Political Campaigning & Policy Implementation: Evidence from US Presidential Elections Ioanna Grypari Max Planck Institute, Bonn Grigorios Siourounis Panteion University Preliminary Draft February 14, 2014 Abstract This paper empirically establishes the relationship between pre-election competition and post-election policy making, in US Presidential elections from 1952 till 2008. Specifically, we account for the effect of political campaigning, through issue emphasis, on bills proposed and passed during a Presidency. First, we construct a campaign effect variable by isolating the fraction of post-election bill proposals that is solely due to campaign promises. We do so by exploiting the observed difference in behavior between the winning and losing party of the Presidential election. Second, we build a dynamic structural model to allow for the endogeneity of bill proposals to campaign promises and vice versa. Using data on bills from the Policy Agendas Project and survey data from the American National Election studies, as well as our constructed variables from the first part of the paper, we estimate the model. We are thus able quantify the effects of political campaigning on bills passed into law, through agenda priorities. Preliminary findings show that campaign effects dissipate or persist depending on the party that controls the Congress, the term of the President and on whether an issue was high on the agenda for both parties. We thank Martin Hellwig, Olga Gorelkina, workshop participants at MPI Bonn and Panteion University, and Georg Treuter for excellent research assistance. 1
1 Introduction This paper quantifies the effect of pre-election political competition on policy changes in the US from 1952 till 2008, in the context of Presidential elections. Specifically, the channel we focus on is that of agenda priorities. During the campaign, Presidential candidates spend varying amounts of time on different issues (to be referred to as issue emphasis from now on), thus promising the voters a specific hierarchy of the agenda if they are to come into office. 1 After the election, the two parties allocate their time across issues by proposing bills (from now on issue effort). First we isolate empirically the campaign part of issue effort, i.e. the fraction of bills proposed due to politicians sticking with their campaign issue emphasis. We then build and estimate a dynamic structural model that allows us to control for the endogeneity of pre- to post- election behavior. Our results allow us to quantify the effect of campaign issue emphasis on actual bills passed into law. Although there is a large theoretical literature on the effects of electoral competition on economic outcomes, this relationship is yet to be established empirically, especially in terms of actual policy changes. 2 The contribution of this paper is three-fold. First, by focusing on agenda priorities, we are able to isolate a force in the legislative process that is tractable. 3 Second, the model we provide is flexible because of the simplifying parameterizations we use, but still captures the basic dynamics of the political process, allowing it to be used in different settings. Lastly, we are able to measure 1. whether there are any campaign effects persisting after the elections and 2. how these vary depending on campaigns, Congress control and the Presidential term. Thus, shedding more light on the way in which the existence of pre-election competition affects observed outcomes. We start by empirically examining the fraction of issue effort that is due to campaign promises. We run a simple regression of bills proposed on an issue on campaign issue 1 The link between actual policy promises and policy making although clearly a very relevant one is harder to measure as 1. there is little data on policy promises and 2. the data that exists shows a large variation in the perceived positions of a Presidential candidate in the electorate. However, although the focus on the paper is on the effect of agenda priorities on policy changes, we do address the issue of exact policy promises in the last section of the paper. 2 There is, however, an empirical literature on the effect of the party of the President on economic outcomes (not focusing on campaigns), as well as a strand on the effects of campaign issue emphasis on government expenditures (an approximation to policy changes). 3 To voters as well. 2
emphasis controlling for year, issue and party fixed effects and observe that the winning and losing parties of the Presidential election behave differently. For the Presidential party there is a statistically and economically significant positive correlation between issue emphasis and issue effort whereas the effect is insignificant for the opposition. Using this difference we construct a variable that isolates the fraction of bills on an issue that are due to the Presidential party sticking with their agenda priorities as set in the campaign (from now on the campaign effect estimates). Second, we build a dynamic structural model that allows us to account for 1. the potential reverse causality between bill proposals and campaign promises as well as 2. the dependence of bills to future election motives. By parameterizing legislative bargaining we are able to significantly simplify the political process and ultimately reduce the infinite horizon model to a dynamic programming problem with two state variables. Using data on bills, voting and our campaign effect estimates variables from the first part of the paper, we estimate the model to match predicted moments. Preliminary findings indicate that pre-election competition has varying effects on bills passed into law depending on whether the issue is major or minor in the previous election. In particular, campaign effects dissipate, or rather, are not identifiable in issues that were highly emphasized from both parties, potentially indicating that these are major state of the world issues and would have been addressed regardless. For issues that were highly campaigned on by the winning party, campaign effects depend on the party that holds power over the Senate and House of Representatives, but are in several cases (i.e. above some threshold level of Congress control ) persistent. As expected campaigns have a smaller effect on policy on years where the President is in his second term. 45 4 But may still exist. 5 Although we can identify the effects of political competition on policy changes through agenda priorities, we are not able to examine whether campaign policy promises are kept as there is no data on that. However, we examine the difference between the position of the original post-campaign bill proposal on an issue and the bill that was actually passed into law. 3
2 Data and Campaign Effects Estimates In this section, we identify the part of the legislative process that is due to campaign promises. Data Campaigns vary in several dimensions, two relevant ones for this paper are the positions and issue emphasis of candidates; the latter referring to the salience of different issues in the campaign. We classify issues according to the Policy Agendas Project of the University of Texas. The handbook contains 21 major topics and a varying number of subtopics for each. It is the same one used for the bills data (see below), and thus covers all possible issues besides campaign oratory remarks which we measure separately. The issue emphasis data is taken from Grypari (2014) and is constructed by counting the number of words in a particular issue in the party nominating convention acceptance speeches given by Presidential candidates. The speeches are thought of us representative of the agenda of the candidate and have the benefit of being the largest campaign event (by number of viewers) after the debates. 6 Data on the positions of Presidential candidates is extracted from the American National Election Studies data-set and it includes the perceived positions of voters by demographic group in the population on several of the major topics. We do not have data on candidates positions in all 21 major categories, but this is not essential in this part of the paper as we are focusing on agenda priorities rather than exact positions. 7 In order to quantify the legislative process we use data on bills proposed in Congress from 1952 till 2008 through the Policy Agendas Project. Each data point 8 contains all relevant data for a particular bill: information on the Congress member that proposed it, date of proposal, the chamber in which it originated, the issue it referred to, whether it was signed into law, its NOMINATE score, 9 etc. We proxy post-election issue effort by the 6 Debates cannot be used as a source of data on agenda/issue emphasis as the issues discussed are at least partly determined by the moderator. 7 We address the issue of missing data in terms of campaign promises in section 4 of the paper. 8 About 410,000 in total. 9 A scale created by Poole and Rosenthal that summarizes the position of a bill in a two-dimensional 4
fraction of bills on a specific issue and we construct this measure separately for each of the four years after the election. 10 As bills are not allowed to be proposed by the President himself and must originate from a member of the congress we use the bills from the President s party as representative of his own preferences, but we take this into consideration in our estimation. Campaign Effects In order to account for the effect of pre-election competition on policy making, we must take into consideration the different forces present in the legislative process. Besides the willingness to pass policy on the issues that were emphasized during a campaign, once elected the winning party is also facing their own party preferences, special interest groups, changes in the state of the world, 11 and future election considerations. In this section of the paper, we identify the fraction of bills proposed in each issue that is solely due to campaign promises. Now, issue emphasis may be correlated with all of the above, as election motives are not necessarily the only incentives present when setting agenda priorities during the campaign. Party preferences for example will affect both the campaign but also bill proposals post-election, 12 therefore the estimates in this section are meant to be interpreted not as exogenous to the rest of the political process, but rather as isolating campaign effects. To construct them, we use the fact the winning and losing party of an election behave differently post-election in terms of the relationship between between campaigning and policy priorities. We regress issue effort (fraction of bill proposals on issue) on issue emphasis (campaign time on issue), controlling for party, year and issue fixed effects. 13 We find that for the winning party there is a significant and positive correlation between campaign and post-election priorities (the coefficient is 0.63, significant at the 5% level), whereas the correlation is insignificant for the losing party. 14 We vary the specification by looking at scale. 10 Noting the different incentives in play during each year of the Presidency. 11 Drawing attention to a potentially different set of issues. 12 Thus, the need of a structural model. 13 Using the Wild cluster-bootstrap percentile-t procedure that allows double clustering in issues and years. 14 The latter is somewhat surprising. Although, the Presidential candidate of the losing party is no longer 5
bills proposed at different years of the Presidency and find these results robust. We construct the campaign effects issue effort variable (for the winning party) simply by multiplying issue effort with the coefficient of the regression. By creating this variable, we are able to significantly simplify the model as we have isolated the campaign part of issue effort and can summarize the rest of the forces in play (which are not of interest to us) by a simple parameter. 3 Model In this section, we present a simple two stage model that is repeated infinitely. For now, we drop the subscript for time, t, from all notation. Agents There are two politicians j {r, d}, two parties J {R, D} and a continuum of voters i [0, 1]. Each voter belongs to a (demographic) group g G where π g is the fraction of voters in g, g π g = 1. For now, we assume that politicians and their parties are identical and thus drop J. Actions Voters vote for one of the two politicians, v ig {r, d}. There are n issues, n N, each politician has an exogenous position on each of these issues, p j n [ 1, 1] and there is a status quo position p 0 n [ 1, 1]. Politicians campaign by spending time on different issues in the campaign, e j n s.t. n ej n = 1, which we call issue emphasis. Politicians also decide how to allocate their time across issues after the election, b j n s.t. n bj n = 1, we call this issue effort and denote it by b as it will be proxied by bill proposals. There is an exogenous function λ n ( ) that summarizes the legislative bargaining process. With p n being the policy that gets implemented at the end of the Presidency we have: ( p n = λ n b j n, b j n, p j n, p j n, p 0 ) n (1) Timing in the picture during the legislative process party preferences were still present when setting the agenda during the campaign. 6
At time t, a two stage period begins. Stage 1: Politicians campaign by setting {e j n} jn and voters vote {v ig } ig. Stage 2: The majority winner j takes office, both politicians allocate effort across issues {b j n} jn and policy positions change to {p n } according to (1). Payoffs Politicians care about the probability of winning elections and their own preferred positions, p j n. Voters utility depends on the salience of different issues as determined by issue emphasis in this and the previous campaign. 4 Empirical Testing In this section, we use a GMM estimator to estimate the model from section 3, using bill proposals data from the Policy Agendas Project, the estimates on campaign effects from section 2 and survey answers from the National Election Studies data set. 15 The predicted moments to match are: 1. the fraction of bills that are passed into law on each issue, 2. the fraction of seats of both parties in the Senate and House of Representatives, 3. the average party position on issues (using NOMINATE scores of party memebers), 4. ultimate re-election probability of President (whether in first or second term and Presidential approval rate at the end of this term). Preliminary findings indicate that pre-election competition has varying effects on bills passed into law depending on whether the issue is major or minor in the previous election. In particular, campaign effects dissipate, or rather, are not identifiable in issues that were highly emphasized from both parties, potentially indicating that these are major state of the world issues and would have been addressed regardless. For issues that were highly 15 For an exact description of the different data-sets please refer to the previous sections. 7
campaigned on by the winning party, campaign effects depend on the party that holds power over the Senate and House of Representatives, but are in several cases (i.e. above some threshold level of Congress control ) persistent. As expected campaigns have a smaller effect on policy on years where the President is in his second term. 8