycle ( ). We also use an indicator for whether districts were represented by a Republican state legislator data based on data from?, which we also use, as above, to characterize state legislators voting records. Figure?? provides a descriptive look at the relationship between term limits and dyadic representation. The plot on the left shows smoothed curves to characterize the relationship between district ideology (on the x-axis) and whether a district is represented by a Republican legislator (y-axis). These curves are fit separately for states with and without term limits. Across both groups of states, more conservative districts are increasingly likely to be represented by Republican legislators. However, because the gray curve (states with term limits) is somewhat steeper than the black curve (states without term limits), the plot also suggests that the match is stronger between a district s preferences and the partisanship of its state legislator in states with term limits. In particular, relatively moderate districts are more likely to elect legislators whose party is a better fit with district ideology: right-ofcenter districts increasingly elect Republicans while left-of-center districts increasingly elect Democrats. Importantly, most state legislative districts are in this moderate range (see Figure??). The right plot in Figure?? shows the relationship between district preferences (x-axis) and legislative voting records (y-axis). While we emphasize that the estimates of district and legislative ideology are not directly comparable, they allow us to evaluate responsiveness by studying whether, for instance, legislators elected from more conservative districts compile more conservative voting records. And indeed this is what we find; both of the fitted curves

24 are positive with fairly steep slopes, indicating that representatives from more conservative districts exhibit greater conservatism in their voting records in state legislatures. We again find that the gray curve (term-limited states) is steeper than the black curve (states without term limits), particularly in the middle region of the plot which corresponds to relatively moderate constituencies. The relationships in this plot therefore suggest that legislators Term Limits 0.50 No Yes Legislator Ideology Legislator is Republican exhibit greater responsiveness to constituency preferences in states with term limits. Term Limits 0 No Yes District Ideology District Ideology (a) Partisan Representation and District Ideology (b) Term Limits and Legislative Responsiveness Figure 4: Term Limits and Dyadic Representation The plot on the left displays the estimated effects of term limits on the ideological positions of Republican and Democratic legislators. The plot on the right displays the estimated effects of terms limits on polarization for states upper and lower legislative chambers. We document these relationships in a more systematic way. To do so, we adjust our identification strategy because our estimates of district ideology begin in 2003, by which time most states with term limits had already adopted them. while most of our withinstate variation in the presence of term limits occurred prior to this.21 Instead, we use two alternative models, each of which requires additional, but somewhat different, assumptions. First, we use OLS regression which retains year fixed effects but omits state fixed effects, which are (nearly) perfectly collinear with the presence of term limits. Second, we estimate 21 Only three states implemented term limits after

25 a multi-level model with year fixed effects and random state intercepts. This model assumes a distribution for state-specific intercepts, adding an additional set of assumptions beyond a fixed effects model; this additional assumption, however, allows us to retain a within-state model interpretation with little within-state variation. While these alternative models rely on different parametric and identifying assumptions, they produce strikingly similar estimates. In both models, we include the full set of control variables used in Column (4) of Table??. 22 The results of this analysis are presented in Table?? and provide strong evidence in support of our argument and the patterns in Figure??. The first two columns show results when predicting whether a district is represented by a Republican legislator. The results indicate that a one-unit increase in district conservatism (refer again to Figure??) corresponds to about a twelve percent greater likelihood of electing a Republican in term-limited states than in states without term limits; these results are statistically significant at the p < 0.05 level in both the OLS and the mixed effects models. The two rightmost columns report how term limits alter the relationship between district ideology and legislator ideology, where the main coefficient of interest is the interaction term between Term Limits and Mean District Ideology. If term limits diminish responsiveness by attenuating the relationship between constituency preferences and legislative behavior, we would expect a negative coefficient on the interaction term. This is not what we find. Instead, and consistent with our account, the coefficients in both models are positive and indicate that term limits increase legislator responsiveness to district preferences. The results from these models therefore provide fairly strong evidence to suggest that term limits help to enhance ideological representation at the dyadic level. 22 For the analysis using legislator partisanship as the outcome, three of these covariates divided government, absolute difference in seat share, and Democratic seat share are clearly post-treatment as their values were only known after the legislator was elected. Accordingly, we use the two-year lagged values of these variables, so as to proxy for their true values with values that are (in almost all states) from the previous legislative session. The results are unchanged by using the post-treatment variables or four-year lags instead. 24