PUBLICLY FUNDED ELECTORAL SYSTEMS EFFECT ON STATE LEGISLATURES' INCUMBENCY TURNOVER

Transcription

1 PUBLICLY FUNDED ELECTORAL SYSTEMS EFFECT ON STATE LEGISLATURES' INCUMBENCY TURNOVER A Thesis submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy in Public Policy By Laura Hatalsky, B.A. Washington, DC April 12, 2011

2 Copyright 2011 by Laura Hatalsky All Rights Reserved ii

3 PUBLICLY FUNDED ELECTORAL SYSTEMS EFFECT ON STATE LEGISLATURES' INCUMBENCY TURNOVER Laura R. Hatalsky, B.A. Thesis Advisor: Ward Kay, PhD. ABSTRACT This analysis focuses on the ties between publicly funded election systems and the reelection chances of state legislative incumbents. Proponents of these systems often argue that these systems can minimize the unsavory political influence of special interest money and wealth; make elections more competitive; and encourage challengers to run elements that may be detrimental to incumbent lawmakers. This analysis finds that pursuing election reform and winning reelection need not be diametrically opposed. This research uses data compiled by the Interuniversity Consortium for Political and Social Research on state legislative election returns and controls for time, geographical region, Presidential election years, population per member in state legislative bodies, the number of opponents, political party, district type (single member district or otherwise), upper chamber seats, and professionalization. By examining election outcomes across the nation from 1990 through 2008, this study concludes that publicly funded election systems produce narrower margins of victory and increased challenger emergence. However, no statistical significance can be tied between publicly funded election systems and increased incumbency loss. Therefore, current lawmakers can enact these systems without resigning themselves to defeat. iii

4 Writing a graduate thesis is much like running a marathon or withstanding a ten round fight it cannot be completed and endured without much love and support. True credit for this success must be extended where it is due. Carl Klarner at Indiana State University showed unparalleled generosity when he agreed to pre-release ground-breaking data through My analysis might ve been a tireless rehashing of past studies without this gift. My advisor, Ward, for infinite patience and guidance. I would ve wandered endlessly in the confusing haze of econometrics without him. My thesis colleagues, Heather Vaughan and Amy Cohen, who made the year-long bout such a delight. My fiancée, better half, and intellectual superior, Lanae. My tenure at Georgetown would ve lasted little more than a week if not for her compassion and pep talks. And the rest of my family that tolerated my explanations of this project, despite having no clue what I was ever talking about. Many thanks, Laura R. Hatalsky iv

5 TABLE OF CONTENTS Introduction and Motivation... 1 Defining Clean Election and Hybrid Systems... 3 Literature... 5 Data Evaluating the Goodness of Fit for Logit Models Results Implications Appendix I: Summary of Clean Election System Operations in Arizona, Maine, and Connecticut Appendix II: Defining Professionalization Appendix III: Population per Member of State Legislative Bodies References v

6 LIST OF FIGURES Figure 1: All Incumbents Election Outcomes ( ) Figure 2: Election System Type Figure 3: Incumbent Loss Rate Over Time ( ) Figure 4: Incumbent Loss Rate by Political Party Figure 5: On Year and Off Year Loss Rates Figure 6: Loss Rate by Region Figure 7: Incumbents Loss Rate by Election Type Figure 8: Fitted Probabilities for Model Figure 9: Fitted Probabilities for Model LIST OF TABLES Table 1: Election Systems Classification Summary... 5 Table 2: Logit and OLS Results vi

7 INTRODUCTION AND MOTIVATION Publicly financed elections are systems whereby taxpayer dollars are directed to public office candidates to run their campaigns. Generally, candidates must first demonstrate their viability in an election before he/she becomes eligible for public funds. Often, this requires a candidate to collect a number of donations in small denominations from constituents throughout the electoral district. In purely publicly funded systems, the candidates may only spend the amount of the public grant while additional private fundraising and spending by the candidate is disallowed. Alternatively, hybrid systems exist where candidates may qualify for public funds, but the grant does not finance the entire operations of the campaign. Therefore, traditional fundraising is pursued in addition to the public grant. Lastly, traditional systems allow candidates to seek contributions from citizens and a variety of organizations without any taxpayer-funded support. While the amount an individual or organization can contribute to a campaign is typically capped (e.g. federal contribution limits are set at $2,400 an election per candidate a ), this is the system that may most obviously lend itself to political quid-pro-quo for campaign contributions. States have adopted a variety of these models to operate elections within their boundaries, making state legislature candidates an ideal population for study to discern the effects of publicly financed electoral arrangements. Proponents of moving towards publicly funded elections such as Common Cause, the Clean Elections Institute, or Public Campaign believe that these schemes have two relevant benefits. Firstly, by granting candidates regardless of political affiliation, wealth, incumbency, or celebrity a basic subsistence amount of funds to run their campaigns, the system levels the a Federal Election Commission. Contribution Limits : 1

8 playing field where electoral outcomes are determined. These potentially fairer elections balance the resources available to campaigns and may increase competitiveness (Francia & Herrnson, 2003). Secondly, proponents believe that politicians, upon reaching elected office, are accountable to and influenced by the individuals or groups that financed their electoral success. Under a traditional system, they believe elected officials will be make governing decisions based on improving the welfare of their campaign s funders. Alternatively, a publicly financed system removes special interests from the electoral success equation leaving politicians free to work towards maximizing public welfare (Francia & Herrnson, 2003). Elected officials that sympathize with the clean elections cause, however, may find themselves balancing their own professional future against electoral reform. A self-interested politician seeking reelection does not want to make the political lives of potential opponents any easier. Incumbents, when compared to their challengers, will likely have a higher public profile, possess more political clout and influence, and enjoy a fundraising network invested in their success. Enacting legislation that would implement publicly funded elections inherently gives would-be adversaries resources. This transfer of power puts the reelection chances at odds with election reform. However, pursuing election reform and winning reelection need not be diametrically opposed. Just because elections are more competitive does not necessarily mean that challengers will win more often. Margins of victory may narrow; challengers may run more often; an incumbent may need to canvass more, talk to constituents more, and campaign harder but the loss of a seat is another matter entirely. Most recent extant literature has observed increased overall competitiveness from these systems (GAO 2010, Malhotra 2008) while examination of the sole 2

9 effects on incumbents electoral road in publicly financed systems is unaddressed. If in fact publicly financed election systems do not increase incumbency loss, politicians can enact these systems without resigning themselves to defeat. DEFINING CLEAN ELECTION AND HYBRID SYSTEMS Using data compiled by the National Conference of State Legislatures, each electoral year in each state is classified into one of three categories: (1) clean election, (2) hybrid system, or (3) traditional election. Maine was the first state in the nation to take steps toward pure public financing availability in 1996, when a ballot initiative passed by a 56 percent to 44 percent margin. The system, first used in state legislature races in the 2000 election, provides candidates with the option to publicly finance their campaigns. Arizona mimicked Maine s steps toward pure public financing in 1998, when voters approved the Citizens Clean Elections Act by 51 percent to 49 percent. The system, first used in state legislature races in the 2000 election, provides candidates with an opt-in system in which they receive public funding to run their campaigns. Lastly, in December 2005, the Connecticut state legislature passed legislation enacting public financing. The system was first used by their General Assembly candidates in Therefore, in this analysis, all Maine and Arizona elections occurring in 2000 or after are classified as clean elections as well as elections in Connecticut in A summary of the monetary amounts involved in each programmatic phase from qualification to primary and general elections is outlined in appendix I. 3

10 Hybrid systems will be those that provide partial public financing to state legislative candidates (not to state political parties). The Hawaii Election Fund was created in This data only includes years after 1990; therefore, across all years in the sample, Hawaii is coded as a hybrid system. Minnesota's public financing system was enacted in the 1970s and significantly reformed in The data set only includes years after Therefore, across all years in the sample, Minnesota is also coded as a hybrid system. Nebraska passed a public financing law in In a phone conversation with the Nebraska Accountability and Disclosure Commission, the official said that while the first election cycle in which funds were available was 1996, they did not disperse any funds until b To be consistent with other coding (which accounts for presence of the statute and availability of funds, not actual usage), elections in 1996 and after in Nebraska are coded as a hybrid system. Lastly, Wisconsin enacted its partial public financing system in the 1970s. As this analysis begins in 1990, Wisconsin is coded as a hybrid system across all years in the sample. For the analysis, all other races are classified as traditional election systems. This, admittedly, glosses over New Jersey s pilot program that offered public financing to candidates seeking election in certain districts in the 2005 and 2007 elections. Since this was only a pilot program in a few districts, its effect may be different than where public financing is enacted statewide. Therefore, New Jersey pilot races are treated as traditional elections. b Conversation was on January 18, 2011 with the Nebraska Accountability and Disclosure Commission Executive Director, Frank Daley. 4

11 Table 1: Election Systems Classification Summary Clean Election Arizona: 2000 election forward Maine: 2000 election forward Connecticut: 2008 election forward Hawaii: across all observations in Hybrid Election data Minnesota: across all observations in data Nebraska: 1996 election forward Wisconsin: across all observations in data Traditional Election All other elections Source: National Conference of State Legislatures LITERATURE I think the purpose of this law is to prevent corruption That s what the purpose of all public financing systems are. Supreme Court Justice Elena Kagan in reference to Arizona s Clean Election System c Campaign finance law begins from the notion that human beings are not angels. Its first motivational premise is that without law, some politicians will veer from the public good to serve their own interests. Kenneth Mayer from The Day After Reform (pp. 2) Much study has been dedicated toward the influential factors determining lawmakers governing behavior. Foremost on our minds of publicly funded elections advocates is the untoward or unsavory influence of money, the wealthy, political action committees, or other organized interest groups. Exploring campaign contributions and their influence on voting behavior more generally, Fellowes and Wolf (2004) tease out the nuanced relationship between business campaign contributions and the of Representatives. As they review extant literature on the political influence of money, most studies have found that business contributions perhaps have no policy influence (Ansolabehere, Figueiredo, and Snyder Jr. 2002) and do little c The Washington Post, Supreme Court raises questions about Arizona law s extra cash for publicly funded candidates. Associated Press. March 28,

12 more than give donors access to the political process (Chin, Bond, and Geva 2000). The influence of business contributions, up until their study, had been limited to special circumstances (pp 316) including low visibility scenarios (Sorauf 1992; Morton and Cameron 1992) and votes on which lobbies are extraordinarily active (Baumgartner and Leech 1998). Fellowes and Wolf also recognize that lawmakers are balancing countervailing interests by not wanting to lose favor with constituents by appearing easily bought and sold and simultaneously needing to build respectable campaign war chests. Therefore, Fellowes and Wolf explain a form of tactical rationality whereby Representatives will pursue a rational strategy of rewarding business contributions with benefits that impose minimal electoral costs (pp. 317). On direct expenditure bills (where the appearance of favoritism most easily lends itself to accusations of quid pro quo relationships with business interests), Fellowes and Wolf hypothesized lawmakers voting behavior will not be systemically related to business campaign contributions. However, in instances where the potential electoral downfall is minimized in less-obvious scenarios found in tax policy or regulatory votes lawmakers who are heavily reliant on business campaign contributions will vote in the industry s best interest. By examining voting behavior in the 105 th session of the of Representatives and campaign contributions collected by the Federal Election Commission (FEC) as well as the Center for Responsive Politics (which collects information specifically on individuals with business-sympathetic interests), their theories held muster. Their regressions also controlled for politicians ideology, an incumbent s margin of victory in the previous election, the length of the lawmakers tenure, and regional dummies. 6

13 The Fellowes and Wolf tactical rationality approach reflects a realistic and nuanced relationship between special interests, contributions, and lawmakers. However, state lawmakers (the population of interest here) may not mimic the behavior of their federal counterparts. Further, the political influence of money may not be the same across different donor industries. While the business lobby may be particularly effective at maximizing the efficacy of their campaign contributions, Fellowes and Wolf do not speak to labor contributions, environmental industries, or other organized lobbies. Speaking more directly to the behavior of state legislators, Monardi and Glantz (1998) discerns patterns of tobacco industry campaign contributions and legislative behavior in California, Colorado, Ohio, Pennsylvania, New Jersey, and Washington. By running a 2-stage least squares method of estimating simultaneous models, Monardi and Glantz specified one equation to account for legislative behavior to attract campaign contributions (what they refer to as the electoral influence model) and another equation to account for campaign contributions to affect political behavior (the legislative influence model). After controlling for partisanship, majority party status, and leadership status of individual members, significant and positive correlations were seen between tobacco industry contributions and pro-industry legislative behavior in all states except Ohio. The authors also had access to additional information regarding the California state legislators and were able to control for constituent attitudes on tobacco and lawmakers personal characteristics. The statistical significance and positive correlation remained strong. While not terribly surprising as this issue may fall under the special circumstances mentioned by Fellowes and Wolf (2004) the work was supported by 7

14 grants from the American Cancer Society and the National Cancer Institute and should be viewed with this in mind. Dow, Endersby, and Menifield (1998) address the scenario from a different perspective and study whether campaign contributions from particular industries are focused on those state lawmakers with the most institutional influence over their cause by sitting on the relevant committee. If this correlation does exist, we could reasonably believe that industries attempt to influence relevant lawmakers with strategic contributions, while the resulting behavior from lawmakers may still be up for debate. The study examines the frequency of contributions made by dozens of industries ranging from agriculture and real estate to entertainment and finance to California Assembly incumbents running for reelection in 1984, 1986, and Using a chisquared test, the authors found that the vast majority of the industries exhibited donation behavior that is skewed by the committee assignment of Assembly members. This study confirms the intent of organized lobbies to curry influence but does not address lawmaker behavior. Anecdotally, the Arizona s clean elections system was put into place after years of corruption and questionable behavior. This included the then- Judiciary Committee Chairman being caught on hidden videotape accepting $55,000 in cash explicitly in exchange for his vote. d Governor Evan Mecham s tenure was riddled with questionable deeds, including accusations of concealing a $350,000 donation from a land developer. Other incidents relating to inappropriate use of government funds surfaced, and the Governor was eventually impeached. e Arizona s d New York Times, Keep Arizona Elections Clean. December 20, 2010 e Sullivan, Patricia. Evan Mecham, 83; Was Removed as Arizona Governor. The Washington Post. February 23,

15 Citizen Clean Elections Commission, which oversees the public financing system, states that the system allows [candidates] to forgo special interest groups. f Additionally, Connecticut s election governing body, the State Elections Enforcement Commission (SEEC), says their program was designed to limit the role of private money in the State of Connecticut s political process. g The Government Accountability Office was instructed to analyze the clean election programs in Arizona and Maine by section 310 of P.L (McCain-Feingold Bipartisan Campaign Reform Act). The 2003 report generally concludes that post-reform elections saw increased candidate participation in public financing; however, the programs effect on competitiveness is minimal, The public financing programs were expected to make elections more competitive, but our analyses were inconclusive (pp 4). Similar conclusions were reached regarding interest group influence in government. Since the 2003 release of GAO s report, information continues to become available on electoral outcomes in Arizona and Maine, increasing opportunities for research on these purely publicly funded electoral systems. Soon after GAO s seminal report, Zuckerman (2003) criticized it for falsely attributing some positive outcomes (such as more incumbents facing challengers and smaller margins of victory) to public financing that truly resulted from term limits. Mayer, Werner, and Williams (2006) found significant methodological shortcomings in GAO s 2003 report namely, they failed to count incumbents that lost in primary elections. Mayer, Werner, and Williams (2006) research examines elections for state seats in Arizona, f Citizens Clean Election Commission website: g State Elections Enforcement Commission website: 9

16 Hawaii, Maine, Minnesota, and Wisconsin from looking for indicators of contestedness, competitiveness, and reelection rates. They reached four conclusions: (1) public funding increased the pool of candidates in elections, (2) public funding has increased the likelihood of an incumbent having a competitive race, (3) reelection rates for incumbents in Maine and Arizona dropped significantly in 2002 and 2004 in comparison to rates before public financed was enacted, and (4) meagerly funded systems are ineffective. The weakness of this article, however, is that they reached these conclusions without multivariate analysis. As late as 2008, Malhotra notes the absence of multivariate regression analysis in extant literature. His analysis distills electoral competition down to two basic components: (1) incumbency, and (2) challenger quality. Through this lens, public financing could serve to increase competition by getting resources to challengers and/or by encouraging high quality candidates to run. Malhotra uses margin of victory and the inverse HHI (Herfindahl-Hirschman Index) to measure competition on Arizona ( ) and Maine ( ) State races. While he found that the mere presence of the statute made no significant difference in competition, instances in which a non-incumbent actually chose to enroll in public financing did exhibit statistically significant more competition in both states. In races where challengers participated in public financing and an incumbent was running in the election, the margin of victory lessened by 13 percentage points in Maine and 47 percentage points in Arizona. Malhotra ran different models on the data for each state, and found statistically significant differences in the coefficients using a Hausman and Chow-type test. He credits the bigger effects in Arizona to the increased professionalism of their legislature (more valuable seats are more influenced by campaign expenditures) and the larger size of Arizona districts when 10

17 compared to Maine. Campaigning to reach a larger number of voters relies more heavily on money, whereas smaller districts can be more easily influenced by retail politics. As Malhotra points out, Holbrook and Tidmarch (1991), Gelman and King (1990), King (1991), Cox and Morgenstern (1993, 1995), and Carey, Niemi, and Powell (2000) all find that incumbency advantage is boosted by professionalism in state legislatures. In addition to professionalization, Carey, Niemi, and Powell (2000) also point to multi-member districts (MMD) and district size as statistically significant sources of incumbency disadvantage. Examining standardized coefficients, MMDs have the largest impact on decreasing reelection likelihood for an incumbent. Further, as district size grows, it undercuts the amount of time each elected official can dedicate to the needs of individual constituents, lessening incumbency advantage. Most recently, the Government Accountability Office updated its 2003 report on Arizona and Maine in May 2010 to include analysis on races from 1996 to They also collected data in four states for basis of comparison (Colorado, Connecticut, Montana, and South Dakota), and found decidedly mixed results. They found quite substantial increases in participation rates of candidates in public financing in clean election states. The percentage of general election races that included at least one participating candidate jumped from 47% in 2000 to 98% in 2008 in Maine. Similarly, Arizona jumped from 53% in 2000 to 82% in Meanwhile, the winner s margin of victory increased in both states over this time period. Improving on their previous analysis, they used multivariate statistical models including fixed effects regression and hierarchical loglinear models. 11

18 This analysis in its entirety closely resembles the aims of Mayer, Werner, and Williams work in 2006 in the dependent variables of interest. However, the study uses more recent data, employs multivariate regression analysis, and examines election factors (challenger emergence, margins of victory and loss, and reelection) only for incumbents. DATA This study aims to discern whether publicly financed election systems lead to increased turnover in state legislatures, but also examines challenger emergence and election margins to check the robustness of my results and validity of my models against previous literature. The data is available from the Inter-University Consortium for Political and Social Research (State Legislative Election returns, ) and an addendum provided by Carl Klarner at Indiana State University. Professor Klarner will be releasing the latest version of the ICPSR data in the near future, but was kind enough to share his data (up to 2008) before its public release. Since clean elections are a relatively recent phenomenon, the analysis only used data from 1990 forward. Additionally, as this analysis is meant to focus on electoral outcomes of incumbents, all non-incumbents were all dropped from the data. The units of analysis are the candidates, and the data includes variables on election year, state, the candidates political party affiliation, the number of candidates in the race, the type of legislative seat (single member district, multimember district free for all elections, or multi-member district post elections), and indicators of whether the election was to fill a lower-chamber or upper-chamber seat. Further, imputed values relating to the professionalization of a state legislature were included (a full list of numerical 12

19 values and each state are included in appendix II) as well as population per member values (the full list by state and legislative body are provided in appendix III). Overall, this analysis is concerned with the rarest of occurrences: incumbent losses in publicly financed election systems. As the graphs below show, unsuccessful campaigns for reelection only happen 5.96% of the time. Further, 92.38% of state legislature incumbents elections occurred in traditional systems, with 2.01% in clean election systems and 5.62% in hybrid systems. Figure 1: 13

20 Figure 2: Election System Type 2.01% 5.62% 92.38% Presence of Clean Election System Presence of Traditional Election System Presence of Hybrid Election System Further, incumbency loss is becoming an increasingly rare occurrence. In the most recent decade, rates have declined steadily from their highest of 10% in the 1990 through 1991 period to a meager 3% in

21 Figure 3: 0.12 Incumbent Loss Rate Over Time ( ) When examined by political party affiliation, Democrats exhibit the strongest incumbent reelection efforts with only a 5.44% loss rate followed by incumbents of non-major political party affiliation with a loss rate of 6.28%. Republican incumbents fare the worst with the highest loss rate of 6.53%. 15

22 Figure 4: 7.00% 6.00% 5.00% 4.00% 3.00% 2.00% 1.00% 0.00% Incumbent Loss Rate by Political Party 6.53% 5.44% 6.28% Republicans Democrats Non-Major Presidential election years also have a compelling anti-incumbency affect for down ballot races. Incumbents seeking reelection in 1992, 1996, 2000, 2004, and 2008 were unsuccessful 7.27% of the time while off year incumbents loss rate falls to 5.28%. Figure 5: 8.00% 7.00% 6.00% 5.00% 4.00% 3.00% 2.00% 1.00% 0.00% On Year and Off Year Loss Rates 7.27% 5.28% Off Year Presidential Election Year 16

23 Further, substantial differences in loss rates for incumbents emerge if the data is examined by region with lawmakers in the West faring the worst with a 6.61% loss rate followed by Midwest incumbents at 6.24%. Incumbents in the Northeast and South fare substantively the same with loss rates at 5.71% and 5.70% respectively. Figure 6: 6.80% 6.60% 6.40% 6.20% 6.00% 5.80% 5.60% 5.40% 5.20% Loss Rate by Region 6.61% 6.24% 5.71% 5.70% Northeast Midwest South West Professionalization of the state legislative profession has been shown in past studies to significantly determine the extent of incumbency advantage (as noted above by Malhotra 2008; Holbrook and Tidmarch 1991; Gelman and King 1990; King 1991; Cox and Morgenstern 1993, 1995; and Carey, Niemi, and Powell 2000). Increased professionalization in this study is also related to stronger reelections with California (the most professional state legislative body) exhibiting a loss rate of 3.07% while New Hampshire (the least professional state legislative body) has an incumbent loss rate of 11.94%. There is a small but significant correlation (0.080) between professionalism and loss rate. Further, the mean professionalization value among 17

24 winning incumbents is 20.5 while this drops to 15.7 among losing incumbents. A t-test on the difference between these means finds it highly significant. As stated above, Carey, Niemi, and Powell (2000) found that growing district size lessens the amount of time the elected official can spend on the needs of individual constituents which, in turn, lessens incumbency advantage. However, in the data spanning from 1990 to 2008, the California State from 2002 to 2008 (which has the highest population per member at 846,791) had absolutely no incidents of incumbency loss. Meanwhile, the New Hampshire State in 1990 (which had the smallest population per member at 2,302) had an incumbency loss rate of 5.9%. The correlation between loss rate and district size (0.051) is relatively small. Though incumbent loss is rare enough that some may call it a mere fluke, patterns of incumbency loss emergence when the data are viewed through these different independent variables. Additionally, a pattern also emerges when loss rate is examined by election type: incumbents in traditional systems fare the best (only 5.92% lose), with those in hybrid systems doing slightly worse (6.53% lose), and clean election incumbents having the hardest time (6.61% lose). This analysis will examine the difference between these election outcomes and determine whether they can be reasonably credited to the election type or whether other relevant variables are driving outcomes. 18

25 % incumbents Lost Figure 7: Incumbents' Loss Rate By Election Type Clean Hybrid Traditional Model 1 (logit): Challenged in election [0-1] = β 0 + β 1 (Clean) + β 2 (Hybrid) + β 3 (UpperChamber) + β 4 (GOP) + β 5 (3 rd Party) + β 6 (multi-member post) + β 7 (professionalization) + β 8 (Presidential year dummy) + β 9 (ln(population per member)) + β 10 (naught-decade dummy) + β k (census regional dummies) In this model, if the number of candidates in the election was greater than one, then it was classified as an election with a challenger. In multi-member districts with free-for-all elections, incumbents are required to run against each other on the same ballot (along with any challengers that choose to run) with the top vote getters assuming office. Because of the nature of these seats and their election design, these incumbents will always have opposition. Therefore, they were not included in this analysis. Although not the main thesis of this paper, this model will discern whether public financing can be reasonably credited for boosting challenger emergence. Model 2 (OLS): Margin of Victory or Loss (percentage point difference) = β 0 + β 1 (Clean) + β 2 (Hybrid) + β 3 (UpperChamber) + β 4 (GOP) + β 5 (3 rd Party) + β 6 (# of opponents) + β 7 (multi-member post) + β 8 (professionalization) + β 9 (presidential year dummy) + β 10 (ln(population per member)) + β 11 (naught-decade dummy) + β k (census regional dummies) 19

26 For incumbents that won reelection, the margin of victory was the difference in percent of the votes they garnered and the second place finisher. For those that were not successful, the margin of loss was the difference between their percent of the vote garnered and the percent garnered by the winner. These values took a negative value. Voting margins were all coded as missing for multi-member district free-for-all elections in the data set. Therefore, those races are not included in this analysis. This model was run only on incumbents that faced opponents. Although not the main reason for this paper, this model will discern whether public financing systems can be reasonably associated with closer reelection vote margins. Model 3 (logit): Election outcome [0-1] = β 0 + β 1 (Clean) + β 2 (Hybrid) + β 3 (UpperChamber) + β 4 (GOP) + β 5 (3 rd Party) + β 6 (# of opponents) + β 7 (multi-member post) + β 8 (multi-member free-for-all) + β 9 (professionalization) + β 10 (presidential year dummy) + β 11 (ln(population per member)) + β 12 (naught-decade dummy) + β k (census regional dummies) In this model, any candidate that won reelection was coded as a one and zero otherwise. This analysis only pertains to incumbents facing opposition. This model discerns whether clean election or hybrid election systems are a statistically significant determinant of incumbents reelection. 20

27 Table 2: Logit and OLS Results Model 1 Model 2 Model 3 (drew opponent) (W/L Margin) (Won reelect) Clean Elections *** *** ( ) ( ) ( ) Hybrid Elections *** *** ( ) ( ) ( ) Upper Chamber *** *** *** ( ) ( ) ( ) Professionalization *** *** *** ( ) ( ) ( ) Multi-member post district (single member seat reference) *** *** ( ) ( ) ( ) Multi-member free-for-all district (single member seat reference) ** ( ) Republican candidate (Democrat as reference) *** *** *** ( ) ( ) ( ) 3rd Party candidate (Democrat as reference) *** *** *** ( ) ( ) ( ) Number of Opponents *** *** ( ) ( ) ln(population per Member) *** *** ( ) ( ) ( ) Presidential Election Year *** ( ) ( ) ( ) Midwest (New England as reference) *** *** ( ) ( ) ( ) South (New England as reference) *** *** *** ( ) ( ) West (New England as reference) *** *** *** ( ) ( ) ( ) Naught-Decade Dummy (1990s reference) ** *** *** ( ) ( ) ( ) Constant * * ** ( ) ( ) ( ) Observations 45,430 21,234 35,744 Adjusted R-squared (OLS model) Pseudo R-squared (logit models) Standard errors are in parentheses *** significant at 1%; **significant at 5%; *significant at 10% EVALUATING THE GOODNESS OF FIT FOR LOGIT MODELS For further attention to the goodness of fit of my logit models, I will review the association of predicted probabilities and observed responses. 21

28 Density For my first model on whether or not the incumbent drew a challenger, I obtained fitted probabilities from my model. I then compared these predictions to the actual challenges to incumbents. For these calculations, we assume that the model is predicting a challenger if the fitted probability is greater than 0.60, and is predicting no opposition if the fitted probability is less than The obvious break between in the distribution of fitted probabilities (shown below) appears to be around Therefore, this is an appropriate cutoff for calculating concordant and discordant pairs. Figure 8: Fitted Probabilities for Model Pr(OPPOSED) Concordance reflects when the model rightly predicts the emergence or non-emergence of a challenger. This occurs on 34,308 occasions. Wrong predictions from this model occurred 22

29 0 5 Density ,965 times. No ties (predicted probability of challenger emergence at 0.60) resulted from this model. Percent concordant is 66.91% while percent discordant is 33.09%. For my last model on whether or not the challenged incumbent won or lost an election, I obtained fitted probabilities from my model. In this instance, the model never makes a prediction below Therefore, using typical measures of concordant and discordant observations is not an appropriate measure of the quality of the model. Instead, I will examine the difference between the fitted probabilities for those candidates that won reelection and those that did not. Their distinctive distributions are shown below. Figure 9: Fitted Probabilities for Model 3 otherwise Predicted Probabilities Candidate won the election Pr(Win) Graphs by Election Outcome 23

30 The mean predicted probability among candidates that actually won reelection is while among candidates that lost it was slightly lower at Using a basic t-test, the difference between these two means (0.018) is highly statistically significant (t-statistic = with a twotailed p-value < 0.000). Predicting rare occurrences will not be easily seen in the model and explains why the difference between the distributions is so slight. RESULTS From the coefficients and corresponding odds ratios, we can discern which variables have an anti-incumbent effect and which help politicians retain their seats. As theorized, clean and hyrbid electoral systems both have an anti-incumbency effect by boosting challenger emergence and shrinking margin of victory: Incumbents seeking reelection in a clean election system are 113% more likely to face an opponent and have fared 7.4 percentage points worse than incumbents running in traditionally funded election systems. Incumbents seeking reelection in a hybrid election system are 70% more likely to face an opponent and have fared 2.5 percentage points worse than those running in traditionally funded election systems. For the model pertaining to election outcome, clean and hybrid election systems both had negative coefficients but were not statistically significant at traditional levels (p-values of and respectively). Upper chamber and non-democratic incumbents also have the hardest electoral roads with higher challenger emergence, declining vote margins, and fewer reelections. Upper chamber incumbents are 36% more likely to face an opponent, have fared 5.4 percentage points worse, and are 29% less likely to win their elections than a lower chamber incumbent. 24

31 Republican incumbents are 12% more likely to face an opponent, have fared 2.3 percentage points worse, and are 11% less likely to win their elections than Democratic incumbents. 3 rd party incumbents are 20% more likely to face an opponent, have fared 4.4 percentage points worse, and are 29% less likely to win their elections than Democratic incumbents. Seeking reelection in a Presidential election year has no statistically significant effect on challenger emergence or vote margin (p-values and respectively). However, there is a lessening of the incumbent advantage as incumbents in these years are 23% less likely to win reelection than those in other years. Unsurprisingly, as the number of opponents an incumbent faces increases, their vote margin declines and reelection chances diminish. For each additional opponent in their race, incumbents have fared 2.3 percentage points worse and are 3% less likely to win their reelections. Alternatively, time trends are both working in incumbents favor. Incumbents running for reelection in the most recent decade (2000 and later) are 5% less likely to face a challenger, have fared 4.1 percentage points better, and are 43% more likely to win reelection than incumbents that ran in the 1990s. Two variables in my model the professionalization of the state legislature and population per member have mixed anti- and pro- incumbency effects. Both variables boost challenger emergence, while also increasing vote margins and reelection chances. Professionalization is a variable running from zero to one hundred and is meant to reflect the extent to which the legislative body is as professional as the United States of Representatives. The minimum is New Hampshire (3.4) while the maximum is California (57.1). For a one-unit increase in professionalism, we expect challenger emergence to increase by 2%, the vote margin to increase by.17 percentage points, and reelection to go up by 2%. 25

32 For a 10% increase in population per member, we expect a 1.5% increase in challenger emergence, vote margin to increase by 0.34 percentage points, and a 170% increase in reelection. Overall, these results are largely consistent with Mayer, Werner, and Williams (2006). Challenger emergence and competitiveness are both boosted by the presence of clean and hybrid systems. However, this analysis diverges from the Mayer, Werner, and Williams study in examining election rates. Because their analysis does not use multivariate analysis, they are observing a one-dimensional drop, which is also seen in this study (loss rates were 6.61% for clean elections, 6.53% for hybrid elections, and 5.92% for traditional elections). But when analysis also controls for other pertinent variables, the presence of the clean or hybrid system no longer has statistical significance. Further, this analysis differs from Malhotra s 2008 analysis which finds no statistical significance in the presence of a clean elections statute (found in Arizona and Maine) in determining vote margins. This analysis found statistically significant and large differences made by publicly funded systems with clean election incumbents faring 7.4 percentage points worse and hybrid election incumbents faring 2.5 percentage points worse. The differences between this study and Malhotra can likely be attributed to the substantially different model specifications. While Malhotra includes analysis of all candidates in the data set, this study only examines outcomes for incumbents. By casting this wider net, Malhotra s model includes a dummy variable for incumbency which is highly statistically significant and driving many of the results in the model. Additionally, Malhotra includes measures of challenger quality which is left unaddressed here. 26

33 The Government Accountability Office s 2010 analysis of Maine and Arizona s clean election systems found statistically significant decreases in the winner s margin of victory. However, they note that this decrease is probably due to other, unobserved factors such as the popularity of the candidates. This analysis is consistent with the GAO s most updated analysis, and also recognizes that the occurrence of incumbent loss is exceedingly rare and perhaps most commonly induced by exceptional circumstances such as scandal or rare talent and charisma in a challenger. IMPLICATIONS This analysis was meant to discern the electoral implications of clean and hybrid election systems for incumbents running for reelection. As shown above, clean and hybrid election systems have a statistically and contextually significant effect an incumbents electoral future. Incumbents in these systems should expect challengers to emerge more often and that the margins of competitiveness work in the newcomers favor. However, when it comes to the most important results of an election which candidate actually emerges the winner clean and hybrid systems do not change the election end outcome. This removes one of the biggest political obstacles towards further adoption of these systems, as incumbents currently in office the audience that most controls public financing election systems future can maintain a self-interested stance by enacting these systems. With this in mind, policymakers in state houses around the country can evaluate publicly financed electoral systems on whether they meet their stated objectives (namely, minimizing the political influence of money) without worrying over their own electoral future. 27

34 While beyond the scope of this study, it has been suggested (GAO 2003) that lobbyists and donors influence has not been diminished by these systems. But these suggestions are mostly anecdotal accounts and are worth more rigorous, academic study. Further, the elections in 2009 and 2010 have been in the midst of an anti-government era with the sentiment reaching a fervent pitch. The clean election systems in Arizona, Maine, and Connecticut are a new enough phenomenon that they did not exist during other anti-incumbent waves of recent memory (such as the 1994 elections). While hybrid systems in Minnesota, Wisconsin, Hawaii, and Nebraska have been time tested, clean election systems have not. Therefore, more robust and reliable analysis of clean election systems should include the historic wave elections in 2009 and Further, previous studies point to the significant anti-incumbency effects of multi-member districts. This analysis contains some commentary on MMDs, but the data do not lend itself to accessible analysis of challenger emergence, vote margins, and their intersection with MMDs. Additionally, each state has such unique election laws, political attitudes, demographics, and countless other factors that potentially sway electoral outcomes. Other analyses have validly taken alternative approach by finding public financing effects within one state, then conjectured effects to other states. The external validity of this method is of concern but it may provide a clearer picture of the program s effects. 28

35 Qualifying Phase Small individual donations required Primary Election Phase Maximum Initial Disbursement Maximum matching funds General Election Phase Maximum Initial Disbursement Maximum matching funds Total Election Cycle Maximum total liability due to individual candidate APPENDIX I: SUMMARY OF CLEAN ELECTION SYSTEM OPERATIONS IN ARIZONA, MAINE, AND CONNECTICUT Arizona Maine Connecticut $1,100 (in $5 bundles) $1,100 (in $5 bundles) $250 (in $5 bundles) $750 (in $5 bundles) $5,000 (in $5- $100 bundles) $15,000 (in $5- $100 bundles) $12,921 $12,921 $1,504 $7,746 $10,000 $35,000 $25,842 $25,842 $3,008 $15,492 $20,000 $70,000 $19,382 $19,382 $4,144 $19,078 $25,000 $85,000 $38,764 $38,764 $8,288 $38,156 $50,000 $170,000 $96,909 $96,909 $16,944 $80,472 $105,000 $360,000 APPENDIX II: DEFINING PROFESSIONALIZATION The scale developed by Squire (2000) has been widely used by subsequent literature. In my analysis, each value in the scale was multiplied by 100 to make it more reminiscent of percentages, rather than proportions. But the significance and substance is unchanged by this manipulation. The measure uses data on 1995 regular session legislative compensation and the mean number of legislative days in regular sessions for 1994 and 1995, both taken from The Book of the States The number of staff members in each state for 1996 is taken from data compiled by the National Conference of State Legislatures, "Size of State Legislative Staff: 1979, 1988, and 29

36 1996," using the total staff during session numbers. The number on each measure for each state is compared to that for the U.S. Congress during the same years. In essence, the final score for each state can be interpreted as how closely the state legislature mimics the U.S. Congress in terms of professionalization. Rank State Score 1 California Michigan New York Wisconsin Massachusetts New Jersey Ohio Pennsylvania Hawaii Florida Illinois Alaska Texas Washington Missouri Maryland Oklahoma Arizona Minnesota Connecticut Colorado Nebraska Nevada Oregon Iowa Delaware Virginia North Carolina Louisiana South Carolina Mississippi Tennessee Vermont

37 34 West Virginia Rhode Island Idaho Kansas Georgia Indiana Arkansas Maine Kentucky Montana Utah Alabama South Dakota North Dakota Wyoming New Mexico New Hampshire

38 APPENDIX III: POPULATION PER MEMBER OF STATE LEGISLATIVE BODIES Leg. Body AK AK AL AL AR AR AZ AZ CA CA CO CO CT CT DE DE Per Capita Representation Population per member ,673 15,673 15,673 15,673 15,673 15,673 15,673 13,751 13,751 13,751 13,751 13,751 13,751 13,751 13,751 13,751 13,751 10,046 10,046 31,347 31,347 31,347 31,347 31,347 31,347 31,347 27,502 27,502 27,502 27,502 27,502 27,502 27,502 27,502 27,502 27,502 20,093 20,093 42,353 42,353 42,353 42,353 42,353 42,353 42,353 38,482 38,482 38,482 38,482 38,482 38,482 38,482 38,482 38,482 38,482 37,085 37, , , , , , , , , , , , , , , , , , , ,254 26,734 26,734 26,734 26,734 26,734 26,734 26,734 23,507 23,507 23,507 23,507 23,507 23,507 23,507 23,507 23,507 23,507 22,864 22,864 76,383 76,383 76,383 76,383 76,383 76,383 76,383 67,164 67,164 67,164 67,164 67,164 67,164 67,164 67,164 67,164 67,164 65,327 65,327 85,511 85,511 85,511 85,511 85,511 85,511 85,511 61,087 61,087 61,087 61,087 61,087 61,087 61,087 61,087 61,087 61,087 45,304 45, , , , , , , , , , , , , , , , , ,174 90,607 90, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,698 66,173 66,173 66,173 66,173 66,173 66,173 66,173 50,683 50,683 50,683 50,683 50,683 50,683 50,683 50,683 50,683 50,683 44,461 44, , , , , , , ,893 94,126 94,126 94,126 94,126 94,126 94,126 94,126 94,126 94,126 94,126 82,570 82,570 22,553 22,553 22,553 22,553 22,553 22,553 22,553 21,769 21,769 21,769 21,769 21,769 21,769 21,769 21,769 21,769 21,769 20,580 20,580 94,599 94,599 94,599 94,599 94,599 94,599 94,599 91,511 91,511 91,511 91,511 91,511 91,511 91,511 91,511 91,511 91,511 86,322 86,322 19,112 19,112 19,112 19,112 19,112 19,112 19,112 16,248 16,248 16,248 16,248 16,248 16,248 16,248 16,248 16,248 16,248 14,496 14,496 37,314 37,314 37,314 37,314 37,314 37,314 37,314 31,722 31,722 31,722 31,722 31,722 31,722 31,722 31,722 31,722 31,722 28,302 28,302 32