Does Responsiveness Stop at the Water's Edge? Examining the Foreign Policy Electoral Connection

Does Responsiveness Stop at the Water’s Edge? Examining the Foreign Policy Electoral Connection

Michael Pomirchy∗ and Bryan Schonfeld† April 3, 2020

Abstract Are legislators responsive to public opinion on foreign policy issues? Despite the preva- lence of preference-based approaches to international relations, no existing study has system- atically assessed the relationship between constituency preferences and foreign policy roll-call votes in the U.S. House. Using multi-level regression and post-stratification (MRP), we create original estimates of constituency preferences, broken down by party, on various trade, security, and immigration bills and measure the responsiveness of members of the U.S. House of Representatives. We find that members are responsive to their median constituent on all issue areas, but there is less responsiveness on trade bills than on security and immigration. We also find that members of Congress are more responsive to co-partisan constituents than to the median on security bills–the reverse holds for trade bills. Additional analyses suggest that cases of greater responsiveness to co-partisans do not result from voters taking partisan cues from their representatives.

∗Michael Pomirchy is a Ph.D. candidate in the Department of Politics at Princeton University. †Bryan Schonfeld is a Ph.D. candidate in the Department of Politics at Princeton University. Helpful comments and suggestions were provided by Brandice Canes-Wrone, Jonathan Kastellec, Helen Milner and Dustin Tingley. Introduction

Do American legislators adhere to public opinion when voting on foreign policy issues? While there is already substantial evidence that American legislators are responsive to their constituencies on policy issues (Canes-Wrone, 2015; Erikson, MacKuen and Stimson, 2002; Stimson, MacKuen and Erikson, 1995), there is less work that investigates responsiveness to constituency (and sub- constituency) preferences on foreign policy. There are reasons to believe that responsiveness on domestic policy may not extend past the water’s edge: voters rank foreign policy low in importance in polls, so the conventional wisdom holds that foreign policy is not salient to the American public (Almond, 1950; Busby and Monten, 2012). However, voters might indicate that foreign policy issues are unimportant to their vote choices because they are satisﬁed with status quo policies: new experimental evidence suggests that Americans care as much about foreign policy when evaluating candidates as they do about economic or religious issues (Tomz, Weeks and Yarhi-Milo, 2020).1 Responsiveness on foreign policy may also be different because voters do not perceive large policy differences between the parties on foreign policy issues (Kertzer, Brooks and Brooks, 2017), though some evidence suggests that voters do take partisan cues from elites on some foreign policy issues (Berinsky, 2007; Trager and Vavreck, 2011). On foreign policy, the literature has shown that the government as a whole is responsive to mass opinion on defense spending and related policies (Bartels, 1991; Wlezien, 2004). On trade, the previous literature has largely used proxies for constituency preferences at the district level to examine responsiveness of U.S. legislators. In particular, studies have focused on district-level exposure to Chinese import competition (Feigenbaum and Hall, 2015; Kuk, Seligsohn and Zhang, 2017) and other district-level economic characteristics (Milner and Tingley, 2015). This analysis has largely investigated the effect of contextual variables on legislative behavior. However, the literature has not examined micro-level representation in this context by measuring the association between constituency and sub-constituency preferences and the votes of

1Gadarian(2010) similarly ﬁnds that foreign policy positions matter for candidate evaluations, but foreign policy is more salient for the evaluation of Republican candidates.

1 U.S. House members on foreign policy bills. To our knowledge, there are no existing estimates of constituency-level opinion of this kind. The question remains: does responsiveness to constituency preferences extend to foreign policy roll-call votes?2 Moreover, are members of Congress more responsive to co-partisans in their districts on these votes than to the district as a whole? In this paper, we create new estimates of constituency opinion broken down by partisan category on several foreign policy roll-call votes in the U.S. House of Representatives. We collect survey data and use multi-level regression and post-stratification (MRP) to measure constituency preferences on various bills over time. In particular, we look at key pieces of legislation, like the authorization of military intervention in Iraq and the North American Free Trade Agreement (NAFTA). Using these data, we assess whether or not legislators’ roll-call voting corresponds to the preferences of their constituents. In addition, we investigate whether or not there is a greater association between roll-call voting and the preferences of co-partisans in the district. Moreover, we investigate the role of four issue-specific characteristics in the levels of responsiveness on these bills: partisan cleavages, interest group activity, issue salience, and cue-taking. We theorize that there should be less responsiveness on trade bills than on the other bills we examine because trade bills are not very salient and many business and labor groups aggressively lobby legislators on these bills. In addition, we expect to see more responsiveness to copartisans on security and immigration than on trade because there are stronger ideological distinctions between the parties on the former set of issues. We find evidence of responsiveness to the median in all policy areas: legislators are more likely to vote in favor of the bill when there is higher support for it in their constituency. We do find, however, there is substantially less responsiveness on trade than on security or immigration. Moreover, legislators are responsive to their co-partisans on security, and they are more responsive to them than they are to their median voters. On immigration, we find more mixed results. We

2Qualitative evidence suggests that American presidents are responsive to public opinion on foreign policy (Foyle, 1999). Outside the U.S., the strongest direct evidence for responsiveness on military intervention comes from Tomz, Weeks and Yarhi-Milo(2020), which presents experimental evidence that Israeli parliamentarians were more supportive of military force when informed that the public supported it. Furthermore, parliamentarians believed going against public opinion on military intervention would be very politically costly. For trade, Kono(2008) presents evidence of a negative cross-national correlation between public support for trade and tariff barriers in democratic countries.

2 examine the role of cue-taking by looking at cases in which parties have shifted their position (like the Republican Party on immigration, or the Democratic party on the Iraq War) and seeing if there is a resulting increase in copartisan responsiveness; we ﬁnd no evidence of this in our data. Our ﬁndings have important implications for the viability of preference-based approaches to International Relations. Liberal scholars of International Relations have called for “taking preferences seriously” (Moravcsik 1997). A vast literature has emerged to explore the determinants of individual preferences on trade, security and immigration issues, but the crucial link between citizen preferences and foreign policy has not been shown. If citizen preferences on issues like trade are largely ignored by legislators, then preference-based approaches would need substantial revision. Alternatively, evidence of disproportionate responsiveness to co-partisans would indicate that domestic partisan divides are an important force in International Relations (Schultz, 2001).

Related Literature

We are most directly engaged with a literature assessing the constituency-level determinants of foreign policy voting by American legislators. Milner and Tingley(2015) looks at the domestic politics of trade, foreign aid, and military intervention. Higher proportions of active-duty military members in a constituency positively predicts legislative voting for military intervention (Hilde- brandt et al. 2013). American legislators are more likely to vote in favor of trade liberalization if they represent more educated constituencies (Owen, 2017; Milner and Tingley, 2011; Bailey, 2001). District-level vulnerability to offshoring (Owen, 2017) and Chinese import competition (Feigenbaum and Hall, 2015; Kuk, Seligsohn and Zhang, 2017) also predict anti-trade positioning and voting.3 Despite the lack of existing evidence linking individual voter trade preferences to trade policy

3The Open Economy Politics (OEP) approach to International Political Economy views foreign economic policy as a process beginning with citizen preferences, which are then aggregated through political institutions, ultimately resulting in the “output” of a given foreign economic policy. For an overview of the Open Economy Politics approach, see Lake(2009). For examples of this approach in the American context, see Milner and Tingley(2011) and Owen (2017). For a critique of this approach, see Oatley(2011) and the response by Chaudoin, Milner and Pang(2015).

3 outcomes, there is a vast literature exploring the determinants of individual-level public opinion on trade. In keeping with the constituency-level analysis mentioned above, college-educated individuals are more supportive of trade (Scheve and Slaughter, 2001; Hainmueller and Hiscox, 2006), individuals whose jobs are more vulnerable to offshoring exhibit greater support for protectionism (Owen and Johnston, 2017), and individuals living in areas negatively affected by trade shocks respond sociotropically by opposing trade (Bisbee, 2018; Alkon, 2017). We assess responsiveness to both the overall constituency and co-partisan constituents. The literature generally finds evidence that American legislators are responsive to their constituencies (Erikson, MacKuen and Stimson, 2002; Stimson, MacKuen and Erikson, 1995). Moreover, there is evidence that there is an electoral penalty for dissonant behavior by politicians (Ansolabehere and Jones, 2010; Canes-Wrone, Brady and Cogan, 2002), such that those who vote against the preferences of their constituency are punished by the voters. However, voters in the United States did not hold Senators accountable for their votes on the Central American Free Trade Agreement (Guisinger, 2009), possibly because voters find it difficult to link trade policy outcomes directly to their personal economic circumstances (Guisinger, 2017).4 Some evidence, however, exists that members of Congress do not represent the median in their district (Bafumi and Herron, 2010; Lee, Moretti and Butler, 2004; McCarty, Poole and Rosenthal, 2009). In particular, these analyses show that legislators from both parties reside on opposite sides of the median voter in their district; as a result, elections are often a choice between two poles of the ideological spectrum. Moreover, some have argued that there is unequal responsiveness such that policies tend to cater to the affluent and the well-connected (Bartels, 2008; Gilens, 2012), though some have challenged this claim (Soroka and Wlezien, 2008). Our focus on the preferences of sub-constituencies comes from Fenno(1978), who argues that legislators think of their district as comprising different sub-constituencies, which they may be differentially responsive to. In the literature, Kastellec et al.(2015) find that Senators respond more to their co-partisans than to their state’s median voters when it comes to Supreme Court confirmation votes; Lax, Phillips and

4Furthermore, there is no evidence of incumbent adaptation to shifting public opinion (i.e. dynamic responsiveness) in the months leading up to the vote on NAFTA (Lee, Pomirchy and Schonfeld, 2019).

4 Zelizer(2019) confirms this result on a broader range of roll-call votes. Similarly, Clinton(2006) finds that Republicans in the House are disproportionately responsive to co-partisan preferences, and Bafumi and Herron(2010) finds that partisans tend to be closer to their legislators’ ideal points than the district median. In this paper, we contribute to the literature on opinion estimation by looking at issue-level roll-call voting in the House with a partisan breakdown. Some papers have looked at district-level estimates of opinion (Tausanovitch and Warshaw, 2013; Warshaw and Rodden, 2012) and state- level estimates (Lax and Phillips, 2009, 2012), but there have not been any estimates of opinion on foreign policy roll-call votes. Moreover, some have looked at partisan breakdowns using MRP (Kastellec et al., 2015; Lax, Phillips and Zelizer, 2019), but such estimates have not been carried out in House districts. To our knowledge, we are the first to create issue-specific constituency opinion estimates broken down by party in the U.S. House.

Theoretical Expectations

We examine several different bills on immigration, security, and trade across the last three decades. Our theoretical foundations are derived from Fenno(1978), who suggests that there are different sub-constituencies that legislators appeal to to win reelection, one of the most important being co-partisans in the district. As such, there has been quite a lot of prior literature corroborating the notion of co-partisan responsiveness (Butler and Broockman, 2011; Clinton, 2006; Kastellec et al., 2015; Lax, Phillips and Zelizer, 2019). In our context, however, our expectations of whether legislators will respond more to the median of their district or to their co-partisans differ across these issues. In particular, we theorize that four issue-speciﬁc characteristics will affect responsiveness: interest group behavior, issue salience, the extent of partisan cleavages, and voters’ cue-taking ability. We note that these characteristics should not be thought of as separate or mutually exclusive but rather as interactive (e.g., voters have higher cue-taking ability when they observe the positions of their party and the parties’ positions are distinct from each other).

5 Scholars have cited interest group behavior as an inﬂuence on legislative decision-making (Fordham and McKeown, 2003; Grossman and Helpman, 2002). Often, interest groups are per- ceived as altering the direction of responsiveness away from the majority of the public (Anzia, 2011; Gilens, 2012). In particular, studies pinpoint the disproportionate inﬂuence of interest groups on trade legislation (Ehrlich, 2008). For instance, Karol(2007) cites the AFL-CIO labor union’s shift on trade policy in 1970 as the primary catalyst for the Democratic party’s shift from general support for trade to more protectionist policy positions. Given this, it is likely that on bills where there is a lot of interest group activity, like trade bills, we would expect less responsiveness to district opinion than on other issues. One other dimension that these issues vary on is the existence of clear partisan cleavages among members of Congress. Some issues are far more politicized than others and spark clearer divides between Republicans and Democrats. This politicization could have multiple sources. It could be the case that interest groups that often align with one party (like the NRA, for instance) mobilize their base to support a bill. Alternatively, the sitting President might support a bill and unify the other party in opposition (Lee, 2008, 2016). On security and trade issues, there is likely to be much more deference to the President, given the asymmetry in information and resources (Canes-Wrone, Howell and Lewis, 2008; Wildavsky, 1966). In some cases, this might be complicated when the President or party leaders take a position in contrast with the rest of the party. One clear example of this is President Bill Clinton championing NAFTA in 1993. Since the Democratic Party was mostly in favor of protectionism, this signal of Democratic support for a free trade agreement likely sparked intra-party disagreement within both parties. Thus, on issues where the partisan distinctions are fuzzy (like on NAFTA), particularly in cases where the position of the President or the legislator’s party clashes with the position of co-partisans in the district, there should be less responsiveness to co-partisans. The reverse should be true as well; on issues where there are clear partisan cleavages, we should expect legislators to be more responsive to their co-partisans. Moreover, the salience of the issue is relevant in a couple of ways: politicians may perceive the consequences of deviating from constituency preferences to be more dire when legislating

6 on issues that are important to the constituency (Arnold, 1990), and voters may understand the positions of the two parties better on such issues (Hutchings, 2001). Indeed, it has been shown that legislators are more likely to be thrown out of office due to discongruent votes on more salient issues (Highton, 2019; Hutchings, 1998).5 If we view the accountability mechanism on high- salience issues as valid, then it is the low-salience bills (and in particular those in which there is high interest group activity) that we should expect the least responsiveness to the median. One can also acknowledge that in measuring responsiveness, the direction of the relationship might be reversed, such that voters take cues from their representatives and change their views accordingly. In particular, this can be a consequence of higher issue salience. Much evidence suggests that voters take cues from politicians that they agree with (Lenz, 2012; Trager and Vavreck, 2011), and they rely on these cues for information (Mondak, 1993). Zaller(1992) showed, using the Vietnam War among other examples, that individuals are most capable of taking cues when the parties are sufficiently far apart on the issue; others have provided additional support for this (Hetherington, 2001; Levendusky, 2009). However, voters by and large do not perceive the parties as having distinct ”types” on foreign policy issues (Kertzer, Brooks and Brooks, 2017). Thus, on issues where Republicans and Democrats hold similar positions (e.g., the Iraq War when it was first authorized in 2002, the North American Free Trade Agreement in 1993),6 voters are less able to take cues on these issues. However, where the parties are far apart ideologically speaking (like on the bill to de-fund sanctuary cities in 2017) and the voters observe the position of the two parties, it should be easier for voters to follow the party or politician they favor and adopt their position (Levendusky, 2009). Thus, using this logic, on issues that are salient and where the parties di- verge ideologically, we would expect the association between co-partisan opinion and legislative roll-call voting to be higher. When there is high cue-taking ability, co-partisans can mirror the preferences of their representative and their party, and we should see higher levels of co-partisan

5However, there exists some research that suggests there is no evidence politicians converge to the middle when the issues are important to their constituencies (Fowler and Hall, 2016). 6Many prominent Democrats, like Senator Hillary Clinton and Senator Joe Biden, were in favor of the authorization of the War in Iraq, and the ﬁnal vote was fairly bipartisan. Moreover, President Bill Clinton supported NAFTA, and the trade agreement had also garnered support among Republicans, who traditionally support free trade.

7 responsiveness in this case.

Empirical Strategy

We ﬁrst test whether or not legislators are responsive to the median of their constituency. We do this by measuring the association between constituency preferences and legislators’ roll-call voting on each bill that we examine. Secondly, we test whether legislators are more responsive to their co-partisan base in their district than to the district as a whole. We do this by comparing the effect of overall constituency preferences on roll-call voting with the corresponding effect of the preferences of co-partisans in the district.

Let Yi be legislator i’s roll call vote, which is equal to 1 when legislator i votes yes on a bill and

equal to 0 when legislator i votes no. Let γb denote the bill-speciﬁc ﬁxed effect. We estimate the following model to test the association between overall constituency preferences and legislators’ roll-call voting:

−1 Pr(Yi = 1) = logit (β0 + β1ProportionConstituencySupport + γb) (1)

In equation (1), if there is responsiveness to the median, we would observe β1 > 0. We estimate the following model to test differential responsiveness to co-partisan preferences, controlling for the size of the copartisan and out-partisan populations in each district.

−1 Pr(Yi = 1) = logit (β0 +β1ProportionConstituencySupport +β2ProportionCopartisanSupport+

β3CopartisanPopulation + β4Out partisanPopulation + γb) (2)

β2 > β1 in equation 2 suggests that co-partisan support has a greater effect on roll-call voting than the support of the district median.

8 Data and Research Design

Table 1 describes the survey data we gathered to assess public opinion. When gathering data on public opinion, we searched for surveys taken before the corresponding House vote. Furthermore, we restricted our analysis to those surveys/questions that either speciﬁcally asked about the bill that was being voted on or addressed the key component of the bill. Finally, we limited our search to surveys that had key demographic characteristics and state-level or district-level indicators. When- ever possible, we pooled multiple surveys together to attain more precise estimates of constituency (and sub-constituency) preferences. We chose particular trade, security and immigration bills because detailed survey data existed on the passage of these bills in Congress and because they were highly consequential foreign policy bills with important implications for the U.S. economy and international security. Trade bills include NAFTA, which liberalized trade between the United States, Mexico and Canada, and the Uruguay Round Agreements Act that established the World Trade Organization (WTO). In addition, H.R.4444 normalized trade relations between the U.S. and the People’s Republic of China, and the Central American Free Trade Agreement (CAFTA) liberalized trade between the United States and Central American nations (and the Dominican Republic). Finally, the U.S.- Korea Free Trade Agreement eliminated 95 percent of bilateral tariffs on goods and strengthened intellectual property protections (Erickson, September 2, 2017). When it comes to security policy, we study two bills related to military intervention in Iraq: the Authorization for Use of Military Force Against Iraq Resolution of 2002, and the Responsible Redeployment from Iraq Act in July 2007 which called on the Secretary of Defense to commence withdrawals of U.S. troops in Iraq within 120 days of passage of the bill. We also analyze voting on H.J.Res.68 which authorized military intervention in Libya in support of the NATO mission to implement U.N. Security Council Resolution 1973. The resolution aimed to protect civilians and target the Gaddaﬁ regime during the Libyan civil war.7 Finally, the immigration bills we examine are HR.4134 (legislation to deny public education

7Libya: UN backs action against Colonel Gaddaﬁ (March 18, 2011).

9 beneﬁts to undocumented immigrants), the Secure Fence Act of 2006 which provided for the construction of fencing along 700 miles of the U.S. border with Mexico, and the No Sanctuary for Criminals Act, which would have penalized “sanctuary cities” (i.e., municipalities not complying with federal immigration law) by cutting off their federal funding. In the Appendix, we document the surveys we use to generate our public opinion estimates.

Table 1: Bills Bill Day of Vote Outcome of Vote NAFTA November 17, 1993 234-200 Uruguay Round Agreements Act (WTO) November 29, 1994 288-146 Denying Education Beneﬁts to Undocumented September 25, 1996 254-175 Normalization of Trade Relations with China May 24, 2000 237-197 Iraq War Authorization October 10, 2002 296-133 CAFTA-DR July 28, 2005 217-215 Secure Fence Act September 14, 2006 283-138 Redeployment from Iraq July 12, 2007 223-201 Authorization of Force in Libya June 24, 2011 123-291 US-Korea Free Trade Agreement Oct 12, 2011 278-151 No Sanctuary for Criminals Act June 29, 2017 228-195

One caveat that we note here is that our overall selection of bills may bias the estimates of responsiveness since by looking at bills that have been asked about in polls, our analysis largely focuses on relatively salient bills. As we mentioned before, some evidence suggests that on bills that are more salient, politicians are more likely to cater to the median (Hutchings, 1998). Our two main results are that there is less responsiveness to the median on trade, and there is more responsiveness to co-partisans on security than there is to the median. We would arguably expect the ﬁrst ﬁnding in a broader universe of bills that included less-salient bills as well. Moreover, with respect to the second result, our estimates might be biased upwards if it was due to voters’ cue-taking; if it is based on the politicization of the issue, then issue salience likely does not bias our estimates. Later in the paper, we make the case that cue-taking is likely not the primary driver of our results.

10 Opinion Estimation

To generate our estimates of public opinion, we use multi-level regression and post-stratification (MRP). MRP proceeds in two stages. In the multi-level regression stage, we model individual survey responses as a function of geographic and demographic factors. We use gender, race, education, and party identification as individual-level predictors and use the location of respondents to estimate district-level intercepts, which are themselves modeled using predictors like region or aggregate demographics, percent senior, and percent employed in agriculture. In the post- stratification stage, we calculate predicted probabilities from the multi-level regression for each demographic-geographic type and weight these types based on their levels in each district.8 In constructing opinion estimates by partisan group, one data barrier to overcome is that the Census does not have information on party affiliation. Therefore, one has to discern the breakdown of partisanship by our various demographic and geographic variables before proceeding to the MRP. In particular, we can extract the number of white men between the ages of 18 and 29 living in the 12th district of New Jersey, but in order to have estimates by partisan group, we need also the number of white Republican men between the ages of 18 and 29 living in the 12th district of New Jersey (in addition to the analogous estimates for Democrats and Independents). To do this, we estimate a second MRP where party affiliation is the dependent variable. We first estimate the probability of identifying as a Democrat on our slate of independent variables. Then, we throw out the Democrats and estimate the probability of being a Republican (where the baseline is being an Independent) and deduce the proportion that fall into all three partisan categories using these two regressions.9 Since we are looking at bills across three decades and congressional districts do not look the same after redistricting, we estimate the breakdown of partisan groups by pooling

8This data comes from American FactFinder. Factfinder includes data on breakdowns of race, education, and gender by district for those over the age of 25, which is useful for post-stratification. Using these data is consistent with past research that estimates constituency opinion at the district level (Tausanovitch and Warshaw, 2013; Warshaw and Rodden, 2012). While it introduces some error into the estimates, only a small percentage of the population falls in the remaining age category, and the demographic characteristics are similar between the age groups. 9In order to ease concerns that starting our regression with affiliating as a Democrat as our dependent variable may affect results, we conduct the same analysis, with Republicans as our starting dependent variable and average the estimates that arise from the two approaches.

11 relevant survey data covering the decade of interest. When post-stratifying, we use Census data corresponding to the year of each particular roll-call vote (or the closest year in which data is available), such that we have estimates of partisan breakdowns for each relevant year. One other difficulty in estimating preferences at the district level is that some surveys do not contain district identifiers. It is sometimes not possible to figure out which district the respondent belongs to. To ameliorate this, we adopt an existing method called “cross-level” regression (Krim- mel, Lax and Phillips, 2016), where we use state-level estimates in the multi-level regression and district-level values to post-stratify.10 For example, one district-level predictor we use is median income. Since we do not know the district that a particular respondent belongs to, we instead use median income for the state that the respondent belongs to instead in the multi-level regression. When post-stratifying, however, we use the coefficient for median income from the regression and district-level median income. Thus, we are modeling the geographic variables at the state level but using district values to extrapolate from the geographic patterns in the data to all districts. One can verify that there is not a substantial difference in using state-level data versus district-level data by exploiting the surveys that contain both state-level and district-level indicators. For these surveys, one can compare the estimates that result from a cross-level MRP with those that result from the original implementation. In the Appendix, we plot these estimates for the surveys/bills that have both indicators and show that the cross-level estimates are highly correlated with the original estimates (r > 0.95). This lends confidence to us that the cross-level method we employ is unlikely to produce substantially different estimates across the bills we examine. For each roll call vote that we analyze, we regress support for the policy on several individual- level and district-level (or state-level) predictors. Denote the value of the dependent variable by

Yi for a given individual i. This value is either 1 if the individual supports the policy or 0 if the individual opposes the policy.11 The individual-level predictors are race (“White,” “Black,” “Hispanic,” and “Other”), education (“No HS,” “High school graduate,” “Some college,” “College

10We use this method for all bills except the No Sanctuary for Criminals Act and the Responsible Redeployment from Iraq Act. Moreover, we use this method to obtain estimates of party afﬁliation for earlier years in our timeframe. 11Respondents who abstained are counted as missing.

12 graduate,” and “Post-grad”), and gender (“Female” and “Male”). The district-level predictors are the percentage of individuals in a district that are senior, median income, the percentage of workers in a district that are employed in agriculture, and the percentage of the district that was born in another country. We include the percentage of senior individuals in a district and median income because age and income are often conceived as being strong predictors of political preferences. Moreover, the percent of agriculture workers is included because occupation predicts political preferences, particularly for trade bills. Finally, immigration status is also predictive of political preferences, and it nicely helps predict preferences on immigration in particular. Formally, we use the following speciﬁcation:

−1 0 f emale race educ party district poll Pr(yi = 1) = logit (β + β ∗ f emalei + αk[i] + αl[i] + αn[i] + α j[i] + αp[i] ) where k denotes the category of race that respondent i falls into, l denotes the category of education i belongs to, n denotes the party i belongs to, j denotes the district that i resides in, and p denotes the poll that i is responding to.12 The district intercepts are modeled as a function of district-level predictors:

district state med.income senior.prop α j ∼ N(αm[ j] + β ∗ med.income j + β ∗ senior.prop j+

agriculture.prop f oreign.prop 2 β ∗ agriculture.prop j + β ∗ f oreign.prop j,σdistrict) with district j located in state m[ j].13 To clarify, the variance of the district coefﬁcient is constant across all districts. Furthermore, the following individual-level and district-level coefﬁcients are

12The poll variable is only included when there are multiple polls being pooled together. 13For some issues, the region variable is included in the speciﬁcation as well.

13 modeled as follows:

race 2 αk ∼ N(0,σrace) for k = 1,...,4

educ 2 αl ∼ N(0,σeduc) for l = 1,...,5

party 2 αn ∼ N(0,σparty) for n = 1,...,3

poll 2 αp ∼ N(0,σpoll) for p ∈ R+

state 2 αm ∼ N(0,σstate) for m = 1,...,51

The state variable includes all 50 states plus the District of Columbia. Using these results, we calculated the predicted probability of supporting the policy for each demographic-geographic type and used Census data to post-stratify. Given 436 districts (435 U.S. House districts plus the District of Columbia), 2 gender categories, 4 race groups, 5 education groups, and 3 parties, we have 436 ∗ 2 ∗ 4 ∗ 5 ∗ 3 = 52,320 demographic-geographic types. Using the model estimated above for respondent preferences, we calculated predicted probabilities for each of these 52,320 categories.14 We weight these probabilities by the recorded population level listed in the Census. Thus, if d denotes a particular congressional district, θˆ j is the predicted probability for a given cell j, Nj is the Census population size for cell j, andy ˆd is the proportion of individuals supporting a given policy for district d, then

ˆ ∑ j∈d Njθ j yˆd = ∑ j∈d Nj

Descriptive Statistics: Sub-Constituency Preferences

Figure1 shows the distribution of constituency support for pro-trade bills among Republicans, Independents and Democrats at the mass level. The boxes show the middle 50 percent of con-

14For the poll coefﬁcients, we take the average of the intercepts, which mathematically is equal to 0.

14 stituency support for the bill for each sub-group (i.e. the 25th percentile to the 75th percentile, with the median district in the middle), and the whiskers extend to the most extreme observations. One striking aspect of this ﬁgure is that there is not much heterogeneity between partisans on many of the trade bills. Republicans, Democrats, and Independents are largely consistent in their levels of support for these bills. We can still point out some trends for the bill to liberalize trade relations with China, the establishment of the World Trade Organization (WTO), and the trade agreement between the U.S. and South Korea. Though Republicans were generally more supportive of free-trade legislation, they were less supportive of trade liberalization with China than Independents or Democrats. Republicans were more supportive than Democrats of the establishment of the WTO. In general, there is substantial cross-district variation in levels of support for these bills. Interestingly, this suggests that there are some constituencies where Republicans are not supportive of these trade bills and some constituencies where Democrats are supportive.

Figure 1: Sub-constituency Preferences on Trade (by Bill)

Figure2 shows the distribution of preferences among partisan sub-groups for security bills. Unlike for the trade bills, there is substantial variation in preferences between partisan sub-groups. As conventional wisdom would dictate, Republicans were signiﬁcantly more supportive of for-

15 eign military intervention in Iraq. However, intervention in Libya garnered greater Democratic support than Republican support. As expected, Democrats were substantially more supportive of withdrawing from Iraq than Republicans; while Republicans exhibit differential levels of support across districts, Democrats were uniformly in support of the redeployment of troops from Iraq.

Figure 2: Sub-constituency Preferences on Security (by Bill)

Finally, Figure3 shows preferences on three restrictionist immigration bills. Republicans were more supportive than Democrats of all three bills, and the divide is clearest on the issue of defunding sanctuary cities. Moreover, there was substantial cross-district heterogeneity in Republican support for the bill limiting education beneﬁts, but virtually no geographic heterogeneity in Re- publican support for the No Sanctuary for Criminals Act. Democrats and Independents exhibited greater heterogeneity in their preferences for defunding sanctuary cities, and both Republicans and Democrats exhibited substantial cross-district heterogeneity in support for the Secure Fence Act.

16 Figure 3: Sub-constituency Preferences on Immigration (by Bill)

Responsiveness

Table2 examines whether higher levels of overall support for a bill in a constituency increase the likelihood of a legislator voting in favor of the bill. The table examines bills pooled by our three issue areas: trade, security, and immigration. We find that in all issue areas, there is a positive and statistically significant relationship between constituency opinion and legislative roll-call voting, a finding which is compatible with much of the literature (Erikson, MacKuen and Stimson, 2002; Lax and Phillips, 2012). However, the relationship is noticeably weaker among trade bills than for the other issue areas. Table3 shows whether or not there is greater responsiveness to a House member’s copartisans in their district than to the median constituent. We find strong evidence that there is disproportionate responsiveness towards co-partisan constituents for security and immigration policy, but not for trade policy. However, on immigration, as we will see in the next section, this result is primarily driven by the bill that denies public education benefits to undocumented immigrants. On

17 Table 2: Responsiveness to Overall Constituency

Dependent variable: Vote in Favor Trade Security Immigration Overall 2.551∗∗∗ 19.849∗∗∗ 14.028∗∗∗ (0.637) (1.545) (1.128) Observations 2,162 1,271 1,274 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 the other two bills, we ﬁnd that there is more responsiveness to the median constituent than to a House member’s co-partisans. One should exercise some caution here with respect to the speciﬁcations including both overall support and co-partisan support, as there might be issues of collinearity (Lax, Phillips and Zelizer, 2019). To address this, we complement our analyses by looking at the number of times that a member of Congress takes the side of the median over co-partisans. One empirical con- cern might be that the median voter in a constituency is in some cases a co-partisan; Republicans likely control constituencies where the median is a Republican, and the analogous may be true for Democratic legislators. Examining cases in which the co-partisan median and the constituency median disagree, we see that legislators side with co-partisans more than 92 percent of the time on immigration (362 times in 386 cases), with Republicans displaying (slightly) disproportionate responsiveness to co-partisans.15

Table 3: Responsiveness to Overall and Co-partisan Constituencies

Dependent variable: Vote in Favor Trade Security Immigration Overall 20.121∗∗∗ 1.198 2.888 (3.672) (2.200) (2.315) Co-partisan −18.105∗∗∗ 12.276∗∗∗ 5.668∗∗∗ (3.559) (0.844) (1.269) Observations 2,162 1,271 1,274 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

15Republicans side with co-partisans in 321 cases and only side with the median voter in 21 cases; for Democrats, it is 35 and 3 respectively.

18 Across the five trade bills, there are 141 cases of disagreement–in 78 of these cases (55 percent), legislators side with the median voter. However, disaggregating by party, we find that this result is driven by Democratic legislators, who side with the median almost two-thirds of the time; Republicans are about equally likely to favor their co-partisans as they are to favor the median voter. On security issues, legislators side with their co-partisan constituents almost 80 percent of the time (371 out of 476 cases). This disproportionate responsiveness to co-partisans is mostly driven by Republicans, who only side with the median voter five times in 216 cases of disagreement (Democrats side with co-partisans on security just over 60 percent of the time). In all three issue areas, Republican members of the House are more likely than Democratic legislators to side with their co-partisan constituents over their district’s median voter.16

Assessing Responsiveness By Bill

We now assess responsiveness by bill. We first examine legislator responsiveness to constituents on five trade bills: the North American Free Trade Agreement (NAFTA), a bill creating the World Trade Organization, trade liberalization with China, and a trade agreement between the US and South Korea. In the pooled trade analysis, we uncovered a positive and significant relationship between overall support in a constituency for a trade bill and the likelihood of a legislator voting to enact that trade bill. The disaggregated analysis shown in Figure4 reveals that this relationship emerges from NAFTA, CAFTA-DR, and the WTO. Table A1 examines co-partisan responsiveness separately for each bill–for all trade bills except the agreement with South Korea, there is more responsiveness to the median than to co-partisans. The bill liberalizing trade with China is somewhat anomalous among the bills we examine here. On the Democratic side, the majority of the House Democrats had voted against the bill, despite the fact that President Bill Clinton supported the bill and ultimately signed it into law. Hence, this

16This may result from misperception of public opinion on the part of Republican legislators; Broockman and Skovron(2018) ﬁnd that Republican politicians perceive their constituencies as being more conservative than they are.

19 is why we see evidence of anti-responsiveness here. Moreover, we see evidence of a disconnect between Republicans in the mass public and Republican legislators on the bill liberalizing trade with China. Republican voters were deeply skeptical of liberalizing trade with China, even though Republican legislators endorsed the bill. Three-fourths of Republicans voted in favor of the bill, but less than 50 percent of Republicans supported the bill in the median congressional district (as shown in Figure1). In Figure5 (which includes the authorization of military force against Iraq in 2002, a bill that would have initiated troop withdrawals from Iraq in 2007, and the authorization of limited force in Libya), we find that for every security bill, there is a strong relationship between constituency opinion and roll-call voting. However, Table A2 provides evidence that representation is biased towards co-partisans for each bill. There is more responsiveness to co-partisans than there is to the median on each of the security bills. Finally, we assess responsiveness on three immigration bills: a bill denying public education benefits to undocumented immigrants, a bill providing for the construction of fencing along the US-Mexico border (the Secure Fence Act of 2006), and the No Sanctuary for Criminals Act, which would have penalized municipalities that were not compliant with federal immigration law (i.e., sanctuary cities). As shown in Figure6, we find evidence of responsiveness on the sanctuary cities and education bills, but not for the Secure Fence Act. We find evidence for disproportionate responsiveness to copartisans in Table A3 for the bill denying education benefits to undocumented immigrants, but not for the sanctuary cities bill or the Secure Fence Act.

20 Figure 4: Responsiveness to Overall Constituency on Trade Bills

21 Figure 5: Responsiveness to Overall Constituency on Security Bills

22 Figure 6: Responsiveness to Overall Constituency on Immigration Bills

It should be noted here that the research design employed in this analysis is descriptive. It is useful, however, to discuss some explanations for why we see these results, drawing on the factors that we mentioned above: politicization of the issue, the role of interest groups, issue salience, and cue-taking. Here, we take a look at the role of each of these factors in determining our bill-speciﬁc

23 results. One of our main findings is that on trade bills, there is a lower level of responsiveness than on security and immigration bills. As mentioned above, on issues that are less salient and that attract high levels of interest group activity, we should see less responsiveness. For example, the South Korea trade agreement was not very salient to voters, but like most trade legislation, it attracted interest group activity, particularly among business and labor groups. The number of newspaper articles about the trade agreement with South Korea was approximately four times smaller than the number of articles on the Iraq War, for instance.17 Moreover, the number of lobbying reports on the trade agreement (260) is much higher than the Iraq war authorization (0) or the redeployment bill (2).18 It is clear that in the context of many trade agreements, interest groups with a vested interest in economic outcomes lobby members of Congress and may be distorting roll-call voting away from constituency preferences towards their own. This is consistent with some existing evidence in the literature (Fordham and McKeown, 2003). This dynamic is also apparent in the other trade bills–while there is a positive and statistically significant relationship between constituency opinion and roll-call voting on the trade bills, the size of the relationship for CAFTA and the WTO is relatively small (compared to the security and immigration bills). On the flip side, one can note that the more salient trade bills are the ones where we see greater levels of responsiveness. In particular, NAFTA was a highly salient issue when it passed, and the level of responsiveness on that bill is higher than on the other bills we examine. Similarly, the votes on the authorization of the War in Iraq and the 2007 vote were also very salient, and both votes yield evidence of high responsiveness. Moreover, one other key aspect of the trade bills we examine here is that the extent of partisan cleavages is much smaller. Trade bills are less “politicized” than other bills. This can be demonstrated by noting the lack of partisan splits in the roll calls for these bills and how closely tied the preferences of Republicans and Democrats are in our opinion estimates. In particular,

17This was calculated using a search of LexisNexis using the keywords “Korea” and “trade” for the South Korea trade agreement and keywords “Iraq” and “war.” 18The number of lobbying reports was taken from the Lobbyview database.

24 the votes on NAFTA and the WTO were not particularly partisan, arguably because a Democratic President had supported these pieces of free-trade legislation, which ran counter to the traditionally accepted views on trade in the Democratic Party. On the WTO, for instance, 163 Republicans voted in favor and 86 voted against, and 125 Democrats voted in favor and 59 against.19 Moreover, on these issues, as one can see in Figure1, the divergence between Republicans, Independents, and Democrats at the mass level is not very large either, as compared to the other issues that we examine. Like we noted in our Theoretical Expectations section, this suggests that we would see less disproportionate responsiveness to copartisans versus the median, which is what we see in the data. On most of the security and immigration issues, we have demonstrated evidence of a link between co-partisan constituent preferences and legislator voting. However, it is possible this evidence of responsiveness actually reflects reverse causation; co-partisan constituents might adopt the positions of their party and their representatives (Achen and Bartels, 2016; Lenz, 2012). In other words, there could be a greater association between the copartisan preferences and legislative voting because copartisans are taking cues from the representatives that they affiliate with. At first glance, this seems quite likely given that we have included contain many highly salient bills, like the authorization of war in Iraq in 2002 and NAFTA. Furthermore, the graph of constituency preferences shows greater divergence between Republicans and Democrats on more recent immigration and security bills (where there was greater elite polarization) than on the earlier bills we examine. However, while voters may be taking cues from their representatives on these bills, this might not be the dominant force underlying our results on co-partisan responsiveness; in particular, one can examine issues where the parties’ positions have drifted apart over time and see whether or not there is a corresponding shift in co-partisan responsiveness. The War in Iraq is a nice case in point. Republicans and Democrats in Congress did not have distinct positions at first, but the parties gradually grew apart over the course of the Bush Administration. In 2002, many Democrats voted

19On NAFTA, 102 Republicans voted in favor and 156 voted against, whereas 133 Democrats voted in favor and 85 voted against.

25 in favor of the authorization, and many prominent Democrats, like then Senator Hillary Clinton and Senator Joe Biden, had voiced support for the war. In the wake of 9/11, there was strong pressure from the public for meaningful government action, and public approval for President Bush was very high. As a result, Democrats and Republicans were largely in consensus in approving resolutions on the wars in Iraq and Afghanistan. As time went on, Democrats in Congress became less sanguine about the war effort, especially in the wake of revelations regarding the absence of weapons of mass destruction. In particular, the 2006 elections, after which Speaker Nancy Pelosi took the majority in the House, was largely seen as a referendum on the War in Iraq. By this time, elite Democrats were opposed to the war, and Republicans were still in support. We have opinion estimates on the authorization of the War in Iraq in 2002, when elites were, for the most part, uniﬁed, and a 2007 vote to redeploy troops to Iraq. Theoretically speaking, we should see much more co-partisan responsiveness in the latter case because Republicans and Democrats at the elite level had more readily discernable positions on the War in Iraq and co-partisans were better able to correctly sort themselves (Levendusky, 2009). However, as one can see in Table A2, there is no distinguishable increase in relative co-partisan responsiveness (if anything, there is more disproportionate co-partisan responsiveness in the 2002 vote than there is in the 2007 vote). There is largely no substantive difference in our ﬁndings between these two votes; in both cases, there is a larger association between copartisan opinion and voting than there is between overall support and voting. Another interesting set of issues we can examine in this setting to further recover evidence of cue-taking is immigration. There have been substantial changes in the Republican Party’s position on immigration over time. In particular, one can note that President Ronald Reagan and President George W. Bush were largely in favor of (and in the former case, presided over) increases in immigration. As late as 2013, the Gang of Eight in the Senate, which included Republicans like Senator Marco Rubio, tried to pass a bill through Congress that included a pathway to citizenship for many undocumented immigrants. However, the Republican Party, led by President Donald Trump, now takes a substantially more hard-line position on immigration. In particular, the President supports

26 deporting mass number of undocumented immigrants, building a wall on the southern border, and restricting funding to sanctuary cities. We have also constructed opinion estimates on immigration bills across three decades, including a bill that was proposed after the election of President Donald Trump that strips federal funding from sanctuary cities. This allows us to see if there is variation in co-partisan responsiveness over time. Theoretically speaking, on the more recent bills, we should see more co-partisan responsiveness since there is a sharper distinction between the two parties on immigration in the wake of Trump’s election than before. While it is evident from our constituency opinion estimates, vi- sualized in Figure3, that Republicans and Democrats hold more divergent opinions over time on immigration, co-partisan responsiveness has not increased over time. On the 2017 bill, members of Congress are far more responsive to the median to their co-partisans, and in 1996, they are more responsive to their co-partisans than to the median. Thus, our results on immigration suggest that cue-taking is likely not the primary driving force behind our results on co-partisan responsiveness. One last check that we do is to assess the relationship between legislative voting and the preferences of those co-partisans who are arguably least likely to know their representative’s position on an issue: those who are not college-educated. While this is by no means a deﬁnitive test of whether or not we observe the mass public taking after the legislators (and not vice versa), this is a reasonable test given that college-educated Americans demonstrate more civic engagement and pay more attention to politics (Prior 2018; Prior and Bougher 2018). Because less educated citizens are unlikely to know how their representative plans to vote on a bill, a relationship between the preferences of less educated co-partisans and legislator voting on foreign policy would provide some additional evidence that voting is responsive to (co-partisan) constituency preferences, rather than the other way around. In Table A4, we demonstrate that we obtain the same ﬁndings as before using less-educated co-partisans instead of the whole population of co-partisans.

27 Conclusion

In this paper, we provide the first empirical estimates of constituency-level and subconstituency- level public opinion on issues of trade, security, and immigration policy in the House. Using these estimates, we examine responsiveness to the constituents of members of Congress and find that legislators exhibit responsiveness to the median on trade bills but primarily respond to co-partisan constituents on security issues (with mixed evidence for immigration). Liberal scholars of international relations had called for “taking preferences seriously” but had not definitively linked foreign policy to voter preferences. Our findings suggest that preference- based accounts of foreign policy-making are indeed based on solid foundations, though our finding that legislators are responsive to co-partisans suggests that partisanship is an under-explored factor in the International Relations literature. With regards to the study of responsiveness in the United States, our results provide further support for the notion that representation in Congress is biased away from the median voter towards co-partisans, at least for security and immigration issues. Finally, our finding of less responsiveness to the median on trade is broadly consistent with a consensus in the literature, in which “the less demanding voice of the American public has been depicted as relatively muted against the intense interests of organized special-interest groups.”20 However, we find evidence suggesting that voter preferences matter for foreign policy issues– according to our analysis, responsiveness does not stop at the water’s edge.

References

Achen, Christopher H and Larry M Bartels. 2016. Democracy for realists: Why elections do not produce responsive government. Vol. 4 Princeton University Press. Alkon, Meir. 2017. “Local Sociotropism: How Community Variation in Trade Exposure Affects Voter Demands.” Available at SSRN 3103852 . Almond, Gabriel Abraham. 1950. “The American people and foreign policy.”.

20Guisinger(2017), page 6

28 Ansolabehere, Stephen and Philip Edward Jones. 2010. “Constituents Responses to Congressional Roll-Call Voting.” American Journal of Political Science 54(3):583–597. Anzia, Sarah F. 2011. “Election Timing and the Electoral Influence of Interest Groups.” The Journal of Politics 73(2):412–427. Arnold, R Douglas. 1990. The logic of congressional action. Yale University Press. Bafumi, Joseph and Michael C Herron. 2010. “Leapfrog representation and extremism: A study of American voters and their members in Congress.” American Political Science Review 104(3):519–542. Bailey, Michael. 2001. “Quiet influence: The representation of diffuse interests on trade policy, 1983-94.” Legislative Studies Quarterly pp. 45–80. Bartels, Larry M. 1991. “Constituency opinion and congressional policy making: The Reagan defense buildup.” American Political Science Review 85(2):457–474. Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age. Stu - student edition ed. Princeton University Press. Berinsky, Adam J. 2007. “Assuming the costs of war: Events, elites, and American public support for military conflict.” The Journal of Politics 69(4):975–997. Bisbee, James. 2018. “What Is Out Your Back Door: How Policy Preferences Respond to Local Trade Shocks.”. Broockman, David E and Christopher Skovron. 2018. “Bias in perceptions of public opinion among political elites.” American Political Science Review 112(3):542–563. Busby, Joshua W and Jonathan Monten. 2012. “Republican elites and foreign policy attitudes.” Political Science Quarterly 127(1):105–142. Butler, Daniel M. and David E. Broockman. 2011. “Do Politicians Racially Discriminate Against Constituents? A Field Experiment on State Legislators.” American Journal of Political Science 55(3):463–477. Canes-Wrone, Brandice. 2015. “From Mass Preferences to Policy.” Annual Review of Political Science 18(1):147–165.

29 Canes-Wrone, Brandice, David W. Brady and John F. Cogan. 2002. “Out of Step, out of Ofﬁce: Electoral Accountability and House Members’ Voting.” The American Political Science Review 96(1):127–140. Canes-Wrone, Brandice, William G. Howell and David E. Lewis. 2008. “Toward a Broader Under- standing of Presidential Power: A Reevaluation of the Two Presidencies Thesis.” The Journal of Politics 70(1):1–16. Chaudoin, Stephen, Helen V Milner and Xun Pang. 2015. “International systems and domestic politics: Linking complex interactions with empirical models in international relations.” Inter- national Organization 69(2):275–309. Clinton, Joshua D. 2006. “Representation in Congress: constituents and roll calls in the 106th House.” The Journal of Politics 68(2):397–409. Ehrlich, Sean D. 2008. “The tariff and the lobbyist: political institutions, interest group politics, and US trade policy.” International Studies Quarterly 52(2):427–445. Erickson, Amanda. September 2, 2017. “A quick guide to the South Korea free trade deal.” Wash- ington Post . Erikson, Robert S, Michael B MacKuen and James A Stimson. 2002. The macro polity. Cambridge University Press. Feigenbaum, James J and Andrew B Hall. 2015. “How legislators respond to localized economic shocks: Evidence from chinese import competition.” The Journal of Politics 77(4):1012–1030. Fenno, Richard F. 1978. Home style: House members in their districts. HarperCollins,. Fordham, Benjamin O and Timothy J McKeown. 2003. “Selection and inﬂuence: Interest groups and congressional voting on trade policy.” International Organization 57(3):519–549. Fowler, Anthony and Andrew B. Hall. 2016. “The Elusive Quest for Convergence.” Quarterly Journal of Political Science 11:131–149. Foyle, Douglas C. 1999. Counting the public in: Presidents, public opinion, and foreign policy. Columbia University Press. Gadarian, Shana Kushner. 2010. “Foreign policy at the ballot box: How citizens use foreign policy

30 to judge and choose candidates.” The Journal of Politics 72(4):1046–1062. Gilens, Martin. 2012. Affluence and Influence: Economic Inequality and Political Power in Amer- ica. Princeton University Press. Grossman, Gene M and Elhanan Helpman. 2002. Interest groups and trade policy. Princeton University Press. Guisinger, Alexandra. 2009. “Determining trade policy: Do voters hold politicians accountable?” International Organization 63(3):533–557. Guisinger, Alexandra. 2017. American opinion on trade: Preferences without politics. Oxford University Press. Hainmueller, Jens and Michael J Hiscox. 2006. “Learning to love globalization: Education and individual attitudes toward international trade.” International Organization 60(2):469–498. Hetherington, Marc J. 2001. “Resurgent mass partisanship: The role of elite polarization.” Ameri- can Political Science Review 95(3):619–631. Highton, Benjamin. 2019. “issue accountability in us House elections.” Political Behavior 41(2):349–367. Hildebrandt, Timothy, Courtney Hillebrecht, Peter M Holm and Jon Pevehouse. 2013. “The domestic politics of humanitarian intervention: Public opinion, partisanship, and ideology.” For- eign Policy Analysis 9(3):243–266. Hutchings, Vincent L. 1998. “Issue Salience and Support for Civil Rights Legislation among Southern Democrats.” Legislative Studies Quarterly 23(4):521–544. Hutchings, Vincent L. 2001. “Political context, issue salience, and selective attentiveness: Con- stituent knowledge of the Clarence Thomas confirmation vote.” Journal of Politics 63(3):846– 868. Karol, David. 2007. “Does constituency size affect elected officials’ trade policy preferences?” The Journal of Politics 69(2):483–494. Kastellec, Jonathan P, Jeffrey R Lax, Michael Malecki and Justin H Phillips. 2015. “Polarizing the electoral connection: partisan representation in Supreme Court confirmation politics.” The

31 journal of politics 77(3):787–804. Kertzer, Joshua D, Deborah Jordan Brooks and Stephen G Brooks. 2017. “Do Partisan Types Stop at the Waters Edge?” Manuscript. Harvard University and Dartmouth College . Kono, Daniel Y. 2008. “Does public opinion affect trade policy?” Business and Politics 10(2):1– 19. Krimmel, Katherine, Jeffrey R. Lax and Justin H. Phillips. 2016. “Gay Rights in Congress.” Public Opinion Quarterly 80(4):888 – 913. Kuk, John, Deborah Seligsohn and Jiakun Jack Zhang. 2017. The Partisan Divide in US Congres- sional Communications after the China Shock. Technical report UC San Diego Working Paper (June 2017). Lake, David A. 2009. “Open economy politics: A critical review.” The Review of International Organizations 4(3):219–244. Lax, Jeffrey R and Justin H Phillips. 2009. “Gay rights in the states: Public opinion and policy responsiveness.” American Political Science Review 103(3):367–386. Lax, Jeffrey R and Justin H Phillips. 2012. “The democratic deﬁcit in the states.” American Journal of Political Science 56(1):148–166. Lax, Jeffrey R., Justin H. Phillips and Adam Zelizer. 2019. “The Party or the Purse? Unequal Representation in the US Senate.” American Political Science Review p. 124. Lee, Boram, Michael Pomirchy and Bryan Schonfeld. 2019. “Differential Dynamic Responsive- ness in the U.S. Congress: Evidence from NAFTA.”. Lee, David S., Enrico Moretti and Matthew J. Butler. 2004. “Do Voters Affect or Elect Policies? Evidence from the U. S. House.” The Quarterly Journal of Economics 119(3):807–859. Lee, Frances E. 2008. “Dividers, not uniters: Presidential leadership and senate partisanship, 1981-2004.” The Journal of Politics 70(4):914–928. Lee, Frances E. 2016. Insecure majorities: Congress and the perpetual campaign. University of Chicago Press. Lenz, Gabriel S. 2012. Follow the leader? how voters respond to politicians’ policies and per-

32 formance. Chicago studies in American politics Chicago ; London: The University of Chicago Press. Levendusky, Matthew. 2009. The partisan sort: How liberals became Democrats and conserva- tives became Republicans. University of Chicago Press. Libya: UN backs action against Colonel Gaddaﬁ. March 18, 2011. BBC News . McCarty, Nolan, Keith T. Poole and Howard Rosenthal. 2009. “Does Gerrymandering Cause Polarization?” American Journal of Political Science 53(3):666–680. Milner, Helen V and Dustin H Tingley. 2011. “Who supports global economic engagement? The sources of preferences in American foreign economic policy.” International Organization 65(1):37–68. Milner, Helen V and Dustin Tingley. 2015. Sailing the Water’s Edge: The Domestic Politics of American Foreign Policy. Princeton University Press. Mondak, Jeffery J. 1993. “Public opinion and heuristic processing of source cues.” Political behavior 15(2):167–192. Moravcsik, Andrew. 1997. “Taking preferences seriously: A liberal theory of international politics.” International organization 51(4):513–553. Oatley, Thomas. 2011. “The reductionist gamble: Open economy politics in the global economy.” International Organization 65(2):311–341. Owen, Erica. 2017. “Exposure to Offshoring and the Politics of Trade Liberalization: Debate and Votes on Free Trade Agreements in the US House of Representatives, 2001–2006.” International Studies Quarterly 61(2):297–311. Owen, Erica and Noel P Johnston. 2017. “Occupation and the Political Economy of Trade: Job Routineness, Offshorability, and Protectionist Sentiment.” International Organization 71(4):665–699. Prior, Markus. 2018. Hooked: How Politics Captures People’s Interest. Cambridge University Press. Prior, Markus and Lori D Bougher. 2018. “Like Theyve Never, Ever Seen in This Country? Polit-

33 ical Interest and Voter Engagement in 2016.” Public Opinion Quarterly 82(S1):822–842. Scheve, Kenneth F. and Matthew J. Slaughter. 2001. “What determines individual trade-policy preferences?” Journal of International Economics 54(2):267–292. Schultz, Kenneth A. 2001. Democracy and coercive diplomacy. Vol. 76 Cambridge University Press. Soroka, Stuart N. and Christopher Wlezien. 2008. “On the Limits to Inequality in Representation.” PS: Political Science amp; Politics 41(2):319327. Stimson, James A, Michael B MacKuen and Robert S Erikson. 1995. “Dynamic representation.” American political science review 89(3):543–565. Tausanovitch, Chris and Christopher Warshaw. 2013. “Measuring Constituent Policy Preferences in Congress, State Legislatures, and Cities.” The Journal of Politics 75(2):330–342. Tomz, Michael, Jessica LP Weeks and Keren Yarhi-Milo. 2020. “Public Opinion and Decisions about Military Force in Democracies.” International Organization pp. 1–25. Trager, Robert F and Lynn Vavreck. 2011. “The political costs of crisis bargaining: Presidential rhetoric and the role of party.” American Journal of Political Science 55(3):526–545. Warshaw, Christopher and Jonathan Rodden. 2012. “How Should We Measure District-Level Public Opinion on Individual Issues?” The Journal of Politics 74(1):203–219. Wildavsky, Aaron. 1966. “The two presidencies.” Society 35(2):23–31. Wlezien, Christopher. 2004. “Patterns of Representation: Dynamics of Public Preferences and Policy.” Journal of Politics 66(1):1–24. Zaller, John R. 1992. The Nature and Origins of Mass Opinion. Cambridge University Press.

34 Appendix

Survey Data

For NAFTA, we use the following surveys from 1993, with a total sample size of 15,982: LA Times (January 14-17), Gallup/CNN/USA Today (March 29-31), Yankelovich/Time Magazine/CNN (June 17-21), CBS News (August 2-3), Yankelovich/Time Magazine/CNN (September 8-9), NBC News/Wall Street Journal (September 10-13), CBS News/NY Times (September 16-19), ABC News (September16- 19), Times Mirror (September 24-27), Los Angeles Times (September 25-28), Gallup/CNN/USA Today (November 2-4), Yankelovich/Time Magazine/CNN (November 11), and CBS News/NY Times (November 11-14). For the Uruguay Round Agreements Act, we use the LA Times survey conducted October 17-19, 1994 and the Gallup/CNN/USA survey conducted November 28-29, 1994 (total sample size 2,292). Our public opinion estimates for the bill denying education ben- eﬁts to undocumented immigrants are based on a survey conducted by the LA Times (September 7-10, 1996, 1,522 observations). For trade liberalization with China, we draw on a Gallup survey (November 18-21, 1999) and a CNN/USA Today/Gallup poll (April 7-9, 2000) for a total survey population of 2,016. For the authorization of force in Iraq, we draw on several surveys from 2002, including an ABC News/Washington Post survey (September 23-26), a CBS/NY Times poll (October 3-5) and a Gallup/CNN/USA Today survey (October 3-8), yielding a total sample size of 3,448 respondents. For CAFTA-DR, we draw on the PIPA/Knowledge Networks survey from September 22-26, 2005 (812 respondents). For the Secure Fence Act of 2006, we draw on survey data from the LA Times (April 8-11, 2006), CBS News/NY Times (May 4-8, 2006), and NBC News/Wall Street Journal (June 9-12, 2006), for a total sample size of 3,600. For the Iraq redeployment bill, we use the 2007 Cooperative Congressional Election Study (CCES), which surveyed 10,000 respondents. To assess support for intervention in Libya, we use a CNN/ORC poll from May 24-26, 2011, and a Gallup poll from June 22, 2011 (2,006 respondents total). For the US-Korea free trade agreement, we draw on the USA Today/Gallup poll from January 14-16, 2011 (sample size 1,032). Finally, for the No

1 Sanctuary for Criminals Act, we draw on the 2017 CCES, which surveyed 18,200 respondents.

Responsiveness Bill by Bill (Overall and Co-Partisan)

Examining legislator voting on each of five trade bills, in Table A1, we find inconsistent evidence of co-partisan responsiveness and responsiveness to the overall constituency. Only Korea exhibits strong responsiveness to co-partisans. We do, however, find consistent evidence of co-partisan responsiveness on security bills in Table A2. Finally, we examine voting on each immigration bill in Table A3 and find support for greater responsiveness to co-partisans for the 1996 bill but more responsiveness to the overall constituency for the sanctuary cities bill).

Table A1: Responsiveness on Trade Bills in the House

Dependent variable: Vote in Favor NAFTA (1993) WTO (1994) China (2000) CAFTA (2005) Korea (2011) Overall 126.573∗∗∗ 17.594∗∗ 12.024 214.425∗∗ −126.124∗∗∗ (44.616) (8.961) (8.276) (85.877) (38.539) Co-partisan −117.926∗∗∗ −12.293 −14.498∗ −216.324∗∗ 128.763∗∗∗ (44.580) (8.689) (8.224) (85.756) (38.630) Observations 434 433 434 432 429 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Table A2: Responsiveness on Security Bills in the House

Dependent variable: Vote in Favor Iraq (2002) Iraq (2007) Libya (2011) Overall 0.268 4.391 −10.457∗ (3.459) (7.609) (6.165) Co-partisan 5.013∗∗∗ 6.514∗∗∗ 18.812∗∗∗ (1.711) (2.167) (3.918) Observations 429 424 418 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

2 Table A3: Responsiveness on Immigration Bills in the House

Dependent variable: Vote in Favor Education Beneﬁts (1996) Secure Fence (2006) No Sanctuary (2017) Overall 7.918∗∗ 3.300 38.105∗∗∗ (3.765) (4.131) (10.988) Co-partisan 8.357∗∗∗ −1.861 14.306∗∗∗ (2.649) (3.203) (3.874) Observations 429 422 423 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Analysis Addressing Reverse Causation

We use the following empirical speciﬁcation. Let Yi be legislator i’s roll call vote, which takes on the value 1 when a legislator votes yes on a bill and the value 0 when a legislator votes no.

−1 Pr(Yi = 1) = logit (β0 + β1ProportionConstituencySupport+

β2ProportionUneducatedCopartisanSupport+

β3CopartisanPopulation + β4CrosspartisanPopulation + γb) (3)

If β2 > β1, then uneducated co-partisan support has a greater effect on roll-call voting than overall constituency support. We explore responsiveness to uneducated co-partisans using the pooled analyses. Table A4 shows that as was the case with the analysis of co-partisans, responsiveness is biased towards (uneducated) co-partisans for security and immigration, but not for trade bills. In the Appendix, we demonstrate that we ﬁnd the same results in the bill by bill analysis using the support of uneducated co-partisans instead of the general level of co-partisan support.

Less-Educated Co-Partisans Bill by Bill

We now assess whether the patterns of responsiveness to co-partisans on trade bills explored above hold when examining responsiveness to uneducated co-partisans in Table A5. Table A6 tests

3 Table A4: Responsiveness to Overall and Uneducated Co-partisan Constituencies

Dependent variable: Vote in Favor Trade Security Immigration Overall 14.823∗∗∗ 2.634 3.244 (2.998) (2.185) (2.267) Uneducated Co-partisans −13.784∗∗∗ 12.668∗∗∗ 5.693∗∗∗ (3.080) (0.880) (1.270) Observations 2,162 1,271 1,274 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 whether the level of responsiveness to uneducated co-partisans is any different from the previous result on responsiveness to all co-partisans on security bills. Responsiveness to uneducated co-partisans is consistent with the prior results. For all three immigration bills, the same results also hold in Table A7 when examining only uneducated co-partisans rather than all co-partisans.

Table A5: Responsiveness on Trade Bills in the House (Uneducated Co-Partisans)

Dependent variable: Vote in Favor NAFTA (1993) WTO (1994) China (2000) CAFTA (2005) Korea (2011) Overall −15.896∗ 17.924∗∗ 14.756∗ 77.920∗∗ −24.479 (8.836) (8.688) (8.057) (36.927) (17.473) Uneducated Co-partisans 32.612∗∗∗ −12.974 −17.905∗∗ −83.023∗∗ 28.865 (11.247) (8.646) (8.280) (38.266) (18.785) Observations 434 433 434 432 429 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Table A6: Responsiveness on Security Bills in the House (Uneducated Co-Partisans)

Dependent variable: Vote in Favor Iraq (2002) Iraq (2007) Libya (2011) Overall 1.129 4.180 −6.077 (3.401) (7.588) (5.642) Uneducated Co-partisans 4.879∗∗∗ 6.545∗∗∗ 20.781∗∗∗ (1.767) (2.196) (4.058) Observations 429 424 418 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

4 Table A7: Responsiveness on Immigration Bills in the House (Uneducated Co-Partisans)

Dependent variable: Vote in Favor Education Beneﬁts (1996) Secure Fence (2006) No Sanctuary (2017) Overall 8.400∗∗ 3.235 38.802∗∗∗ (3.742) (3.828) (11.044) Uneducated Co-partisans 7.914∗∗∗ −1.940 14.667∗∗∗ (2.603) (3.052) (3.964) Observations 429 422 423 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Overall, we ﬁnd strong evidence of responsiveness to uneducated co-partisans, i.e. those co- partisans who are least likely to know in advance how their legislator is going to vote.

Accounting for Educated Inﬂuence

One alternative explanation to our findings is that the federal government is more responsive to the affluent than to the middle or lower classes (Bartels, 2008; Gilens, 2012).21 Because we do not have sufficiently detailed Census data to estimate district-level preferences for income groups,22 we use college education to proxy for affluence. Though we cannot directly test whether the affluent have disproportionate influence, we can examine whether our results are robust to accounting for the potentially disproportionate influence of college-educated voters. In Table A8, we test whether the pooled results hold when including the preferences of the college-educated. We do in fact find the same relationships as before, i.e. responsiveness to co-partisans on security and immigration issues (and in the sample including all bills).

21However, Lax, Phillips and Zelizer(2019) show that disproportionate responsiveness to affluent citizens is actually driven by co-partisan responsiveness by Republicans. 22Specifically, the Census does not provide granular enough data to give us, for example, the number of college- educated white women between the ages of 18 and 29 who have an income of between 50,000 dollars and 59,999 dollars in the fifth district of New Jersey.

5 We use the following empirical speciﬁcation:

−1 Pr(Yi = 1) = logit (β0 +β1ProportionConstituencySupport +β2ProportionCopartisanSupport+

β3ProportionEducatedSupport +β4CopartisanPopulation+β5CrosspartisanPopulation+γb) (4)

If β2 > β3, then co-partisan support has a greater effect on roll-call voting than the support of highly educated constituents. In Table A8, we test whether the pooled results hold when including the preferences of the college-educated. We do in fact ﬁnd the same relationships as before, i.e. responsiveness to co- partisans on security and immigration issues (and in the sample including all bills).

Table A8: Responsiveness to Overall, Co-partisan and Educated Constituencies

Dependent variable: Vote in Favor Trade Security Immigration Overall 27.135∗∗∗ −15.019∗∗ 3.803 (6.241) (6.494) (12.974) Co-partisan −18.000∗∗∗ 12.160∗∗∗ 5.685∗∗∗ (3.552) (0.860) (1.293) Educated −7.854 19.263∗∗∗ −1.029 (5.652) (7.273) (14.362) Observations 2,162 1,271 1,274 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Cross-level MRP Validity Checks

In this section, we present some validity checks for the cross-level MRP methods used in the paper. There are two bills in our sample that allow us to perform the MRP using both state-level and district-level identifiers and compare how different the corresponding estimates are. In Figure7, the estimates using state identifiers for Iraq (2007) are plotted against the estimates using district identifiers. One can see that these estimates are highly correlated (r = 0.98). The same is true in

6 Figure 7: Iraq (2007) (Cross Level by Normal Estimates)

Figure8 for the sanctuary cities bill in 2017 (r = 0.96).

7 Figure 8: No Sanctuary (Cross Level by Normal Estimates)