Congressional Gridlock: The Effects of the Master Lever Olga Gorelkina Max Planck Institute, Bonn Ioanna Grypari Max Planck Institute, Bonn Preliminary & Incomplete February 11, 2015 Abstract This paper accounts for the effects of the master lever, aka the Straight-Ticket Voting Option, on policy making, by focusing on the positions of elected US congress members from 1952 to 2012. The master lever is an option offered in some US states allowing voters to vote for a specific party for all offices listed on a ballot by selecting a box at the top, as opposed to filling out each office individually. We build a novel theory of pre-election competition where parties select their candidates (and equivalently their positions) for state elections by trading off the probability of winning against the party bliss point. The presence of the master lever makes voters less sensitive to candidates positions and more focused on their party affiliation. This changes the parties tradeoff. For example the party favored in a partisan state will become more extreme in the presence of the master lever and the other party more moderate in order to attract the (even stronger now) partisans. By allowing for two-dimensional politician positions, we are able to get predictions for swing states, where voters have conflicting interests across parties and dimensions. We use data from the American National Election Survey, Poole and Rosenthal s two-dimensional congress member positions, and the exogenous variation with respect to the existence of a master lever across states and congresses to test the predictions of the model. We then parameterize the legislative process and using data on congressional bills from the Policy Agendas Project we calibrate an extended version of the model. We quantify the effects of the master lever on congressional gridlock, where the tradeoff arises from the benefit of having less extreme politicians in congress and the disadvantage (in terms of gridlock) of having more unified parties. The authors thank Derek Stemple and attendees of the Max Planck Institute s joint seminar, as well as Georg Treuter for valuable research assistance. 1
1 Introduction 2 Master Lever and Politician Selection 2.1 Theory Fix state s, election period t and a voter i s. The voter s position on the set of issues {1, 2,...N} is described by the vector: x it = (x it1, x it2,...x itn ) (1) x jt is j s preferred point in the policy space. Candidate j representing party P (P {R, D}) is running for state s at t. The candidate s position on the set of issues is given by: y jt = (y jt1, y jt2,...y jtn ) (2) The position of the Democratic party at time t is denoted Y Dt and the position of the Republican party at time t is denoted Y Rt. Voter i s utility from politician j s term in office depends on the difference in their positions on the set of issues: u j i = n (x itn y jtn ) 2 (3) On the election day, the voter receives an imperfect signal of her true utility. The signal is affected by (i) the candidate s party affiliation P, if the voter is a partisan of P, and (ii) a candidate-specific preference shock δ j common to all voters 1 ũ j i = uj i + µip i + (1 µ) δ j (4) where: 1 It is possible to include the voter-specific idiosyncratic noise in equation (4). If the noise term has mean zero, such change will have no implication on (5) and the following results. 2
δ j follows a uniform distribution on ( 1 2, 1 2) and denotes a random advantage over the opponent; I P i = 1 if voter i is a partisan of P, and I P i = 0 otherwise; µ > 0 is the weight of the politician s party affiliation in the utility of a partisan voter. The positions x jt of voters in a specific state are distributed according to an exogenous function f st (x it ) (or f s (x it )). This distribution function reflects the political climate in a state. Fat tails imply a large fraction of extreme views, positive or negative mean imply leaning towards the right or the left, respectively, in the state (in period t). Looking at aggregate levels, we assume that there is mass 1 of voters in state s and denote the voters aggregate utility from assigning the seat to politician j P by Ũ j : Ũ j = = ũ j i di u j i di + µπp + (1 µ) δ j, (5) where π P is the fraction of P -partisan voters in state s. Consider d D, and r R: two candidates representing the Democratic and the Republican parties, respectively. The probability of winning the seat in state s depends on the difference between the total voter satisfaction from each candidate subject to a random noise. To fix ideas, think of candidate d from the Democratic party. Her winning probability is given by: 0, if U 1 µ 1 2 Pr (Ũ d > Ũ r) = 1 2 + U 1 µ, if 1 2 < U 1 µ 1, (6) 2 1, if U 1 µ > 1 2 where U = u d i di + µπd ( u r i di + µπr) is the difference in aggregate utilities between parties, excluding the random preference bias. 3
Each party maximizes the probability of winning the seat in the election. At the same time, the parties incur a cost when the chosen candidate s position is too different from the party position (the party cares about the implemented policies after the election). For example, the Democratic party; it chooses a candidate whose position solves the optimization problem: max y dtn Pr (Ũ d > Ũ r) c Y Dtn y dtn, (7) n where n Y Dtn y dtn is the cost of departing from the party s bliss point. If Y Dtn > x stn n, that is, the Democratic party is on the right from the average voter on every issue, then the interior solution has to satisfy the following first order condition: y dn x sn 1 µ = c m n (Y Dn y dn ) (8) for all n. For example, if there are two issues, 8 is equivalent to the following: y d1 x 1 = c (1 µ) (Y D2 y d2 ) y d2 x 2 = c (1 µ) (Y D1 y d1 ) We use these first order conditions to analyze the effect of the master lever on the party s choice. In our model, the master lever translates into the parameter µ. When the master lever is present, partisan voters are more likely to vote for the party s candidate. This implies that the parameter µ is higher in those states than in the states without the master lever. Figure 1 illustrates what the different levels of µ imply for the solution. Consider a state that starts as a master lever state. The first order conditions are represented by two solid lines with the negative slope c(1 µ) and 1 c(1 µ). Here, µ corresponds to a higher level of partisanship effect due to the master lever. The intersection of the solid lines gives us the solution ȳ to the party s optimization problem. Next, suppose that the state abandons the master lever. The first order conditions corresponding to the new level of 4
µ = µ are represented by the dashed lines. The new solution y is located to the right and above the master-lever solution ȳ. This implies that abandoning the master lever induced the party to make the choice of a more extreme candidate. Or, in the other direction, master lever leads to more moderate candidates chosen in the state. Y 2 y 2 c(1 µ)y 1 c(1 µ)y 1 ȳ y x = 0 c(1 µ)y 2 c(1 µ)y 2 Y 1 y 1 Figure 1: The effect of the master lever on policy choice. However, the master lever may also have an ambiguous effect, leading to the choice of candidates that are more moderate in one dimension, while becoming more extreme in the other. Figure 2 illustrates this point. Here, we start with a constellation of parameters c, µ where ȳ is the choice under the master lever. Abandoning the master lever corresponds to decreasing µ to the level µ. As in Figure 1, this shifts the lines outwards, and leads to a new intersection point y. Similarly to Figure 1, the new solution y lies above the master-lever solution y, corresponding to a more extreme position in dimension 2. However, in dimension 1 we observe the opposite effect: removing the master lever induced the party to choose a candidate that is more moderate on the first issue. 5
Y 2 y 2 c(1 µ)y 1 y c(1 µ)y 1 ȳ x = 0 c(1 µ)y 2 c(1 µ)y 2 Y 1 y 1 Figure 2: The effect of the master lever on policy choice. 2.2 Empirics 2.2.1 Data We use data from the 83rd (1953) to the 112th (2012) congresses and their preceding elections on voters, congress members and parties by state and congress. Voter data on self-declared bliss points, split tickets, 2 interest in the election and other controls, was constructed from the National Election Studies. As data was not available for every state and congress, we used other characteristics to impute the missing observations. Data of vote shares by state for presidential elections is taken from the US Elections Atlas. Congress member data comes from Poole and Rosenthal s DW-Nominate scores that offer a two dimensional mapping of a politician s position based on their voting history in congress. The two dimensions are economics and ideology. To construct party bliss points we use two measures: 1. the average two-dimensional position of all elected members of that party and 2. the party position from the Manifesto Project database. The latter is constructed using content analysis of party political platforms issued in the presidential nominating conventions. Lastly, a binary variable of the presence of the master lever by state and congress was constructed by hand by researching individual states. For now, in everything that follows we use data for the US Senate only. 2 Voting for a different party for the presidential and congressional elections. 6
2.2.2 Summary Statistics The number of states with a master lever has changed from 20 in the 1952 election to 13 in the 2010 midterm election. Figure 3 shows the evolution of the average position of the Republican and Democratic parties, averaging over states depending on their master lever status. Overall Senators of both parties have become more extreme in terms of economics (dimension 1) and more moderate in ideology (dimension 2). In Figure 4, showing the standard deviation from party averages, one can see that parties are more unified in economics and more dispersed in ideology. On the flip side, figure 5 showing voter bliss points, constructed using their self-declared partisanship where 1 is strongly Democratic and 1 is strongly Republican, indicates that on average voters are becoming more moderate. Evolution of Party Averages Dimension 1 Dimension 2 Average Position -.6 -.4 -.2 0.2.4.6 Average Position -.6 -.4 -.2 0.2.4.6 83 87 91 95 99 103 107 111 Congress 83 87 91 95 99 103 107 111 Congress Dems with ML Reps with ML Dems w/o ML Reps w/o ML Dems with ML Reps with ML Dems w/o ML Reps w/o ML Figure 3 7
Evolution of St.Dev. from Party Average Dimension 1 Dimension 2 StDev from Average 0.2.4.6 StDev from Average 0.2.4.6 83 87 91 95 99 103 107 111 Congress 83 87 91 95 99 103 107 111 Congress Dems with ML Reps with ML Dems w/o ML Reps w/o ML Dems with ML Reps with ML Dems w/o ML Reps w/o ML Figure 4 Average Voter Bliss Points -.5 -.25 0.25.5 83 87 91 95 99 103 107 111 Congress Voter Bliss Points Figure 5 2.2.3 Results In the following regressions we run the master lever variable (binary) on the absolute positions of elected senators, controlling for the average voter bliss point in that state. 3 A 3 So that a Democratic state, for example, denotes one where the average voter is more than one standard deviation below the swing voter. 8
negative (positive) effect implies senators are more moderate (extreme) when there is a master lever. We include state trends in all specifications. These results are robust to including congress and state fixed effects, which we have excluded from the main specification as we use other congress and state controls. Table 1: ML on Dimension 1 Positions Republican States Swing States Democratic States Democrats Republicans Democrats Republicans Democrats Republicans Regression (1) (2) (3) (4) (5) (6) ML on Dim 1.1461.2325.0382.0784.0185.0288 ML on Dim 2.1929.281.0101.1497.0936.2667 Senators only. Controls: general or midterm, average party position in both dimensions The results in Table 1 confirm the dynamics in the theory. Specifically in column (1) we see that Democratic candidates become more moderate in the presence of the master lever in Republican states. This is because the master lever only strengthens the Republican partisanship of the state and thus for a Democratic candidate to win they have to be even closer to the state average voter bliss point. The opposite is true for the party that is favored in the partisan state. In columns (2) and (5) we see that Republicans and Democrats respectively become more extreme in the presence of the ML. In this case the strengthening of the effect of parties works in their favor as they can go closer to the party bliss point. In columns (3) and (4) we observe that in swing states the results vary depending on party and dimension. This is because swing voters have by definition conflicting interests, preferring one party in one dimension and one in other, therefore it is not clear how the presence of the master lever would affect states. In order to understand this we must first examine how the effect of changes depending on the bliss point of the party as seen in figures 1 and 2 in the theory. 9
2.2.4 Endogeneity of Master Lever In this section, we consider empirically the potential endogeneity of the master lever. We start by allowing states that have the same master lever status (always had, never had, had and removed) to have common unobserved characteristics and by clustering errors at that level to redo the above estimation. 3 Congressional Gridlock In this section, we estimate the effects of the master lever on congressional gridlock which we define as the difficulty of passing laws in the legislature due to the individual positions of US congress members. As shown in the previous section, on aggregate, more states having a master lever implies a more moderate congress thus facilitating the legislative process. On the other hand, it also implies more unified parties which could increase gridlock. First, we extend the theory from Section 2 to include a parameterized version of the legislative process that follows the elections. Then, using data from the Policy Agendas Project on congressional bills we calibrate the model and quantify the effect of the master lever on congressional gridlock and thus policy making in the US, through politician selection. 10