Indirect Presidential Influence,State-Level Approval,and Voting in the U.S. Senate

Theoretical Framework

In this article, we are interested in the extent to which senators account for their constituents’ approval of the president when casting a roll-call vote on key legislation. We theorize that presidents can have influence and gain additional support (i.e., votes) on key issues when they are popular. More specifically, we examine an indirect presidential influence on senators, one that operates through constituents. A senator’s constituents can communicate a signal to the senator through their approval or disapproval of the president and can consequently persuade their senator to support the president.

As constituency preferences, defined here as voters’ opinions on issues, policy, and the direction of the country, is a broad concept, we posit that state-level presidential approval ratings serve as a useful indicator for one distinct component of this complex concept. We contend that state-level approval of the president is an informative indicator for the amount of indirect influence a president has on a senator and his vote. State-level presidential approval represents how much the senator’s constituents approve of the president, his policies, and his job performance. This serves as a cue for the senator, one that is informative and useful above and beyond the partisan composition of his state.

We present our theoretical framework in three parts. First, we discuss the role of the president and his desire to persuade members of Congress and the public to support his policy positions. Next, we outline the multitude of factors that influence senators when they cast roll-call votes. Last, we look at the role of the public in the process and consider the proposed relationships that tie the actors and institutions together, examining the indirect influence that the president can have on a senator by being popular with the senator’s constituents. In this section, we also outline our expectations for the conditional nature of this relationship. More specifically, we theorize about when senators should be more or less likely to support the president on key issues, depending on the party affiliation of the senator and whether the senator is up for reelection. After outlining our basic theoretical framework, we consider how our analysis contributes to the existing literature on presidential approval and voting in Congress.

The President’s Indirect Influence

As a key player in the legislative process, one of the president’s main goals while in office is to persuade the public and members of Congress to support his preferred policies. Popular presidents frequently use high approval numbers to invoke the idea of an electoral mandate or to “go public” to help ensure success for their initiatives and legislation (Kernell, 1997). Although Edwards (2007) finds that presidents have difficulty persuading the public to support detailed policy initiatives, we expect the president may have more success when trying to shape the actions of senators by being popular with their constituents.

Presidential influence on senators can operate directly and indirectly. Directly, a president can pressure and try to persuade a senator to support a specific policy through conversations. A president can also influence senators by having an indirect influence which operates through the mass public. Past studies suggest a popular president can leverage high national-approval numbers in the public to win support in Congress (Edwards 1976, 1980; Ostrom & Simon, 1985; Rivers & Rose, 1985). As Kernell (1997, p. 250) notes, “All the president need do is convince a sufficient number of politicians that the political cost of resisting his policy is greater than any political gain.” Although the president tries to convince senators to vote in a manner that supports his preferred policies, there are a multitude of factors that affect senators’ votes.

Senators’ Voting Decisions

There are many potential explanations for a senator’s vote in support of a bill. Among these are explanations that focus on factors within the Senate, such as the senator’s personal preferences and party leaders, and those that look to influences outside of the chamber, such as constituency preferences and electoral concerns (Kingdon, 1989). Within the chamber, individual preferences and party effects are likely to influence voting decisions. There is little doubt that personal preferences influence a senator’s vote, despite the difficulty in separating party effects and individual preferences (Krehbiel, 1993). After all, most senators enter Congress with policy preferences, at least in certain areas. Additionally, the majority party in the Senate exerts control over its party members through its ability to control the agenda (Campbell, Cox, & McCubbins, 2002), and the influence of party leaders is the strongest when the party can offer the senator something in return for her vote (e.g., money, media coverage, committee assignments; Smith, 2007).

Although these factors clearly account for much of the influence on senators (and much of the literature), there are other factors—those that occur outside of Congress—that can equally influence a senator’s vote. Of these, the most important is the senator’s constituency, which is a key component of our analysis. Senators need to take constituent preferences into account when deciding whether or not to support the president on key votes because of the electoral connection.

Without the support of the voters back home, the senator will find herself out of office and thereby unable to accomplish her policy objectives (Arnold, 1990; Fenno, 1973; Mayhew, 1974). The literature on congressional voting behavior suggests that how members of Congress vote may affect their electoral fortunes (Bovitz & Carson, 2006; Carson, 2005; Fiorina, 1974). As members of Congress have strong desires to achieve reelection, it is conceivable that actions which are right for any given member of Congress are ones that will please his constituency and guarantee reelection. Thus, the relationship between a senator’s vote and constituency preferences is likely to be strengthened as the senator approaches reelection. After all, if an incumbent is thought to be “out of touch” with constituents, she is more likely to face a quality challenger in the upcoming election (Jacobson, 2001). Recognizing this electoral connection, members of Congress seek out information about their constituents’ preferences to aid in their voting decisions.

Constituents and the Indirect Influence

Although it is clear that a senator must stay attuned to the needs and wishes of her constituency, it may not always be a simple task for her to take constituent preferences into account. Constituency preferences have a variety of components, including constituents’ issue opinions, policy preferences, partisan and ideological identifications, and approval of the president. The majority of the public may not have clear, stable, easily identifiable issue opinions on complex policy proposals (Converse, 1964), but most can express approval or disapproval of the president and his job performance. Thus, presidential approval of the president is not a comprehensive indictor, but is an important and informative indicator of constituent preferences, and one that is a vital signal to senators about how they should vote on important legislation.

We contend that presidential approval is a proxy for one component of constituent preferences, as it indicates the amount of indirect influence a president is likely to have on a senator’s vote independent from constituency preferences and partisanship. More specifically, we suggest that state-level presidential approval can be used as one indicator for constituency preferences, particularly when analyzing the voting decisions of senators, as state-level approval perfectly overlaps with constituency approval. Whereas high approval numbers lead senators to conclude that their constituents prefer positions that the president supports, low approval numbers indicate that not only is the president unpopular, but that supporting legislation the president favors may be problematic for the senator. Approval numbers serve as a cue to a senator, above and beyond his likelihood to support the president based on the partisanship of his state, his party affiliation, and personal policy preferences. In sum, we argue that to increase chances of reelection, senators consider their constituents’ preferences, one component of which are reflected, summarized, and readily available in presidential approval. Consequently, we can utilize state-level presidential approval as an indicator of constituency preferences.

We further contend that this relationship between the president, constituency preferences, and a senator’s voting decision is conditional on party membership. Due to similar ideologies and goals, we expect that senators of the president’s party would agree with the president on key legislation most of the time, regardless of the degree to which their constituents approve of the president. We posit that senators who do not share the president’s party affiliation will respond to the president’s approval rating to an even greater extent than senators of the same party. In other words, due to ideological differences and varying preferences, senators who are not of the president’s party will vote for the president’s position less often, but an increase in the president’s approval in their home states will result in a larger increase in the percent of time they support the president than will occur for same party senators. We expect this for two central reasons. First, senators of the president’s party already support the president a majority of the time making a large increase in agreement unlikely. Second, if an opposite party president is popular in a senator’s home state and the senator does not support the president on key legislation, he needs to be concerned about electoral repercussions.

These theoretical expectations lead us to posit two primary hypotheses:

Although we agree with past research that finds that “presidential popularity may operate at the margins” (Bond & Fleisher, 1984, p. 304), we theorize and demonstrate that it is an important, albeit indirect, influence. That is, presidential popularity should not be the largest factor weighing on a senator’s vote, but one that is significant enough to have an effect on bipartisnaship and cooperation, even in the face of strong polarization.

Previous Investigations

The work of Edwards (1976, 1977, 1997) suggests that our theoretical framework is well-founded, as he finds that members of Congress do not respond to the president’s overall (i.e., national) approval, but are likely to respond to his popularity among subgroups, especially those that are part of their own electoral coalitions. Although previous scholarship supports our contention that presidential approval is a useful, informative indicator for the amount of indirect influence a president has on a senator’s vote, the explicit connection has proved difficult to measure and has led to conflicting findings. Part of this difficulty lies in the use of national-level data rather than district-specific measures and a focus on the House of Representatives. ¹ For example, Canes-Wrone and de Marchi (2002) examine roll call votes in the House during the George H.W. Bush and Bill Clinton administrations and find that the relationship between national public approval and presidential support only exists on salient or complex issues. Borrelli and Simmons (1993) find that the national electoral context from which a member of Congress emerges influences his responsiveness to presidential approval. Bond and Fleisher (1980) also find that overall presidential approval affects presidential support, but the relationship is conditioned on the partisan composition of Congress. More specifically, they find that if Congress is controlled by the president’s party, then public opinion influences presidential support in the House, but if the opposite party controls Congress, there may be no effect on presidential support.

In later work, Bond and Fleisher (1984) examine the effect national-level presidential popularity has on presidential success in Congress. The authors acknowledge that national-level presidential approval is a rudimentary measure of constituent preferences as it may not be directly relevant to members of Congress as sentiment in their district could vary widely from national opinion. Because of the limitations of this measure, Bond and Fleisher’s (1984) finding that popularity is related to support for the president, but that it cannot overcome the partisan and ideological leanings of members of Congress, is unsurprising. Edwards (1997) suggests that to truly test the effect of approval on presidential support, we would want to measure approval over an extended period of time. By examining annual, national-level approval over several decades, Edwards (1997) finds that members of Congress are responsive to those in the electorate who share their party affiliation.

We refine past literature connecting presidential approval and congressional support by measuring presidential approval at the level of the member’s constituency, focusing on the Senate, extending the time period analyzed, and examining individual senators’ votes. Using more specific, state-level data, and examining votes at the individual level, we find, contrary to Bond and Fleisher (1984), that the president is able to exert indirect influence on senators when he is popular in their states. We also find that senators heed the approval of particular subgroups, similar to the work of Edwards (1997). Where Edwards (1997) found an effect for members of Congress responding to those in the national electorate who share their party affiliation, we find an effect for senators responding to their constituents rather than national public opinion.

State-Level Presidential Approval and the Senate

Although presidential approval is frequently measured at the national-level, state-level polls are now readily available and are likely to be a more relevant cue to senators. After all, the constituents back home have the ultimate say as to whether the member will be reelected and able to continue his work (Carson 2005; Mayhew, 1974). Cohen, Bond, Fleisher, and Hamman (2000) use state-level public opinion data to test the relationship between senators’ support for the president and public approval in 1996. Although they do not find evidence of this relationship in that year, we believe there are strong theoretical reasons, as mentioned above, to expect this relationship to emerge when using longitudinal data. Additionally, using longitudinal data allows us to explore how reelection factors into this relationship and the existence of this relationship over several administrations.

Presidential approval ratings, especially at the state level, are one way for members of Congress to gauge their constituents’ feelings, particularly as they are readily available measures. Recognizing this, we model the relationship between indirect presidential influence, measured using state-level presidential approval, and voting in the Senate. The Senate is understudied when it comes to analyzing voting behavior and electoral outcomes. As Carson (2008) states, “While students of congressional politics have examined the representational connection between voting behavior and electoral outcomes in the context of House elections, virtually no systematic attention has been given to the same relationship in the context of Senate elections” (p. 36). By studying the Senate we are able to take advantage of the variance provided by the upper chamber’s staggered election cycles. Additionally, the Senate is the perfect place to test our research question, as the state’s approval of the president maps directly onto the senator’s constituency. Thus, a senator can use state-level presidential approval as a cue when determining her constituents’ preferences. As we control for partisanship of the state in subsequent analyses, the influence captured by state-level presidential approval is above and beyond the partisan aspect of constituent preferences. To reiterate, we hypothesize that senators not of the president’s party are more likely to vote in accordance with his position if state-level approval in their state is high, holding all else equal. Also, we contend that senators facing reelection are more likely than senators not up for reelection to vote with the president’s position if the president is popular in their home state, holding all else equal.

Data, Method, and Analysis

To examine how indirect presidential influence, captured by state-level presidential approval, affects votes in the Senate, we focus our analysis on roll calls. We look at roll calls as senators are unable to avoid expressing positions that may be divisive with voters on roll calls (Bovitz & Carson, 2006). More specifically, we analyze all key votes, as identified by Congressional Quarterly (CQ), from 1989 to 2004. ² CQ’s criteria for selecting key votes is one or more of the following: (a) they are controversial; (b) they provide a test of presidential or political power; (c) they have the potential to greatly impact the nation and lives of Americans. Given these criteria, key votes are ideal for this analysis for several reasons. First of all, past research shows that these votes do an especially good job measuring the balance of power between Congress and the president (Shull & Vanderleeuw, 1987). Second, key votes represent issues that are important to voters and are likely to be raised in competitive elections. Third, we focus on key votes because these are votes on which there is an identifiable position for the president. In sum, key votes provide us with issues that are important, controversial, and visible to the general public and the president. ³ For each key vote during this time period, we code how the senator voted and whether or not the vote fell in line with the president’s position. To measure constituent preferences, we utilize state-level presidential approval from the Job Approval Ratings (JAR) collection (Beyle, Niemi, & Sigelman, 2002). ⁴ For our analyses, we use the percentage of respondents indicating that they approve of the president’s overall job performance. We also use only those respondents who identify themselves as voters. ⁵ These data allow us to test our theoretical expectation that presidents can indirectly influence senators’ vote decisions when they are popular.

Although the JAR collection compiles state-level presidential approval ratings several times a year, we average the presidential approval rating for each state to create a yearly approval score. We use a yearly measure of state-level presidential approval for two primary reasons—the first methodological and the second theoretical. First, we look at a senator’s level of presidential agreement per year since CQ compiles key votes by year. ⁶ Methodologically, the benefit in averaging approval over the course of a year is that the main independent variable (averaged state-level approval) and dependent variable (agreement with the president’s position on key votes) reflect the same period of time. The second reason we employ a yearly measure of state-level approval is theoretical. We believe the yearly average of presidential approval is a more realistic indicator of the information that senators utilize. We do not theorize that senators respond to weekly changes in presidential approval in their state. A senator does not change a vote on an important policy initiative merely because his constituents’ approval of the president changed slightly in response to an incident that occurred in recent days. Rather, we contend that senators look at the broad, overarching trends of support among their constituents and base their votes on these more stable preferences because they represent sustained support or disapproval for the president’s agenda (Edwards, 1997; Neustadt, 1960, 1990). In sum, for both methodological and theoretical reasons, we believe that yearly state-level approval is the appropriate indicator to use in our analysis.

In the models that follow, the unit of analysis is each senator in each year. The dependent variable presidential agreement is measured as the percentage of time over the course of a year a senator’s position on a key vote is the same as the president’s. ⁷ There are four independent variables of primary interest. The first is state-level presidential approval, which is coded as the percentage of respondents who approve of the president’s overall job performance. ⁸ Higher values indicate higher levels of support for the president. Second, we include an indicator variable for reelection, with a value of 1 indicating that a senator is up for reelection. ⁹ The third is the interaction term presidential approval × opposite party. To create this interaction term, we include a dummy variable to indicate whether a senator is of the opposite party of the president. ¹⁰ Last, we include the interaction term reelection × presidential approval. In our analysis, we also control for whether the Senate is controlled by the opposite party of the president ¹¹ and the state’s partisanship by including the president’s percent of the vote in the last presidential election. ¹²

We utilize OLS regression to capture the strength of the relationship between the independent and dependent variables while controlling for individual senator effects. As we have multiple observations for each senator, we calculate robust standard errors, clustering on each senator to control for correlated errors across multiple observations for a single senator. The estimated coefficients and their corresponding robust standard errors for Model 1, which examines the additive effect of state-level presidential approval ratings on a senator’s agreement with the president on key votes, are reported in Table 1. ¹³

OPEN IN VIEWER

Table 1.

Effect of State Approval on Agreement with the President

	Model 1	Model 2	Model 3	Model 4
Variable	Coefficient (Robust Std. Err.)	Coefficient (Robust Std. Err.)	Coefficient (Robust Std. Err.)	Coefficient (Robust Std. Err.)
State approval	0.346* (0.075)	0.125 (0.102)	0.110 (0.103)	−0.520 (0.419)
State partisanship	−0.172 (0.114)	−0.174 (0.112)	−0.174 (0.113)	−0.304* (0.138)
Opposite party	−7.536* (1.665)	−30.151* (6.308)	−30.253*(6.317)
Opposite party* approval		0.432* (0.126)	0.432* (0.126)
Reelection			−6.038 (14.548)	18.816 (21.673)
Reelection* approval			0.116 (0.262)
Common space difference				−6.004* (2.377)
CS difference* approval				0.119* (0.045)
CS difference* reelection				−3.347 (2.155)
Divided	−9.839* (1.478)	−9.669* (1.405)	−9.670* (1.448)	−7.003* (1.804)
Constant	52.744* (5.086)	64.527* (6.480)	65.297* (6.349)	93.822* (21.822)
Number of cases	817	817	817	426
Number of groups	169	169	169	139
R 2	.062	.073	.074	.096
	F (4,168)	F (5,168)	F (7,168)	F (7,138)
	22.76	23.36	16.60	7.86

The dependent variable is the percentage of the time the senator agreed with the president on key votes and ranges from 20.25 to 92.00. Cell entries are unstandardized OLS regression coefficients clustered by senator.

*

p ≤ .05.

The positive, statistically significant coefficient for state approval in the first model suggests that as approval of the president, a proxy for constituent preferences, within a senator’s home state increases so does the senator’s agreement with the president. The negative coefficient for opposite party indicates that if a senator is of the opposite party of the president, he is less likely to agree with the president on key votes than a senator of the president’s party. The negative coefficient for divided signifies that when the Senate is not controlled by the president’s party, a senator is less likely to agree with the president. The coefficient for state partisanship is not statistically significant, offering no evidence that the partisan leanings of a state are related to the senator’s rate of agreement with the president. ¹⁴ However, looking at a simple additive model does not adequately test the theory and hypotheses previously outlined, as it does not take into account the conditional relationship between the party of the senator and the approval rating of the president. As we would expect, the state-level approval variable in the additive model indicates that senators whose constituents support the president agree with the president at a higher rate. Yet, without knowing how changes in presidential approval affect senators of the president’s party and senators of the opposite party, this model is of finite value. Thus, Model 2 in Table 1 examines the conditional relationship between a senator being of the opposite party of the president and the president’s approval rating in the senator’s home state. This model includes the aforementioned interaction term presidential approval × opposite party. This is the independent variable that is of greatest importance to testing our first hypothesis that a change in state-level presidential approval will be associated with a change in a senator’s agreement rate with the president, dependent on whether the senator is a member of the president’s party. The interaction term opposite party × approval indicates that the slope is significantly different for members of the president’s party and members of the opposite party. In other words, a one unit increase in presidential approval results in a .432 increase in agreement with the president for opposite party senators above and beyond the effect for senators of the president’s party. By adding the coefficients for state approval and opposite party × approval, it is possible to determine that .557 is the partial slope coefficient for senators of the opposite party. ¹⁵ This means that while senators who are not members of the president’s party are less likely to agree with the president than senators of his party, as the approval rate increases in the opposite party senators’ home states, their agreement rate increases by a greater percentage. For every one point increase in presidential approval in an opposite party senator’s home state, agreement with the president on key votes will increase by approximately half a percentage point, whereas it increases by approximately a 10th of a percentage point for senators of the president’s party. Although half a percentage point may seem like a small increase, it is substantively consequential, particularly considering that we control for the partisanship of the state. If presidential approval increases by 10 percentage points in an opposite party senator’s state, this leads to a five percentage point increase in agreement with the president. Thus, our expectation that higher levels of presidential approval will lead to higher levels of agreement, among senators who do not belong to the president’s party, is confirmed.

Our second hypothesis contends that senators up for reelection are more likely to vote with a popular president than those not up for reelection. Thus, Model 3 in Table 1 incorporates reelection and the interaction term reelection × approval. ¹⁶ Although the reelection coefficient is not statistically significant, we also test the conditional nature of this hypothesis. To examine whether the extent to which a senator’s agreement with a popular president at election time varies by party, we test whether the coefficient on the interaction term reelection × approval is different between groups (in this case, parties). We find that senators from the same party as the president are more likely to support a president who is popular in their home state when they are up for reelection than their colleagues of the opposite party who are also up for reelection. ¹⁷ It is not surprising that this finding only exists for members of the president’s party. After all, senators not of the president’s party must be wary of primary challengers and cannot always vote with a president of the opposite party even if he is popular in their home state. When facing reelection, opposite party members may be reluctant to cast a vote with the president for fear of drawing the fire of fellow partisans, including those who would consider challenging the senator in a primary. To better understand how opposite party senators are behaving in this situation, we include Common Space scores in our reelection model (Model 4, Table 1). ¹⁸ These scores allow us to calculate the spatial distance between senators and the president. The expectation is that opposite party members who are farther from the president on this measure of ideology are less likely to vote with the president’s position. This model only includes those senators not of the president’s party. The results indicate that as the absolute distance between a senator’s common space score and the president’s common space score grows, it becomes less likely that the senator will vote with the president. The interaction term CS difference × reelection indicates that for a senator not of the president’s party, the farther her preferences are from the president, the less likely the senator is to support the president when she is up for reelection. ¹⁹ This finding may suggest that these senators fear the emergence of a primary challenger if they deviate from the party and support an opposite party president on key legislation, demonstrating that a president, even a popular president, can only exert so much influence on senators.

To better illustrate the effect of changes in approval on a senator’s agreement with the president, we have selected various values of the independent variables to determine the effect of the dependent variable using the equation from Model 2, Table 1. The predicted values of the senator’s agreement rate illustrate that changing the president’s approval rate in a senator’s state alters the percentage of time the senator votes with the president on key legislation. As we theorize, and find, that whether or not the senator is a member of the president’s party is a crucial component of this relationship, we examine the predicted values of the agreement rate first for a senator of the president’s party and then for a senator who is a member of the opposite party of the president.

As can be seen in Figure 1, when divided government is present and state partisanship is held at its mean, a senator of the president’s party has a predicted agreement rate of 51.44% when the president’s approval rating in his state is 39% (one standard deviation below the mean). When the president’s approval rating rises to 65% in a senator’s state (one standard deviation above the mean), the senator is predicted to agree with the president 54.66% of the time. Although this increase is not large, it shows that a senator of the president’s party will agree with the president more often as presidential approval in the senator’s home state increases, demonstrating indirect presidential influence on voting in the Senate.

OPEN IN VIEWER

Figure 1.

Predicted probabilities of agreement with the president. The dependent variable is the percentage of the time the senator agreed with the president on key votes. In the example reported above, calculations are conducted based on the equation for Model 2, Table 1 when divided government is present, state partisanship is held at its mean of 48.08, and the senator is either a member of the president’s party or a member of the opposite party. The mean value of our approval variable is 52.25 with a standard deviation of 12.87. The values of state approval depicted above represent values one standard deviation above and below the mean.

Although the findings for same party senators are informative, we are more interested in the indirect influence the president can assert on opposite party senators. Looking at opposite party senators (Figure 1), when a president has an approval rating of 39% (one standard deviation below the mean), a senator will only vote with an opposite party president 38.29% of the time on key votes. However, when an opposite party president is popular in a senator’s state with an approval rating of 65% (one standard deviation above the mean), a senator is predicted to agree with the president 52.62% of the time. Thus if the president is popular among a senator’s constituents, even when the senator is of the opposite party of the president, the senator is likely to support the president on more than half of the key votes in a given year, which is significant in this era of polarization.

These predicted probabilities illustrate that as a president’s popularity in a senator’s home state increases, a senator is more likely to agree with the president. However, this effect is greater for senators who are not members of the president’s party. A president can exert indirect influence and gain support for key legislation from a senator of the opposite party if he is popular among the senator’s constituents. Yet, a president’s popularity does not persuade senators of his own party to the same extent, most likely because they are already likely to agree with the president on key issues.

National-Level and State-Level Approval

Although we demonstrate that state-level approval affects senators’ voting decisions, as previously discussed, numerous studies posit and examine the role of a president’s national approval on voting in Congress. Recognizing this, Table 2 compares a state-level approval model to a model using national-level approval data. This comparison helps to illustrate that state-level approval data are tapping into something other than the president’s national-approval and are a suitable cue to senators for one component of constituency preferences.

OPEN IN VIEWER

Table 2.

Effect of State-level and National-level Approval on Agreement with the President

	State–level approval	National–level approval
Variable	Coefficient (Rob. Std. Err.)	Coefficient (Rob. Std. Err.)
State approval	0.125 (0.102)
National approval		0.548* (0.066)
State partisanship	−0.174 (0.112)	0.150* (0.063)
Opposite party	−30.151* (6.308)	−8.579 (5.720)
Opposite party* approval	0.432* (0.126)	−0.165 (0.109)
Divided	−9.669* (1.405)	−9.542* (1.330)
Intercept	64.527* (6.480)	29.490* (5.188)
Number of cases	817	1307
Number of groups	169	184
R 2	.073	.152
	F (5,168)	F (5,183)
	23.36	99.17

The dependent variable is the percentage of the time the senator agreed with the president on key votes and ranges from 20.25 to 92.00. Cell entries are unstandardized OLS regression coefficients clustered by senator. The N is larger in the national approval data models because of missing data from some states in our state approval models.

*

p ≤ .05.

The first model displayed in Table 2 is a pure replication of Model 2, Table 1, depicting the relationship between state-level approval and agreement with the president on key votes. The national-level model included in this table uses Gallup/USA Today Job Performance data. ²⁰ As is evident from Table 2, national approval appears to be a statistically significant positive predictor of a senator’s agreement with the president. However, as previously discussed, until the partisan nature of the relationship is taken into account, the national-level approval coefficient is not particularly informative. When the interaction term opposite party × national approval is included in the model, the coefficient for this variable is not statistically significant, but the partial slope coefficient is .383 and is statistically signficiant. ²¹ To better compare the state-level and national-level data used in these two models, we use the Wald test. ²² From this, it is possible to conclude that once party is taken into account, state-level approval data are telling senators something different than national-level approval data. Thus, it seems reasonable to expect that state-level presidential approval acts as an informative cue to senators as this data aligns perfectly with their constituencies and is more relevant to their electoral fortunes.

Discussion and Conclusion

Although a president may not be effective in garnering support by going public on detailed policy proposals (Edwards, 2007), he can create support for his preferred legislation indirectly by being popular with senators’ constituents. Senators, who are concerned with the preferences of their constituents and their own electoral security, can utilize state-level presidential approval as a signal for their constituents’ opinions of the president, his policies, and the direction of the country. If the president is popular in a state, the senator may support the president on key legislation to represent his constituents’ opinions.

We find support for the relationship between indirect presidential influence, state-level approval, and voting in the Senate. Controlling for other factors, such as ideology and state partisanship, we find that senators representing states where the president enjoys high approval numbers vote with the president more often than senators whose constituents disapprove of the president. We also confirm our theoretical expectation that the relationship is conditional on party. Although senators of the opposite party are less likely to agree with the president than senators of the president’s party, as presidential approval increases, opposite party senators’ support increases by a greater margin. This normatively reassuring finding indicates that even in times of polarization, opposite party senators are willing to cast bipartisan votes if their constituents approve of the president and his job performance. A popular president may be able to foster bipartisanship in the Senate without directly reaching out to opposite party senators.

Reelection, however, tempers the bipartisanship a popular president can incite. We conclude that senators of the president’s party who are facing reelection vote with a popular president more often than opposite party senators who are up for reelection. Using common space scores we test the possibility that opposite party senators may be less likely to support a popular president in an election year for fear of a quality primary challenger. We find that the farther apart the preferences of the president and the senator, the less likely the senator is to support the president during a bid for reelection, even if the president is popular. This suggests that even a popular president will have a more difficult time overcoming partisanship and polarization in election years.

Our findings also provide insight into how senators might utilize state-level presidential approval figures to be responsive to the wishes of their constituents. Of course, the state-level approval data used in this analysis is not perfect, though it is the best currently available. The number of state-level polls varies across states and years, but we have attempted to resolve this issue by creating yearly approval averages. We believe this is a satisfactory solution because senators are unlikely to respond to small shifts in presidential approval, but are more concerned with the trends and large changes (Edwards, 1997). It would be ideal if the data allowed us to explore whether senators were more concerned with the preferences of specific subconstituencies, such as fellow partisans. Unfortunately, the data do not allow us to parse out the approval ratings of Republicans and Democrats. Although this distinction is possible with national-level or individual-level data (Bond & Fleisher, 2001; Edwards, 1997), we contend that state-level approval data is a useful, informative signal to senators as it directly maps onto their constituencies and reflects the preferences of people they were elected to represent. Hopefully, more detailed state-level opinion data will become available in the coming years and will allow for further investigation of this important relationship.

Overall, we find that constituents’ approval of the president, one distinct component of constituency preferences, affects senators’ vote choices, which is critical for the discipline’s understanding of public opinion, the relationship between senators, the public, and the president, and representation. Even in highly polarized, partisan times, senators will cross the aisle to support a popular president and represent the wishes of their constituents.