Ideological Donors, Contribution Limits, and the Polarization of State Legislatures

Ideological Donors, Contribution Limits, and the Polarization of State Legislatures Michael Barber This Draft: September 4, 2013 Abstract Can campaign contribution limits affect political polarization? In this paper I show that legislators respond to the ideological preferences of those who fund their campaigns individual contributors who are ideologically motivated when contributing, and political action committees, who tend to be less ideological in their giving. Furthermore, I show that limits on campaign contributions dramatically affect the way in which candidates fund their campaigns, yielding the prediction that limits on contributions may also affect legislator s ideology and voting behavior. To test this prediction, I use an original dataset of campaign contribution limits in the states over the last 16 years to estimate a fixed-effects model and find that higher individual contributions lead to more extreme legislators, while higher limits on contributions from PACs yield more ideologically moderate legislators. These results suggest that the connection between donors and recipients plays a role in the story of the polarization of American legislatures. Many thanks to Brandice Canes-Wrone, Nolan McCarty and Kosuke Imai for helpful feedback. Additionally, thank you to members of the Princeton University Graduate Student Seminar in American Politics and the members of the Princeton American Politics Dissertation Working Group. Ph.D. candidate, Department of Politics and Center for the Study of Democratic Politics, Princeton University, Princeton, New Jersey 08544. Email: barber@princeton.edu URL: http://michaeljaybarber.com

1 Introduction Scholars, pundits, and politicians have universally lamented the degree of polarization that is present in American politics today (McCarty et al., 2006; Levendusky, 2009; Bishop, 2009), and a variety of research shows that political polarization extends beyond the U.S. Congress to many of the legislatures of the American states (Masket, 2009; Shor and McCarty, 2011; Shor et al., 2010; McGhee et al., 2011). Furthermore, this research suggests that polarization leads to a variety of adverse outcomes. Polarization decreases legislative productivity (Binder, 1999), and is strongly related to income inequality in the United States (McCarty et al., 2006). Moreover, polarization of legislatures indicates a growing disconnect between public opinion and policy making, resulting in potentially biased representation in Congress (Bonica et al., 2013; Bartels, 2010). While scholars are quite united in decrying the negative effects of polarization, they are much less unified as to its causes. Previous investigations into the underlying causes of polarization have shown that several intuitive culprits gerrymandering (McCarty et al., 2009), primary election systems (McGhee et al., 2011), polarized voters (Ansolabehere et al., 2006; Bafumi and Herron, 2010), and changes in congressional rule-making (McCarty et al., 2006) are likely not as responsible as initially hypothesized. Lacking from these explanations is a story of how money influences the ideology of candidates who run for office, and the effect contributions have on the behavior of legislators once they are in office (but see McCarty et al. (2006), and LaRaja (2008) for notable exceptions). This paper provides such an explanation by tracing a chain of influence from ideological donors to politicians desire to raise money from these donors. I show that the two largest sources of campaign money, individual donors and political action committees (PACs), have dramatically different ideological motivations when giving: individual donors tend to be more ideological than PACs, who are more moderate on average. Given this fact, I demonstrate that candidates ideologies reflect the ideology of their primary contributors. That is, candidates who receive a greater proportion of their money from individuals (PACs) are also more ideologically extreme (moderate). I then turn to the various legal limits on contributions in the states that can either mute or amplify this connection between donors and recipients. I demonstrate that lowering limits on individual or PAC contributions leads to candidates receiving smaller average contributions from those groups, more contributors being constrained by the lower contribution limit, and candidates raising a smaller proportion of their funds from the group being limited. Finally, I show that as a result, lowering individual 1

limits moderates legislators in office while lowering PAC limits leads to more ideologically extreme representation. The results in this paper contribute to our understanding of the effect of contribution limits on legislative behavior. This is an emergent area of research for a variety of reasons. First, contribution limits at the federal level have been held constant for several decades. This lack of variation across the states and time limits the ability of scholars to identify the effects of contribution limits. It is partly for these reasons that there are no existing empirical studies investigating the effects of campaign finance on political polarization. However, several notable studies (Primo and Milyo, 2006; Hamm and Hogan, 2008; Stratmann et al., 2006) look to the states for variation in contribution limits to investigate the effect of contribution limits on electoral outcomes. I also focus on state legislatures in this paper. However, rather than simply looking at whether or not limits exist, I improve our ability to use contribution limits as an explanatory variable by creating an original dataset that records the actual dollar amount of each limit. The US states vary widely in their contribution limits (from $200 to more than $10,000 for example), and thus being able to account for the actual limit rather than simply noting the presence or absence of a limit is a dramatic improvement over the heretofore available data. I discuss this data in more detail in Section 6.1. I begin by presenting a theory of how contributors influence the ideology of candidates. I then outline the degree to which state legislators are ideologically polarized in Section 3. To measure polarization, I use two large datasets. The first is a massive collection of millions of roll call votes cast by over 11,000 state legislators in 49 states over the last 16 years. 1 Using these votes, I estimate the ideal point of each legislator. These scores are similar to NOMINATE scores which are frequently used to describe the ideological position of members of Congress (Poole and Rosenthal, 1997) While impressive for its comprehensiveness, measuring ideology using roll call votes precludes investigating the ideology of unsuccessful candidates who lose their elections and thus never enter office to cast a vote. To overcome this hurdle, I incorporate a second, equally large dataset that includes both incumbents and challengers. This second dataset is a collection of over 5.5 million contributions given to state legislators by individuals, political action committees, and party organizations. Using this dataset, I create ideological measures based on the networks 1 I am very grateful to Nolan McCarty and Boris Shor for granting me access to this incredibly comprehensive dataset that is the result of thousands of hours of tedious effort. See http:\americanlegislatures.com for the most recent publicly available version of these ideal point estimates. 2

of people and organizations from whom candidates receive money. With both of these ideological measures, I show that over time polarization in the states does not follow a uniform pattern. Many states see increasingly polarized candidates and legislators, while many other states either remain about the same over time or even see a decrease in the average distance between the parties. After establishing the degree of polarization present in state legislatures, I turn my attention to the ideology and donating behavior of individuals and PACs. In this paper I focus primarily on individual contributors and political action committees while setting aside contributions from parties, unions, corporations (outside of a union or corporate PAC), or candidate self-funding. As I will show later, contributions from these other sources make up a very small percentage of the average candidate s fundraising portfolio (usually less than 5%). 2 Thus, I choose to focus my attention in this paper to the influence and behavior of the two largest sources of money for state legislative candidates. In Section 4.1 I explore the degree to which individual donors differ from non-donating voters and political action committees in their ideological preferences. Using survey data from the Cooperative Congressional Election Survey (CCES) and American National Election Study (ANES), I show that over time individuals have been consistently more ideological than voters who do not contribute money to political candidates. Furthermore, using the dataset of campaign contributions, I show that individuals exhibit distinctly different donating behavior than PACs. While most PACs contribute money to candidates from both political parties, individuals are much more likely to concentrate their contributions on candidates from one of the two major parties. Furthermore, over the last 25 years, individual contributions have become a much larger share of the average candidate s fundraising portfolio. Next, I show that the ideology of individual donors relates directly to candidate s ideology and voting behavior. Section 5 shows that legislators who raise more money from individuals are also more extreme. Using the database of campaign contributions, for each election cycle I calculate the share of a candidate s money that comes from individuals, PACs, political parties and other sources. I then show that those candidates who favor individual donors in their campaign portfolios are more ideologically extreme. This is the case for legislators from both parties. Having established that there is a connection between individual (PAC) contributions and ide- 2 Just as an investment portfolio contains a mixture of stocks, bonds, mutual funds, and other money, candidates fundraise from a variety of sources. For the remainder of the paper, I will refer to a candidate s fundraising portfolio as a summary of the share of the money she raises from individuals, PACs, parties, self-funding, or other sources. 3

ologically extreme (moderate) candidates, I then investigate whether or not campaign contribution limits influence the contributing behavior of donors and the fundraising patterns of candidates. In Section 6.1 I show that contribution limits vary dramatically across the states and within any given state over time. This natural variation across and within states allows me to test the ability of limits to affect the behavior of contributors and legislators. Pairing the dataset of contributions with an original dataset of contribution limits in all 50 states from 1990 to the present, I show that contribution limits affect the way in which candidates raise money. Lowering the limits on contributions from individuals leads to candidates receiving smaller average donations from individuals, while also leading to more individuals bumping up against the maximum allowable contribution, thus constraining contributor behavior. Likewise, lower limits on individuals leads to candidates raising less money overall from individual donors. The same patterns hold for political action committees as well. Given that limits affect the composition of candidates portfolios, Section 6.3 completes the story by showing that contribution limits affect the behavior of legislators in office. I combine the contribution limits dataset with the ideal point estimates of state legislators to test this relationship. The results reveal that when contribution limits on individuals (who tend to be more ideological) are lowered, we observe less ideological voting behavior from representatives. Similarly, when limits on political action committees (who tend to be more moderate) are raised, we observe more moderate legislators. On average, changing from limited to unlimited individual contributions increases polarization by nearly one standard deviation on the ideal point scale. This effect is roughly the same for both Republicans and Democrats and represents nearly one third of the average distance between Republicans and Democrats. The results in this paper showing that legislators ideology closely aligns with contributors preferences complements many of the findings in other studies of representation and responsiveness and provides a first look at this relationship among state legislators. Underlying this theory is the foundational question of who legislators respond to when creating public policy. Without question, legislators are intensely interested in being reelected, and given this incentive, we expect them to be responsive to those people who can most credibly keep them in office. As Schattschneider (1942) stated, He who has the power to make nominations owns the party. Moreover, the power to determine the nomination (or election) is not necessarily distributed equally among the electorate (Fenno, 1978). For example, Gilens (2012) finds that public policy closely mirrors the preferences of the wealthy. Furthermore, when the preferences of the affluent diverge from the rest of the 4

population, legislators still more often respond to the preferences of the wealthy. This comports with the results of this paper since wealthy people are much more likely to contribute money to political candidates (Verba et al., 1996). In fact, Gilens suggests that one mechanism for the relationship between the preferences of the wealthy and public policy outcomes is through political contributions. These results suggest that legislators pay particular attention to the preferences of those who are instrumental in funding their campaigns. This work is also related to the literature on fundraising and the effects of campaign expenditures on legislative outcomes. Decades of theoretical and empirical research agree that having more money to spend in a campaign increases a candidate s probability of winning the election (Green and Krasno, 1988; Levitt, 1994; Gerber, 1998; Erikson and Palfrey, 2000; Ashworth, 2006) (Although, see Brown (2013); Jacobson (1978, 1990) for notable exceptions to this rule). The mechanisms underlying this relationship range from increasing voters knowledge of the candidate s policy positions (Austen-Smith, 1987; Herrnson and Patterson, 2000), to signaling to voters the candidate s quality through flattering advertising (Gerber et al., 2011), or decreasing voters assessments of the challenger through negative advertising (Ansolabehere and Iyengar, 2010). Additionally, legislators may simply frighten off quality challengers by amassing a sizable campaign war chest (Epstein and Zemsky, 1995; Goodliffe, 2001). Given the value of campaign money, it logically follows that candidates would invest a significant amount of time in fundraising (Francia and Herrnson, 2001; Herrnson, 1995), and even go so far as to adjust ideologically in order to increase the size of their campaign bank account (Stone and Simas, 2010). 2 Money and Ideology In this section I develop a theory of how legislators respond to the preferences of their most influential contributors, and how, given this relationship, limits on campaign contributions will affect the ideology of legislators in office. I begin by outlining several common assumptions that are necessary for the theory. In each case, I present evidence from previous work that justifies the use of the assumption. In areas where there is little existing work, I cite later sections of this paper where I empirically illustrate the assumptions I make when outlining the theory. I begin by making the common assumption that voters, donors, and legislators can be place on a unidimensional policy space and that voters select candidates based on minimizing the spatial distance between themselves and their candidate of choice (Downs, 1957). Furthermore, I assume that candidates compete on this policy space and derive utility from being elected (Banks and 5

Duggan, 2005; Hinich and Ordeshook, 1970). Thus, this standard Downsian model of political competition stipulates that ceteris paribus, those candidates that take ideological positions closer to the median voter will be more likely to win (Downs, 1957). Turning now to the ideology of contributors, I begin by assuming that individual contributors are on average more ideological than voters and political action committees. Furthermore, I assume that individual donors are more likely to be sincere in their giving patterns: that is, they choose to give to the candidate that is ideologically closest to them without considering the probability of that candidate winning the election. Existing research and new results in this paper show that this is indeed the case. For example, Francia et al. (2005) and Bafumi and Herron (2010) show in surveys of voters that donors have more ideologically extreme views than non-donating voters. Furthermore, Stone and Simas (2010) and Ensley (2009) show that not only are individual donors extreme, but they are also less concerned with the strategic consideration of a candidate s proximity to the median voter of the district. They find that as challengers move away from the median voter, they collect more campaign resources from individuals. These ideologically driven donors are contributing as a consumption activity rather than a strategic investment (Snyder, 1990). However, these results are from federal contribution data. I provide evidence in Section 4.1 showing that this is also the case for contributors to state legislators. Furthermore, previous research suggests that political action committees have different motives when contributing money to political candidates. For some time, theories and data have suggested that PACs are interested in gaining access to legislators in office in an effort to both craft legislation that is favorable to their interests, and simply to ensure that legislators are aware of their preferred policies (Hall and Wayman, 1990; Romer and Snyder, 1994; Ansolabehere et al., 2003). Thus, we observe that PACs tend to be non-ideological in their donation behavior (McCarty and Poole, 1998), often giving to candidates from both parties (Jacobson, 1978). Furthermore, if PACs are primarily interested in access to legislators in office, we can see how they would be motivated to support incumbents or ideologically moderate challengers who are closer to the median voter of their district, and are thus, according to Downsian logic, are more likely to win the election (Canes-Wrone et al., 2002). Only then can a PAC gain access to the lawmaking process (Snyder, 1990, 1993). Given these findings, PAC money acts as a substitute for ideological individual money. That is, when candidates raise money from PACs rather than individual donors, they no longer need to appease ideological demands that come with money from individual donors. This allows candidates to focus their attention to the concerns of the median voter. 6

Finally, I make several assumptions regarding candidates need for campaign money and the way in which contribution limits may impact their ability to raise money from various donors. While countless theories of representation begin with the Downsian result that the policy positions of candidates should match the preferences of the median voter in their district, ideology cannot be considered in a vacuum. Candidates must have the resources to increase their name recognition and inform voters of their positions in order to win elections. No amount of ideological positioning will help a candidate that is unknown to voters. This is one reason that fundraising is such a key component of the campaigning process and occupies such a large portion of representatives schedules (Herrnson, 1995). Thus, I assume that all things equal, candidates with more money are more likely to win an election. This assumption is supported by a vast literature showing that candidates with fundraising advantages are more likely to win elections (Jacobson, 1990; Gerber, 1998; Green and Krasno, 1988). This could be the results of advertising (Stratmann, 2009), mobilization efforts (Green and Gerber, 2008), or other methods of persuading voters to support the candidate s campaign (Gerber et al., 2011). Given these assumptions, one puzzle is why candidates would endorse ideological policies in pursuit of money from ideologues rather than simply remaining closer to the median voter and funding their campaigns with non-ideological PAC money. If the sheer amount of individual money available at the ideological poles is large enough, then such a move may be utility maximizing. In this case the electoral penalty of deviating from the median voter can be outweighed by the ability of a candidate to raise even more money from ideological individuals and thus increase his probability of winning. In Section 4.3 I show that individual money has outpaced PAC money to become the dominant source of campaign funding. Thus, in this model, candidates face a trade off between Downsian pressures to locate at the median voter and fundraising pressures from ideologically extreme donors (Baron, 1994; Moon, 2004; Grossman and Helpman, 1996). With a large enough pot of cash at the ideological fringes, the rational candidate should deviate from the median voter in pursuit of a sizable campaign war chest. However, the ability of candidates to raise funds from any one of these groups is affected by the amount of money they can legally raise from any one individual or PAC. For example, as a candidate raises money for her campaign, she may do so by combining money from individual donors, PACs, party groups, or funding her campaign from her own personal wealth. Finally, I assume that decreasing contribution limits, on individual contributions for example, increases the costs to the candidate of raising the same amount of money from individuals as she did under 7

higher limits. Under the more restrictive limits, she now needs to persuade more individual donors to contribute. Thus, we can interpret a lower limit as an increase in the marginal cost of fundraising from that particular group. Facing these new costs, the candidate will turn her attention to PACs rather than raising this money from individual donors since the relative cost of fundraising from PACs has decreased in comparison to fundraising from individuals. In the process of changing the sources of her campaign money, she has simultaneously changed the ideological composition of her contributors. Freed from the pull of ideologically extreme individual donors, she also moves towards the center under both Downsian pressures and incentives to raise money from more moderate political action groups. The converse is true for lower PAC contribution limits, which will induce candidates to increasingly raise money from individual contributors. In this case, candidates funded primarily by individual donors will be pulled from the median voter of their district towards the preferred ideology of individual contributors. Taken together, these assumptions lead to the following predictions. First, candidates who are more ideological should raise more of their money from individual contributors. Additionally, the opposite will also be the case more moderate candidates will raise more of their money from PACs. Second, as contribution limits are imposed, the ideology of candidates and thus those elected to the legislature will change as well. Lowering limits should lead candidates to decrease the portion of their fundraising that comes from the group facing these new restrictions. Furthermore, this shift in portfolio composition translates into an ideological shift among candidates as well. For example, I expect lower limits on individuals to lead to candidates moving away from individual contributions as well as more moderate candidates being elected to the legislature. Similarly, I expect lower limits on PACs to translate into more ideological candidates. I test these predictions in Sections 5 and 6.1. 3 Polarization in the States In this section I introduce the measure of ideology that I use to show polarization in state legislatures. I demonstrate that legislators in the states vary widely in their ideology and that polarization across states is equally wide ranging. Finally, I show that in states that have polarized over time, Republicans and Democrats are equally responsible for the divergence between the parties. The primary quality of interest I consider is the ideology of state legislators. Until recently, no data existed to measure the ideology of state legislators over time. However, Shor and McCarty (2011) have created a database of the roll call votes of state legislators from 1996 to 2009. Shor and 8

McCarty (2011) develop ideal points for nearly every state legislator that has served between 1996 and 2009 by using these millions of roll call votes to estimate a latent ideology of state legislators. 3 These ideal points are similar to NOMINATE scores, which are frequently used to describe the ideological position of members of Congress (Poole and Rosenthal, 1997). In addition to the roll call votes from 1996 to 2009, I supplement the dataset with roll call votes cast by state legislators in 49 states between 2010 and 2012. 4 These votes are recorded online by the individual state legislatures and are then collected and stored online by OpenStates.org. Pairing the roll call votes from Shor and McCarty with the roll calls cast from 2010 to 2012, my data include the estimated ideal points of more than 11,000 state legislators who served in office over a 16 year period. 5 Appendix A details the exact method by which I estimate the ideal points and how I ensure comparability across time and geographies. However, in short, I use a Bayesian model to estimate one ideal point for each legislator who served in office, and these estimates are comparable across states and time. Figure 1 shows the distribution of these ideologies across the states. Each box in the plot shows the interquartile range of the ideal points of legislators who belong to the Republican or Democratic party in that state. While Republicans are on average always more conservative than Democrats in each state, it is not always the case that Republians are more conservative than Democrats across the states. For example, the median Republican in New York is more liberal than the median Democrat in Oklahoma. On the other hand, the median Democrat in Kentucky is more conservative than the median Republican in Massachusetts. This conforms with research showing a high correlation between geography and ideology in the United States (Gelman, 2008), and also suggests that the relationship between partisanship and ideology is much stronger within states versus across states. Figure 1 also shows the variation in how polarized each state legislature is. While the median Republican and Democrat are quite close to one another in states such as Rhode Island, Hawaii, and West Virginia, there is a much larger distance between the parties in Arizona, Washington, and California, which is by far the most polarized state according to these data. 3 See? for a detailed discussion of their estimation process and Poole (2005) and Clinton et al. (2004) for a thorough discussion of models of latent variable estimation. 4 In this paper I only consider legislators who served in the lower chamber of the state legislature. Since Nebraska has a unicameral legislature, I do not consider this state in the subsequent analysis. I am currently collecting the roll call votes and contribution limits for the upper houses. 5 For the following states, I have more roll call votes extending further into the past: 1986 - SC, 1992 - UT, 1994 - CA, 1995 - AK, DE, MN, NC, OH, TX, WA, WI 9

AK AL AR AZ CA CO CT DE FL GA HI IA ID IL IN KS KY LA MA MD ME MI MN MO MS MT NC ND NH NJ NM NV NY OH OK OR PA RI SC SD TN TX UT VA VT WA WI WV WY -2-1 0 1 2 Figure 1: Distributions of ideologies in the States. 10

Republicans Democrats 2012 Party Median 0.0 0.5 1.0 1.5 2.0 2.5 RI CT NY HI DE CA CO AZ MI OH OK METN WA AL GA NM NC KS INAK WI SC FLID UTSD MD NH WV ILPA IA VT WY 2012 Party Median -2.0-1.5-1.0-0.5 0.0 0.5 NY CA PA INWV NC SD SC RI UT TN GAID HI KS AK NH VT IA IL WY MEOH DE MD CT WAMI NMFL WI AZ CO OK AL 0.0 0.5 1.0 1.5 2.0 2.5 1996 Party Median -2.0-1.5-1.0-0.5 0.0 0.5 1996 Party Median Figure 2: Difference Party Medians in 1996 and 2012 - Points above the line indicate parties becoming more conservative and points below the line indicate parties becoming more liberal. In the left panel, we see that Republicans in 2012 are more conservative than in 1996 in most states. In the right panel, we see that Democrats are more liberal in 2012 than in 1996 in most states. California and Colorado stand out as the two states that have polarized the most during this time. Additionally, there are few states that have seen a decrease in polarization. Figure 2 shows the change in party medians across time for each state. The left panel shows Republicans and the right panel shows Democrats in each state. In each plot, the x-axis shows the 1996 party median in each state and the y-axis shows the 2012 party median. If the party has not changed ideologically, the point will be near the 45 degree line. We see that this is the case for many of the states. However, there are many more states for which the median party member has moved substantially. Among Republicans, points above the 45 degree line show the party becoming more conservative over time while for Democrats, points below the 45 degree line illustrate the party becoming more liberal. There are very few states in which either party became more moderate during this time period, and when this is the case (Wyoming Republicans for example) the change is much smaller compared to the degree of polarization seen in other states. Similar to Figure 1, California stands out as a unique case. Republicans in California experienced the most change 11

between 1996 and 2012 and Democrats in the state are second only to Colorado in the amount of ideological movement observed during these 16 years. Another interesting feature of Figure 1 is the level to which both parties polarize during this time. This result contrasts with similar analyses at the federal level that suggest that the increasing distance between the Republican and Democratic parties in Congress is primarily due to movement to the right by Republicans (McCarty et al., 2006). To test this more thoroughly, Table 1 shows the results of a regression in which each legislator s ideal point is regressed on a variable indicating the year of the observation, a Republican dummy variable, the interaction of these two variables, and an indicator for the legislator s state. The year variable therefore indicates the average yearover-year within-state change in the average ideology of legislators from that party. More formally, the model is as follows: Ideal P oint it = α State[it] + β 1 Republican i + β 2 Y ear t Republican i + β 3 Y ear t + ε it (1) where the dependent variable of the model is the estimated ideal point of legislator i in year t. Dependent Variable: Legislator Ideal Point Republican Year Year * Republican -33.3 (1.205) -0.008 (0.0004) 0.017 (0.001) Fixed Effects State Observations 66,674 pr ( (Y ear + Y ear Republican) = 0 ) <.001 p<0.1; p<0.05; p<0.01 Table 1: The Average Within-State Year-Over-Year Polarization by Party - Republicans are moving in a conservative direction at a rate of -.008 +.017 =.009 while Democrats are moving in a liberal direction at a rate of -.008. While these differences are statistically different from one another, substantively they are nearly identical, with Democrats polarizing about 90 percent as much as Republicans in each year. Using this model, the estimated average yearly change in ideology among Democrats is β 3 while the change among Republicans is β 2 + β 3. To test whether the parties are moving apart at a different rate, I test the linear hypothesis β 2 + β 3 = 0. If it is the case that β 2 + β 3 = 0, then the average change in party ideology is statistically equivalent across the two major parties. 12

Table 1 shows these results. We see that Republicans are moving in a conservative direction at a rate of -.008 +.017 =.009 while Democrats are moving in a liberal direction at a rate of -.008. While these differences are statistically different from one another, substantively they are nearly identical, with Democrats polarizing at roughly 90 percent as fast as Republicans in each year. Taken together, these results suggest that on average, each year the parties move.017 (.009 +.008) further apart from one another. To put this number in context,.017 is slightly more than 1 percent of the average of the distances between the state party medians in 2012. Over a 10 year period this distance (.017 * 10 =.17) accumulates to nearly 15 percent of the average of the distances between the state party medians in 2012. This model is not meant to account for the variety of reasons for why legislators may be polarizing. It is however, meant to show that over time, at the state level, the parties have moved apart at roughly the same rate. This implicates both Republican and Democratic legislators in the rise of polarization in state legislatures. In Section 6.3 I investigate the degree to which limits on contributions can either slow or accelerate this rate of polarization in the states, and how this may vary by political party. 4 The Ideology of Contributors In this section, I describe the contribution data and present its results. I begin by discussing the ideology of donors by showing that individual donors are more ideological than non-donating voters and political action committees. I then show that individual donors make up an increasingly large share of candidates fundraising money. Finally, I show that candidates who raise more money from individual donors are more ideological than candidates who raise money from other sources. 4.1 Individual Donors and Voters Here I present evidence that individual contributors are ideological actors who hold political opinions that are more extreme than voters who do not contribute money. An important component of the theory outlined in Section 2 is that candidates face a trade-off between supporting policies that align with the median voter of their districts and espousing policies that are congruent with the ideology of more ideological individual contributors. Previous work has noted that contributors tend to be more ideological than non-contributing voters (Bafumi and Herron, 2010; Francia, 2003) and I present additional evidence here that comports with this finding using data from the American National Election Survey (ANES) and Cooperative Congressional Election Study (CCES). Furthermore, I extend these results by showing that contributors are more ideological than active partisans, which I define as voters who identify with a political party and engage in political activ- 13

ities outside of simply voting. The ANES is a biannual survey conducted during election years, and in every iteration of the survey from 1972 to 2008, respondents have been asked if they have contributed money to a political candidate. 6 Additionally, respondents are asked to place themselves on a 7-point ideological scale that ranges from Extremely Liberal to Extremely Conservative. 7 I identify survey respondents who report having contributed money to any political candidate. I then compare the self-reported ideology of contributors with non-contributors by partisan affiliation. The CCES is a similar biannual survey of Americans. However, the CCES sample contains more respondents in each iteration than the ANES (30,000 on average compared to 2,000 in the ANES), allowing for additional analysis that would not be possible with a smaller sample. However, the drawback of the CCES is that the first iteration of the survey is in 2006, which only allows for a few years of comparison. Figure 3 shows the results of the ANES analysis from 1972 to 2008 by plotting the difference in average ideology between donors and non-donors by party. In every year (except one) contributors are on average more ideological than non-contributors. These differences between donors and non-donors are statistically significant at the.05 level in 75% of the cases. While the results in Figure 3 suggest that donors are more ideological than non-donors the ANES survey does not allow for more fine-grained analysis because the number of voters sampled each year is quite small, and the proportion of donors in the sample is even smaller. For example, in 2008, 2,322 people were surveyed and 11 percent of respondents reported contributing money to any candidate or political organization. In the entire sample of respondents, 10 percent reported contributing. However, the large sample size of the CCES survey provides an opportunity to investigate the ideology of donors at a more detailed level (CCES, 2012). Pooling the data from the CCES in 2006, 2008, 2010, and 2012, I conduct a similar analysis as before. 8 Figure 4 shows the results of the CCES analysis. I divide the data according to respondent s state to show that within states donors are more ideological than non-donors and that the relationship is not simply the result of donors residing in more liberal or conservative parts of the country. The red points in Figure 4 show the difference in the average ideologies of Republican donors and non-donors, and the blue points show the same differences for Democratic donors and non-donors. We see that in 6 with the exception of 2006 7 The question comes from the ANES time series data and is a 7-point scale with the following response options: Extremely Liberal, Liberal, Slightly Liberal, Moderate/Middle of the Road, Slightly Conservative, Conservative, Extremely Conservative 8 However, in the CCES the self-reported ideology question asks respondents to identify their ideology on a 5-point scale rather than the 7-point scale used by the ANES. Total N from the pooled CCES data is 110,220. 14

Difference in Average Ideology -1.0-0.5 0.0 0.5 1.0 Republicans Democrats 1970 1980 1990 2000 2010 Figure 3: Donor and Non Donor Ideology Across Time - The vertical axis shows the difference in average ideology taken from a 7-point ideological scale used in the ANES survey from 1972 to 2012. We see that in every year (except one), those who contribute money are more ideological than those who do not contribute money. The vertical lines show the 95% confidence intervals on the differences in ideologies. Total N = 29,959. nearly every state contributors are more ideological than non-contributors. In some states (such as Republicans in Montana or Vermont) the difference is quite small while in other cases (such as Democrats in North Dakota or West Virginia) the difference is much larger. The vertical lines on the plot display the 95% confidence intervals for each point. In every case (except Arkansas Republicans) contributors are more ideological than non-contributors, and in 87% of the cases, the different is statistically different from zero at the.05 level. These results support the ANES data described earlier and suggest that contributors are more ideological than non-contributors and have been so for the past several decades. To ensure that donors truly are more ideological, I not only account for the partisanship of the voter, but I also account for the level of involvement voters have in political campaigns. It could be the case that those who are actively involved in politics are more ideological than those who are not and that being a contributor is not tied to more ideological opinions once we account for political activity. However, this is not the case. Table 2 shows that even among those who are very involved in politics, those who contribute money are more ideological. To conduct this analysis I consider only the ANES respondents who are in the top quartile of involvement and regress their ideology on a variable indicating whether or not they contributed money in a political campaign. 15

Difference in Average Ideology -1-0.75-0.5-0.25 0 0.25 0.5 0.75 1 AK AL AR AK AL AR AZ CA AZ CO CT CA DE CO CT DE FL GA FL GA HI HI IA IA ID IL IN KSKYLAMA ID IL IN KS KY Republicans LA MD ME MIMN MOMSMT NC ND NE MS MD MA MN MO MEMI Democrats MT NC ND NE NH NH NJ NM NV NY OH OK OR PA NJ NM NVNY OH OK OR PA RI RI SC SD TNTX SC SD TN UT VA VA TXUT VT VT WAWI WA WI WV WY WV WY Figure 4: Difference in Donor and Non-Donor Ideology by State - The vertical axis shows the difference in average ideological values taken from a 5-point ideological scale used in the CCES survey in 2006, 2008, 2010, and 2012. We see that within each state, those who contribute money are in nearly every case more ideological than those who do not contribute money. The vertical lines show the 95% confidence interval on the differences in ideologies. Total N = 110,220. I also account for the year and state of the respondent as well as indicators for other types of political activity such as attending a rally or volunteering for a political campaign. We see that among Republicans and Democrats, politically active donors are more ideological than those who are equally politically active, yet do not contribute money. 4.2 Individual Donors and PACs In this section I show that individual donors are also distinct in their donating behavior from political action committees. Figure 5 illustrates this by simply looking at the percent of individual donors and PACs who gave to candidates from both major parties, weighted by contribution amount. We see immediately that PACs are much more likely to contribute to legislators from both parties. This aligns with a wealth of previous research suggesting that the motives behind donations from political action committees are much less aligned along partisan lines (Hansen, 1991; Romer and Snyder, 1994; Ansolabehere et al., 2002). Instead, scholars suggest that PACs are much more interested in contributing to legislators on either side of the aisle with connections to policies they view as important. For example, these connections include committee assignments, 16

Dependent Variable: Respondent Ideology Republicans Democrats Donor.17 -.18 (.06) (.07) Attend Event -.05 -.05 (.06) (.07) Work for Campaign.03 -.02 (.07) (.07) Year.02 -.01 (.003) (.003) Fixed Effects State State Observations 1,531 1,726 p<0.1; p<0.05; p<0.01 Table 2: Donor Ideology Among Politically Active Voters - These models consider only respondents to the ANES survey who are in the top quartile of political activity. In other words, among the politically active, donors are still more ideological than non-donors. This pattern holds for both parties. irrespective of the legislator s party (Romer and Snyder, 1994; Kroszner and Stratmann, 1998). This is in contrast to evidence that individuals care primarily about legislators ideology and are not concerned with committee assignment nor other ancillary traits. If it is the case that individuals are less strategic and more ideological when contributing, we would expect two empirical patterns to emerge. The first is that the proportion of money individuals contribute to challengers should be higher than the proportion of money PACs contribute to challengers. Since PACs seek access to legislators in office, the only way to gain such access is through candidates who win the election and go on to be seated in the legislature. Given the substantial incumbency advantage in legislative elections (Ansolabehere et al., 2000), we would expect PACs to first target incumbents rather than challengers who are much more likely to lose. In contrast, individuals should place less importance on the incumbency status of candidates if they are primarily motivated by the candidate s ideology. We see evidence of the first relationship in Figure 6 which plots the percentage of money spent by PACs and individuals on challengers. To determine the incumbency status of each state legislative candidate in these years, I match the data in the state contribution dataset to data collected by Klarner et al. (2013). This data contains an exhaustive list of every candidate for state legislative office between 1970 and 2010 and 17

Percent 0 20 40 60 80 100 PACs Individuals 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 Election Cycle Figure 5: Percent of Donors Giving to Candidates from Both Parties - Weighted by donation amount. In every election cycle, PACs are much more likely to give to candidates from both parties. This shows that PACs and Individuals exhibit very different strategies when contributing and suggests that the two groups have different considerations when deciding who to contribute to. includes a variable indicating if the candidate is an incumbent or challenger. After determining the incumbency status of each candidate, I calculate the percent of individual money contributed to incumbents and challengers for each election cycle. I then do the same for PACs. As predicted, in every election cycle a greater percentage of individual money than PAC money flows to challengers. The second empirical pattern we expect to see is that individuals prefer giving to more ideologically extreme candidates than PACs. If individuals are primarily motivated by ideology when giving, and, as shown in Section 4.1, individuals tend to be more extreme, we would expect these two patterns to translate into few individual donors contributing money to ideologically moderate candidates. On the other hand, if PACs are motivated primarily by accessing legislators in office, and given the previous literature that suggests more extreme candidates are more likely to suffer electorally (Canes-Wrone et al., 2002), we would expect PACs to favor contributing to more moderate or centrist candidates. To assess this prediction I create a measure of ideology for each contributor to state legislative candidates from 1996 through the 2012 election cycle. 9 Using the 9 I borrow this measure from (McCarty et al., 2006), who calculate a similar statistic for federal contributors using 18

Percent 0 20 40 60 80 Individuals PACs 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Election Cycle Figure 6: Percent of Contributions Going to Challengers - In every election cycle a greater portion of individual money than PAC money is spent on challengers. This suggests that PACs are more interested in assisting incumbents than challengers in their campaigns and lends evidence to the idea the PACs are primarily interested in gaining access to legislators in office rather than promoting a particular ideology. legislator ideal points described in Section 3 I estimate the ideology of contributors by taking the average of the ideal points of those they contribute to, weighted by the amount of each contribution. Equation 2 shows this money weighted ideal point more formally. mwip = N i=1 d i x i N i=1 d i (2) In Equation 2 the contributor s money weighted ideal point (mwip) is calculated from N contributions of amounts {d 1,..., d n } to candidates {x 1,..., x n }. The left hand panel of Figure 7 shows the distribution of donors money weighted ideal points. I plot the distribution of individuals (in green with solid border) separately from PACs (orange with dashed border) to illustrate the difference in ideologies between the two different groups. Doing so reveals a distinct pattern. Individual ideal points are distributed more uniformly across the entire range of ideologies with two modes over more conservative and liberal scores. The nadir of the distribution falls over the NOMINATE scores. 19

most moderate values. On the other hand the distribution of PAC ideologies is unimodal and centered around more moderate to slightly conservative scores. This provides evidence that PACs favor more moderate candidates while individuals prefer more partisan or ideological candidates. Density 0.0 0.2 0.4 0.6 0.8 1.0 PACs Individuals Density 0.0 0.2 0.4 0.6 0.8 1.0 PACs Individuals -2-1 0 1 2 Money Weighted Ideology Score -2-1 0 1 2 Iterated Money Weighted Ideology Score Figure 7: Individual and PAC Money Weighted Ideology Distributions - Figure on the left shows the average ideology of donors based on the money weighted average of donations only to legislators who have ideal points estimated from voting records. The right panel shows ideologies estimated through an iterated money weighted average process, which includes all donations, not just candidates with voting records. In both cases, PACs display a moderate, unimodal distribution, while individuals ideologies are bimodal around the ideological extremes. The money weighted ideal point provides more information than simply looking at the incidence of bipartisan giving and allows us to see that there is significant variation in the preferences of individuals and PACs. However, the major limitation of this measure is that it ignores the contributions made by donors to candidates without voting records (which are used to estimate ideal points). Thus the left panel of Figure 7 omits a significant amount of information that could help us understand the differences between individual and PAC contributions. This is especially true since individuals give more often to challengers who do not have ideal points estimated through roll call voting. To address this concern, I estimate a second measure of contributor ideology, the iterated money weighted ideal point (imwip), which was developed by Adam Bonica (2010). Following Bonica, I construct the measure by performing the following steps: 1. Assign each candidate a value of 1 if Republican, -1 if a Democrat, or 0 otherwise. 2. For each contributor, calculate the average of all recipient scores, weighted by the contribution 20