Companion Bills and Cross-Chamber Collaboration in the U.S. Congress

Credit Claiming and Cross-Chamber Collaboration

Prior research on collaboration in Congress provides us little guidance on developing a theory of cross-chamber interactions. Thus, we offer an initial set of hypotheses regarding which legislators are likely to introduce companion bills and who they are likely to choose as partners. We expect these senators and House members to face the same tradeoffs when deciding whether or not to engage in cross-chamber collaboration. Thus, the hypotheses will apply equally to both senators and House members. We begin this discussion with a simple observation: Most bills introduced in Congress do not have opposing chamber companions. Indeed, in our empirical investigations, we find that fewer than 10% of bills introduced in the House have a Senate companion. This low level of companion bill introduction must be because either (a) the introduction of a bill with an opposing chamber companion provides no benefits to the authors of the bills, or (b) there are benefits to introducing companion bills, but the costs of that introduction frequently outweigh the benefits. The development of companion bill partnerships and legislation is likely to be significantly more time-consuming than the development of legislation without members of an opposing chamber. Opportunities for cross-chamber interaction are rarer than interactions between same-chamber colleagues, and as our introductory examples demonstrated, there is a notable degree of animosity between the House and Senate. Given these constraints, it is hard to imagine a member of Congress developing a companion bill unless there was some significant institutional reward for such behavior. That is, to motivate this additional effort, we expect to see the benefits of developing a companion bill in terms of increasing the odds that her companion bill would receive legislative attention in her home chamber. This implies our first hypothesis:

Assuming that companion bills are indeed more likely to receive legislative attention in their home chambers, it is reasonable to then ask who might introduce such legislation. As we mentioned, the development of companion legislation is likely more difficult or time-consuming than the development of legislation in isolation from an opposing chamber. As such, the development of a companion bill is likely to impose a nontrivial opportunity cost on a legislator electing to develop such a bill. The time and effort required to secure a companion bill partner is likely to divert a member’s attention from other legislative activities such as the development and introduction of noncompanion legislation, consultation with same-chamber colleagues, re-election-oriented activities such as responding to constituent requests, and so on. ⁶ Thus, a legislator should be most likely to engage in companion bill development when paying such opportunity costs would be least problematic. A legislator who is secure in her re-election is likely to be most comfortable paying some costs to collaborate across chambers at the expense of more individually focused within-chamber activities such as introducing her own legislation with no potentially competing sponsors (Anderson, Box-Steffensmeier, & Sinclair-Chapman, 2003). In addition, a legislator with a longer tenure is likely to have a considerable personal vote and a high level of general legislative aptitude such that the costs of introducing companion bills is likely to be lower for a highly experienced legislator relative to her more junior counterparts (Coates, 1995; Parker & Parker, 2009; Schiller, 1995). Finally, a more ideologically moderate legislator is likely to find it considerably easier to secure a companion bill partner in an opposing chamber, as an ideologically moderate legislator is more proximate to more members in an opposing chamber than an ideological extremist (Fowler, 2006a; Rocca & Sanchez, 2008). Such ideological proximity ought to reduce the costs of finding an opposing chamber partner, and thus, an ideologically moderate legislator ought to be more likely to introduce companion bills.

This discussion of who is likely to introduce companion bills implies three additional hypotheses:

Beyond the decision to introduce a companion bill, a member of Congress must also secure a companion bill partner in the opposing chamber. Such partner selection is likely to be driven by the homophilous processes that drive partner selection in many other social networks (Bratton & Rouse, 2011; Kirkland & Williams, 2014; Louch, 2000; McPherson, Smith-Lovin, & Cook, 2001). Legislators with similar constituencies and similar ideologies are themselves likely to find more common ground on which to agree about legislation. Thus, legislators with similar constituencies or ideologies ought to be more likely to introduce companion bills together. Also, legislators with some preexisting familiarity with one another ought to be more likely to introduce companion bills together, as such preexisting familiarities are likely to lower the costs of the relationship-development process. This should serve to make legislators who have served together in the same chamber in the past more likely to introduce companion bills together. This homophily-driven partner selection process implies our final four hypotheses:

Taken together, these hypotheses would imply that a legislator pays a cost to introduce companion bills when compared with the introduction of bills without a companion, she chooses companion bill partners in her opposing chamber with an eye toward minimizing those costs, and she is willing to pay these costs because companion bills provide her with significant gains in the probability her proposal will receive legislative attention.

Data and Method

The hypotheses we intend to test regarding cross-chamber collaboration focus on two empirical models: (a) a model of bill success and (b) a model governing the network formation process between legislators introducing companion bills. Indeed, we have several hypotheses regarding who introduces companion bills and who forms companion bill partnerships across chamber lines. Both of these types of hypotheses can be easily translated into expectations regarding the structure of the cross-chamber companion bill network. For example, asserting that legislators with greater tenure are more likely to introduce companion bills is identical to suggesting that legislators with longer tenure are more likely to have many connections in the companion bill partnership network, as one can only introduce a companion bill by finding a companion bill partner. Thus, all of our expectations regarding who introduces companion bills can be expressed as expectations regarding who has the most connections in a network connecting companion bill partners, and all of our expectations about the types of legislators to form partnerships can be expressed as expectations regarding which dyads build connections in such a network.

As seven of our eight hypotheses focus on this network formation process, we begin by describing our data and methods for testing these hypotheses. To test these hypotheses, we gather data on companion bill introductions from the 111th and 112th U.S. Congresses. In the 111th Congress, Democrats controlled both chambers, whereas the Republicans formed a majority in the 112th House and Democrats held onto majority control in the Senate. Our analysis focuses on these two legislative sessions, limiting our ability to explore temporal claims about companion bill patterns; however, the changing conditions between these two sessions enable us to examine these relationships under different ideological regimes. Using the RCurl package in the R statistical environment, we create a web-scraper to gather data on U.S. House of Representative bills from govtrack.com (Lang, 2007). ⁷ For every bill introduced in the House, our scraper records data on the bill’s date of introduction, the bill’s sponsor, the committees to which the bill was referred, whether the bill received a final passage vote in the House, and importantly, whether the Congressional Research Service identified a companion piece of legislation introduced in the contemporaneous session of the Senate. ⁸ If a House bill is reported to have a companion piece of legislation introduced in the Senate, our scraper records the same information about the Senate bill as was recorded for the House bill.

This process provides us with every instance of companion bill introductions over two congressional sessions. The sponsors of the House and Senate versions of a companion bill form an undirected dyadic connection in that their contemporaneous introduction of identical legislation implies some level of cross-chamber collaboration. We use these dyadic connections to construct a social network of cross-chamber collaboration between House members and senators for each session. As an example, on January 24, 2011, Representative Phil Gingrey (R-GA) introduced H.R. 5, the Protecting Access to Healthcare Act. Three days later, Senator John Ensign (R-NV) introduced S. 218, the Help Efficient, Accessible, Low-Cost, Timely Healthcare Act. The language of S. 218 is identical to H.R. 5. This introduction of companion legislation creates a network connection between Gingrey and Ensign in the 112th cross-chamber network. We code 770 House bills with Senate companions in the 111th Congress and 697 House bills with Senate companions in the 112th Congress. The leaders in cross-chamber collaboration in the 111th Congress were Dan Heller (R-NV) who sponsored 11 House bills with Senate companions and Charles Schumer (D-NY) who sponsored 30 companion bills in the Senate. For the 112th Congress, the companion bill leaders were Don Young (R-AK) in the House with 19 companion bills and Charles Schumer (D-NY) again led the Senate with 22 companion bills.

Figure 1 plots the full cross-chamber companion bill networks for the 111th and 112th U.S. Congress. ⁹ Democrats are colored blue whereas Republicans are colored red. Senators are represented by triangles whereas House members are plotted using squares. Isolates in the network are the actors who have no connection to other members of the network. We observe 155 and 125 isolates in the 111th and 112th sessions of Congress, respectively. These networks represent the dependent variables for our analysis of cross-chamber collaborative partner selection.

OPEN IN VIEWER

Figure 1.

The cross-chamber companion bill sponsorship networks for the 111th and 112th U.S. Congresses.

Our hypotheses suggest that legislators with longer tenures and more secure electoral margins are less likely to be isolates and more likely to have many connections in the network, while ideologically extreme legislators are more likely to be isolates and less likely to have many connections in the network. Furthermore, our hypotheses about these networks also suggest that legislators with similar average constituency ideologies should be likely to form connections and state delegations should be likely to build connections to each other. We also expect legislators who formerly served in one chamber together should be likely to have connections. To measure average congressional district and state ideology, we utilize survey responses to the 2008 and 2010 Cooperative Congressional Elections Study (CCES). Following Carsey and Harden (2010), we measure a state or congressional district’s ideology by factor analyzing responses to social policy questions in the CCES. Each respondent receives a factor score on the first latent dimension of the factor analysis. We code the ideology of districts or states as the average first dimension factor score of respondents in that state. ¹⁰ We also gather data on each legislator’s time of service in both chambers and code dyads as having served together if their periods of service in either chamber overlap. In addition, we code legislators’ tenure as the years of service in their current chamber, their electoral security as their two-party vote share in their most recent election, and their ideological extremity as the absolute value of their lagged DW-Nominate score. ¹¹

Having constructed a network of companion bill sponsors in Congress and developed measures of the necessary exogenous variables to test our hypotheses, we turn to our estimation strategy for these networks. Typical approaches to relational data in political science make use of dyadic logistic regressions. Unfortunately, social networks such as our companion bill networks are typically driven by both exogenous characteristics and endogenous patterns of interdependence. This interdependence violates the traditional regression assumption of conditionally independently distributed outcomes. Instead, we make use of the exponential random graph model (ERGM) to estimate the effects of our exogenous variables of interest while controlling for patterns of actor-interdependence in the network. ERGMs have become increasingly common in political science (Alemàn & Calvo, 2013; Box-Steffensmeier & Christenson, 2014; Bratton & Rouse, 2011), and Cranmer and Desmarais (2011) provide an excellent introduction to the logic of ERGMs.

To briefly review, ERGMs create a distribution of graphs against which the likelihood of the observed graph can be maximized. Because the ERGM treats the entire network as the observation of interest, this reference distribution allows an analyst to determine which patterns of interdependence (i.e., reciprocity or transitivity) are more common in the observed network than we would expect from the set of randomly generated graphs without such interdependence. With a reference distribution of possible graphs and the observed graph of interest, calculating the influence of either exogenous variables or endogenous interdependence patterns on the observed network is a simple maximum likelihood problem. For example, an analyst may be interested in the levels of reciprocity in a network. An ERGM counts the number of reciprocal dyads in the observed network, and then determines the reciprocity coefficient by finding the value of reciprocity across the distribution of random networks that makes the observed network most likely. In the absence of interdependence patterns in a network, ERGMs simplify to dyadic logistic regressions.

Our focus on cross-chamber collaboration through companion bill sponsorship necessitates a concern for a very particular pattern of interdependence. In our companion bill networks, House members cannot form connections to other House members. They can only build cross-chamber connections to senators. In addition, there are many more House members than senators in the cross-chamber network. Because House members cannot form within-chamber connections, we should observe an unusually high level of House members with a connection to one senator and no one else (Kirkland & Williams, 2014). Figure 2 demonstrates the logic of this expectation. By forming cross-chamber connections but not within-chamber connections, House members are never able to close off the potential relational triangles they form. Because there are many more House members than senators, this should result in many network structures with senators forming the “pivot” of a potentially closable triad. Indeed, our network plots in Figure 1 have a number of these types of formations that are easily spotted around the periphery. In many networks, the process of triangle closure would close these off (Cranmer, Desmarais, & Menninga, 2012), but our focus on cross-chamber connections prevents this from occurring. This preponderance of potentially closable triads is usually captured with an endogenous network term called “two-stars,” which measures the frequency of unclosed triads in an observed network. Given the much greater number of senators, we expect senators to be the “pivot” of these stars more often than House members. With a fixed number of House member-to-senator connections, but fewer senators among whom to distribute those connections, we ought to observe many two-star patterns in the network (small cross-chamber collaborative teams), with senators at the center of these stars.

OPEN IN VIEWER

Figure 2.

The formation of a preponderance of concurrent nodes in the companion bill network.

To control for this particular form of interdependence implied by Figure 2, we incorporate an offset term set at −∞ on within-chamber connections that prevents our ERGMs from simulating networks with within-chamber connections, and incorporate a term for the number of two stars and three stars in our network. These terms ensure that no within-chamber connections appear in the reference distribution of our ERGMs, and evaluate the likelihood of our network containing many unclosed triads. Between these three controls, we should be able to account for the unique types of interdependence created by our focus on cross-chamber activity. Once we control for the likely interdependence in the companion bill network, our measures of state and congressional district ideology, vote share, ideological extremity, and periods of service in each legislative chamber provide us with all the necessary measures to test our hypotheses concerning partner selection in the companion bill network. Finally, we incorporate a covariate measuring the number of total bills an actor introduced in their legislative chamber. It should be the case that legislators who introduce more legislation are more likely to also introduce companion bills. Thus, all of our network results account for the fact that some actors generally introduce more bills than other actors.

Our final expectation, and in many ways the key motivating assumption of our research, is that the introduction of a companion bill will elevate the probability that a bill will survive the legislative process. We expect that legislators are sometimes willing to pay significant opportunity costs associated with developing companion bills to pass legislation. This is only a sensible perspective if those sacrifices actually bear the fruit of bill passage. To test this expectation, we merge our data on companion bill introductions with data on the fate of legislation from the Congressional Bills Project (Adler & Wilkerson, 2013a, 2013b). The Congressional Bills Project records a host of information about the history of every bill in Congress. We are primarily interested in three variables: (a) the bill’s number of cosponsors, (b) whether a bill is reported out of committee, and (c) whether a bill ultimately passes in the House. Our theory suggests that a House member will seek out a Senate companion as an effort to signal to her home chamber that a bill has support in the Senate, and is thus worth the House’s scarce plenary time. If such an approach is successful, a House bill with a Senate companion ought to be more likely to accrue support in the House (i.e., have more cosponsors), be more likely to emerge from committee deliberation, and be more likely to ultimately pass in the House. Pairing our data on companion bills with the Congressional Bills database allows us to test the effect of a House bill having a Senate companion on each of these stages of the legislative process.

Results

The bulk of our hypotheses focus on who introduces companion bills, and with whom they partner when introducing companion legislation. In other words, most of our hypotheses focus on the formation of the companion bill network. Thus, we begin by evaluating our hypotheses regarding the companion bill network’s structure using an ERGM. To review, we expect that legislators with larger two-party vote shares and more seniority are more likely to have connections in this network (and thus to have introduced companion bills), while ideological extremists are less likely to have connections in this network. We also expect that legislators from the same party and state are more likely to develop connections in this network, and that controlling for those factors, legislators from ideologically similar constituencies will also be more likely to collaborate. Finally, we also expect that legislators who have served in the House together are more likely to develop cross-chamber connections. ¹²

To test these hypotheses, we develop an ERGM predicting the existence of a connection between two actors for the 111th and 112th Congress’ companion bill network. In addition to the terms meant to test our hypotheses, we include model terms for two- and three-star formations in our network to capture the particular form of interdependence in the companion bill network reviewed in the prior section, and an offset term set to −∞ on a same-chamber covariate that forces our ERGMs to avoid simulating connections within chambers. The results of our estimations appear in Table 1. The first and second columns of the table present results for the companion bill networks of the 111th and 112th Congress, respectively.

OPEN IN VIEWER

Table 1.

Exponential Random Graph Model Predicting Cross-Chamber Ties in the 111th and 112th U.S. Congress.

Variable name	111th Congress	112th Congress
Dyadic covariates
Same party	1.054*	0.882*
	(0.126)	(0.100)
Same state	3.391*	3.344*
	(0.124)	(0.113)
Absolute difference in constituent ideology	−0.390	−0.836*
	(0.266)	(0.264)
Service time overlap	0.014	−0.010
	(0.018)	(0.014)
Actor-specific covariates
Two-party vote share	0.004	0.012*
	(0.002)	(0.003)
Tenure	0.001	0.003
	(0.004)	(0.003)
Ideological extremity	−0.311	−0.711*
	(0.179)	(0.167)
Total bills introduced	0.027*	0.016*
	(0.003)	(0.002)
Interdependence
Two stars	0.157*	0.208*
	(0.044)	(0.039)
Three stars	−0.033*	−0.034*
	(0.006)	(0.006)
Edges	−7.754*	−7.763*
	(0.410)	(0.378)
% Correct Ties model	28.70	25.52
% Correct Ties null	0.45	0.37
AIC	4,408	6,017
BIC	4,513	6,100
Number of actors	544	542

Note. Cell entries report coefficient values from an exponential random graph model predicting the structure of cross-chamber collaborations in the 111th and 112th U.S. Congress. MLE standard errors are reported in parentheses. An unestimated, offset coefficient of −∞ is included in the model on “Same Chamber,” which ensures that the ERGM will not simulate within-chamber connections. A burn-in period of 75,000 MCMC iterations and an MCMC sample of 50,000 were used. AIC = Akaike information criterion; BIC = Bayesian information criterion; ERGM = exponential random graph model; MCMC = Markov chain Monte Carlo; MLE = Maximum likelihood estimation.

*

p < .05.

As the interpretation of coefficients from an ERGM can be somewhat challenging, we walk through several results carefully. Dyadic covariates in ERGMs typically reveal whether the network features an unusually high number of connections of a particular type. Thus, the positive and significant coefficient on the same-party covariate implies that the companion bill network has an unusually large number of same-party connections relative to a randomly generated network. In other words, connections in the companion bill network are generally quite partisan. Legislators from the same state have a strong penchant to introduce companion bills together. Two legislators’ concurrent service in the same chamber has little impact on the probability that they are connected to one another in the companion bill network, and we see some evidence that the ideological similarity between two legislators’ constituents played a role in structuring companion bills relationships in the 112th, but not the 111th session of Congress, controlling for the same-state covariate. ¹³ Ultimately, two of our four dyadic hypotheses received strong support in the data, while a third (constituent ideological influences) received some support. We have clear evidence that legislators from the same party and same state introduce companion bills together quite often, but our other expectations regarding the overlap of service in a chamber as a driver of partner selection in the network failed to see much empirical support.

The actor-specific covariates evaluate whether actors of a particular type were unusually likely to have connections to anyone in the network, rather than other actors of a similar type. Thus, the positive effects of the total bills introduced covariate indicate that legislators who introduced more legislation generally in their chambers are more likely to develop companion bill connections to other members. In addition, our results indicate that in the 112th Congress, legislators with larger two-party vote shares were also more likely to introduce companion bills, while ideological extremists (measured by lagged DW-Nominate scores) were less likely to introduce companion bills. ¹⁴ While the total bills introduced covariate is the only nodal covariate to be statistically significant in the 111th Congress, both the two-party vote share and ideological extremity covariates are in the expected direction. Thus, we see some evidence that legislators with larger vote shares were also more likely to introduce companion bills and ideological extremists were less likely to introduce companion bills. These empirical results square extremely well with our hypotheses, and suggest that legislators who are paying significant opportunity costs to increase the odds of their own legislative success are most likely to engage in cross-chamber collaboration.

Terms capturing the interdependence in the network are the final type of covariate in the model. In the companion bill network, our results indicate that our networks contain a large number of two-star formations relative to networks with randomly generated ties. This is precisely what we expected based on our discussions of Figure 2. Our focus on cross-chamber collaboration has induced a tendency toward star formations. Interestingly, our results indicate that controlling for the prevalence of two stars in the companion bill network, the network shows a strong tendency against three-star formation. This suggests that companion bill clusters tend to form between small numbers of actors (two House members and one senator, for example), and there is a tendency in the network against developing large numbers of different companion bill partners. Said differently, actors in the companion bill network tend to form connections to a limited number of partners, rather than having expansive networks in the opposing chamber.

In sum, our results provided some support for three of our four dyadic hypotheses, and for two of our nodal hypotheses. While the coefficients and standard errors from ERGMs are useful for determining statistical significance, the magnitude of these effects is difficult to gauge from coefficients alone. One useful way to make dyadic and interdependence coefficients from an ERGM more easily interpretable is to evaluate the multiplicative change in the predicted network using the model’s coefficients relative to a network simulated from a model with the relevant coefficient set to zero (Desmarais & Cranmer, 2012; Kirkland & Williams, 2014).

The method for interpretation works as follows. First, using the model’s actual coefficients, simulate a predicted network. Second, set a coefficient of interest to zero while keeping all the other coefficients at their observed values, and simulate a second network using the altered set of coefficients. Finally, evaluate the relationship between the simulated network from the observed coefficients and the simulated network from the altered coefficients. An analyst can repeat this process hundreds of times to develop confidence intervals around the predicted multiplicative changes. So, if we were interested in seeing the magnitude of the same-party coefficient, we would compare the number of same-party connections in a simulated network using the observed coefficients and a simulated network with the same-party coefficient set to zero. This would allow us to assess how many more same-party connections are implied by our model relative to a model with no party effects.

To show the magnitude of some of the effects in Table 1, Figure 3 presents these simulated multiplicative changes for the same-party coefficient and the two- and three-star coefficients from the model of the 112th companion bill network. We focus our interpretation on the 112th Congress, as that model’s predictive performance is slightly better than the model of the 111th network, and because for the dyadic and interdependence coefficients, our results were largely the same across both sessions. We simulate 2,500 predicted networks to provide confidence intervals around the multiplicative changes. As the figure indicates, there are roughly 12 times as many two-star formations in the network as there would be in a network with no tendency toward star formation, and a fraction very close to zero of the number of three-star formations in the network relative to a network with no three-star formation pattern. Again, this suggests an active resistance toward three-star patterns in the network, suggesting that companion bill groups tend to be quite small, typically involving only three members. Finally, we see that there are roughly 4 times as many same-party dyads connected in the network as we would expect in a network where party did not affect companion bill network connections at all. This suggests that party is an extremely strong predictor of companion bill partnerships. ¹⁵

OPEN IN VIEWER

Figure 3.

Magnitude of network model effects on multiplicative scales.

While this simulation approach is useful for summarizing the substantive effects of dyadic and interdependence coefficients in an ERGM, nodal covariates can be interpreted in a slightly more straightforward way. The connections between ERGMs and logistic regressions are such that nodal covariates in an ERGM can be re-expressed as the change in the log-odds of a connection between the actors in the network, much like coefficients from a logit model. For nodal covariates, this immediately implies that we can express these coefficients as changes in the log-odds or predicted probability of an actor in the network having a connection to anyone in the network. ¹⁶ Accordingly, Figure 4 plots the change in the predicted probability of an actor having any connection in the 111th and 112th companion bill networks as that actor’s ideological extremity, tenure, and two-party vote share move from their minimum observed value to the mean observed value of the relevant covariate, along with its 95% confidence interval. For example, moving from the minimum to the mean value of tenure among members of Congress is associated with an increase in the probability of having a connection in the 112th companion bill network of roughly 0.08, and an increase in the probability of having a connection in the 111th companion bill network of 0.11. Importantly, we see that while two of the observed effects are statistically insignificant, the results appear to be of similar magnitudes, and are in the same direction across models. This suggests that these patterns are likely to be consistent across Congresses. The composition of the 112th Congress is different from the 111th, given that the 112th was a divided Congress and the median member shifted, making this consistency in the findings indicative that these patterns hold across divergent settings.

OPEN IN VIEWER

Figure 4.

Changes in predicted probability of having a connection in the companion bill network for changes in nodal covariates from minimum to mean values.

As a final step in presenting our models of the companion bill network, Figure 5 provides goodness-of-fit distributions for our networks. A valuable way to demonstrate the goodness-of-fit for ERGMs of a network is to plot the distribution of network characteristics not estimated by the model, and compare them with the predicted distribution of those characteristics from networks generated from the model’s coefficients. If the model does a good job of recovering unmodeled characteristics of the network, it is likely the model fits the network well. In our goodness-of-fit plots, we provide degree and geodesic distance distributions from our models of the 111th and 112th companion bill networks. The solid line in the figures represents the distribution in the actual companion bill networks, while the box-and-whiskers plot provides the estimated proportion of nodes having that value in our predicted networks. The predicted distributions of both degree and geodesic distance from the models track the actual values of these distributions from the true network quite well. Table 1 also provides statistics on the number of connections in the network correctly predicted by the model relative to a model with just the edge coefficient. Each model makes substantial gains in predictive accuracy with the model of the 112th Congress companion bill network correctly predicting nearly 26% of ties in the network, while the null model predicts less than 1% of ties correctly. This all suggests that our models are substantial improvements over null models of the companion bill networks.

OPEN IN VIEWER

Figure 5.

Goodness-of-fit distributions for companion bill network models.

Companion Bills and Bill Survival

While the bulk of our hypotheses focused on the formation of the companion bill network, an important hypothesis focuses on the consequences of companion bill introductions. Our theory suggests that legislators introduce companion bills to increase the odds that their legislation will survive deliberation in their chamber, and that long-tenured, electorally secure legislators are most likely to develop such bills. This suggests that the introduction of a Senate companion to a House bill will increase the odds of that House bill’s survival during House deliberations. As we described earlier, to test this hypothesis, we combined our data on companion bill introductions with data on bill survival from the Congressional Bills Project. Using these data, we create three distinct regression models.

In the first model, we predict the number of cosponsors that appear on a bill as a function of the presence of a companion bill in the Senate. As we assert in our theoretical development, the introduction of a companion bill is intended to signal to a legislator’s home chamber robust support in the opposing chamber. This signal ought to generate greater levels of prefloor support for a bill. In other words, companion bills ought to have more support via cosponsorship than House bills without Senate companions. We then predict the probability that House bills will be reported out by a committee in the House as a function of both the presence of a Senate companion to the House bill and the number of cosponsors. By incorporating the number of cosponsors at this stage of modeling, we allow the presence of a Senate companion to a House bill to have both direct effects on the probability of a bill being reported out of committee, and indirect effects on that probability through increased cosponsorship. In the final stage of our model, we predict the probability that a bill will pass in the House as a function of the presence of a Senate companion to the bill, the number of cosponsors on the bill, and whether the bill has been reported out by a committee. ¹⁷ This again allows the presence of a Senate companion to have a direct effect on bill passage, and indirect effects on bill passage through cosponsorship and committee reports. Thus, we develop a three-stage model of legislative success in which we predict early support for a bill, committee support for a bill, and floor support for a bill as a function of the presence of a Senate companion to that House bill. This approach should allow us to capture the multistage direct and indirect benefits of pairing a House bill with a Senate companion.

In addition to our key covariates, our models also include control variables measuring characteristics of the sponsors of each bill that might influence the support a piece of legislation has in the House. We include the party of the sponsor (coded one for Democrats and zero for Republicans), a dummy variable indicating whether the bill’s sponsor is a legislative leader, the ideological extremity of the bill’s sponsor measured as the absolute value of the sponsor’s lagged DW-Nominate score, a dummy variable indicating the sponsor’s status as a committee chair, a dummy variable indicating whether the sponsor of the bill was female, the tenure of the sponsor of the bill, and the two-party vote share of the sponsor in his or her most recent election. Each of these controls is included at all three stages of our multistage model. Each has also been found to play a role in the probability of bill survival in prior research on legislative success (Anderson et al., 2003; Kirkland, 2011; Volden, Wiseman, & Wittmer, 2013). ¹⁸

The results of these models appear in Table 2. We model the number of cosponsors on a bill using a quasi-Poisson link function that accounts for potential overdispersion in our data (King, 1989), and model the probability of a bill being reported out by committee and the probability of a bill passing in the House using a standard logistic regression. For both the 111th and 112th Houses, we see a strong positive and significant effect of Senate companions on the number of cosponsors a House bill receives. Second, the results suggest that there is no discernible direct effect of Senate companions on either House bills being reported out by committee or passing in the chamber. Importantly, however, the number of cosponsors on a House bill is a strong positive and significant indicator of House bills being reported out by committee and ultimately passing in both the 111th and 112th Houses. This suggests that the pathway through which Senate companions relate to bill survival at the committee and floor stages in the House is indirect, and is generally realized through the increase in prefloor support expressed through cosponsorship that a House bill with a Senate companion receives. ¹⁹

OPEN IN VIEWER

Table 2.

The Effect of Companion Bills on Several Stages of the Legislative Process (111th and 112th U.S. House).

Variable name	111th House			112th House
	No. of cosponsors	Committee report	Passage	No. of cosponsors	Committee report	Passage
Senate companion	0.641*	0.209	−0.067	0.867*	0.224	0.008
	(0.073)	(0.157)	(0.213)	(0.079)	(0.188)	(0.263)
Sponsor is Democrat	0.181	0.039	0.131	−0.218*	−1.534*	−1.039*
	(0.101)	(0.186)	(0.242)	(0.073)	(0.167)	(0.252)
Sponsor is leader	0.806*	0.348	−0.744	1.008*	2.111*	2.762*
	(0.148)	(0.361)	(0.641)	(0.283)	(0.514)	(0.641)
Sponsor’s ideological extremity	0.314	−1.361*	−0.483	−0.007	−0.119	−0.080
	(0.228)	(0.422)	(0.552)	(0.200)	(0.345)	(0.469)
Sponsor is committee chair	0.068	1.393*	0.659*	0.224	1.535*	0.171
	(0.124)	(0.181)	(0.245)	(0.123)	(0.151)	(0.240)
Sponsor is female	0.150*	−0.106	−0.174	0.217*	−0.094	−0.226
	(0.071)	(0.139)	(0.183)	(0.081)	(0.180)	(0.269)
Sponsor’s tenure	0.008*	0.011	0.011	0.005	0.018*	−0.004
	(0.004)	(0.006)	(0.008)	(0.003)	(0.007)	(0.010)
Sponsor’s two-party vote share	−0.001	0.007	−0.012*	0.005*	−0.005	0.011
	(0.002)	(0.004)	(0.005)	(0.002)	(0.005)	(0.007)
No. of cosponsors on bill	—	0.003*	0.005*	—	0.003*	0.011*
	(—)	(0.001)	(0.001)	(—)	(0.001)	(0.001)
Bill reported out by committee	—	—	2.863*	—	—	3.427*
	(—)	(—)	(0.146)	(—)	(—)	(0.187)
Intercept	2.294*	−2.850*	−2.782*	2.139*	−2.247*	−4.795*
	(0.182)	(0.327)	(0.426)	(0.199)	(0.388)	(0.573)
AIC	199,069	2,551.8	1,601.5	192,131	2,162.2	1,075.1
BIC	199,127	2,616.7	1,672.9	192,189	2,227.3	1,146.7
Number of bills	6,562	6,562	6,562	6,618	6,618	6,618

Note. Cell entries report coefficient values from three models of legislative support in the 111th and 112th U.S. House. Columns 1 and 4 report results from a quasi-Poisson model of the number of cosponsors appearing on a bill. Columns 2 and 5 report logistic regression coefficients predicting the probability that a House bill will be reported out by committee. Columns 3 and 6 report logistic regression coefficients predicting the probability that a bill will pass in the House. Standard errors are reported in parentheses. AIC = Akaike information criterion; BIC = Bayesian information criterion.

*

p < .05.

Thus, our models suggest a chain of events surrounding House bills with Senate companions. When a House bill is introduced with a Senate companion, that House bill receives much greater early support from House members, and ultimately winds up with more cosponsors. This increased cosponsorship activity is strongly associated with bills being reported out of committee, and is strongly associated with bills eventually passing in the House. Thus, while we do not see a direct significant effect of Senate companions on a bill being reported out by the committee or passing through the House, the presence of a companion is significantly positively related to the number of cosponsors on the bill. In turn, the size of the cosponsor coalition is significantly related to passing through these legislative hurdles. Given that passing a bill is a highly difficult and low-probability event, the circuitous process of lawmaking, including initial support for the bill, is an important process to consider in the analysis of a bill’s progress.

This indirect effect of Senate companions on House bill survival makes it difficult to assert just how much Senate companions matter for House bills’ survival directly from the model results. To make this indirect effect more readily apparent, we choose several House members from the 111th House, and estimate based on the model’s results the number of cosponsors, probability of a committee report, and probability of passage of a bill introduced by that member with and without a Senate companion. ²⁰ We choose four members from each party in the House, and then estimate predicted values from our multistage model of bill survival in the 111th House. ²¹

The differences in our predictions for each member are based only on the differences created by the introduction of companion, rather than independent legislation, and the differences that introduction creates compounding through each stage of the series of models. For example, for Jeff Flake (R-AZ), the introduction of a House bill with a companion rather than without, is worth an additional 13 cosponsors for his legislation. This companion bill with its 13 extra cosponsors increases his odds of having his bill reported out by a committee from 0.027 to 0.035. While this is not a large jump in an absolute sense, it is more than a 25% increase in his odds of having his bill reported out by a committee. Flake’s bill with its Senate companion, additional 13 cosponsors, and slightly elevated chance of being reported out by a committee has an increased predicted probability of passing in the chamber from 0.020 to 0.021. Again, this change is tiny, but is a 5% increase in the probability that Flake’s bill will pass in the House (Table 3). ²² Given the remarkably low odds of a sponsor having his or her bill pass in the House, a 5% increase in the probability of bill survival is likely to be one House members value.

OPEN IN VIEWER

Table 3.

The Predicted Change in Bill Support as a Function of Senate Companions (111th U.S. House).

Variable name	Predicted no. of cosponsors	Predicted probability of committee report	Predicted probability of passage
Jeff Flake (R-AZ)
No companion	14.4	0.027	0.020
Companion	27.3	0.035	0.021
Adam Putnam (R-FL)
No companion	12.8	0.043	0.028
Companion	24.4	0.055	0.029
Joseph Pitts (R-PA)
No companion	14.09	0.037	0.028
Companion	26.77	0.047	0.029
Joe Wilson (R-SC)
No companion	13.37	0.035	0.027
Companion	25.41	0.044	0.028
Mark Critz (D-PA)
No companion	13.00	0.064	0.036
Companion	24.69	0.080	0.038
Kendrick Meek (D-FL)
No companion	13.77	0.084	0.023
Companion	26.16	0.105	0.025
Ed Pastor (D-AZ)
No companion	14.23	0.063	0.028
Companion	27.04	0.080	0.029
Adam Smith (D-WA)
No companion	14.09	0.079	0.039
Companion	26.78	0.100	0.042

Note. Cell entries report predicted values of the dependent variable for specific House members based on the models reported in Table 5. Each member’s covariate value is used to predict each dependent variable under the conditions in which a member sponsors a bill with and without a Senate companion.

To summarize our prediction exercise succinctly, for the eight members we consider here, simply pairing a House bill with a Senate companion creates somewhere between 12 and 13 additional cosponsors in the 111th House, increases the chances that a bill will be reported out by committee by between 25% and 28%, and increases the chances that a bill will ultimately pass in the House by between 3.5% and 8.7%. These effects all accrue indirectly through companion bills’ increased support in the early stages of the legislative process in the House. This does suggest that there are discernible differences in the probability that companion bills will survive the legislative process in their home chambers, though the odds of survival are small for all legislation. ²³ This offers empirical support to our hypothesis suggesting that legislators introduce companion bills as an effort to increase the chances that their legislation will receive consideration and potentially pass in their home chambers. ²⁴

Hearings as Informal Support

Our primary models of bill passage focus on the effect of opposing chamber companions on bill passage through increased cosponsorship. We suggest that the presence of an opposing chamber companion bill helps increase a bill’s coalition of support within its own chamber, the downstream effects of which are considerable. However, it is possible that cosponsorship is a poor indicator of informal chamber support for a piece of legislation. The motivations behind cosponsorship are nuanced, and some suggest that cosponsorship is simply cheap talk from legislators (Koger, 2003), as opposed to a real credible signal of support (Bernhard & Sulkin, 2013). As an alternative indicator of informal support for a bill, Table 4 replicates our models of passage from Table 2, but replaces the number of cosponsors on a bill with a dummy variable indicating whether the bill received a hearing in committee. We should note before discussing results that there is a strong association between the number of cosponsors and bills receiving hearings, with bills in the 111th House receiving hearings averaging 24 cosponsors and bills without hearings receiving 16 cosponsors, while bills in the 112th House receiving hearings averaging 19 cosponsors and those without hearings receiving an average of 13 cosponsors.

OPEN IN VIEWER

Table 4.

The Effects of Companion Bill Presence on Bill Passage.

	111th House			112th House
	Hearings	Committee report	Final passage	Hearings	Committee report	Final passage
Companion bill	0.819†	−0.118	0.067	0.318†	0.259	0.252
	(0.157)	(0.182)	(0.212)	(0.184)	(0.198)	(0.259)
Sponsor is a Democrat	0.422†	−0.152	0.128	−0.908†	−1.416†	−1.065†
	(0.228)	(0.204)	(0.242)	(0.149)	(0.178)	(0.247)
Sponsor is leader	−0.210	0.630	−0.564	0.591	2.447†	3.253†
	(0.520)	(0.393)	(0.627)	(0.763)	(0.541)	(0.639)
Sponsor’s extremity	−0.248	−1.660†	−0.495	−0.316	−0.095	−0.189
	(0.503)	(0.475)	(0.553)	(0.350)	(0.369)	(0.486)
Sponsor is committee chair	0.523†	1.543†	0.650†	0.546†	1.692†	0.156
	(0.250)	(0.202)	(0.246)	(0.201)	(0.170)	(0.241)
Sponsor is female	0.084	−0.140	−0.134	−0.431†	0.032	−0.182
	(0.153)	(0.155)	(0.182)	(0.195)	(0.192)	(0.264)
Sponsor’s tenure	−0.0004	0.016†	0.011	−0.006	0.024†	−0.010
	(0.008)	(0.007)	(0.008)	(0.008)	(0.007)	(0.010)
Sponsor’s two-party vote share	−0.003	0.011†	−0.012†	−0.010†	−0.006	0.011†
	(0.005)	(0.005)	(0.006)	(0.005)	(0.006)	(0.007)
Committee hearing for bill		3.313†	−0.250		2.908†	−0.549†
		(0.151)	(0.219)		(0.154)	(0.240)
Bill reported out by committee			2.965†			3.570†
			(0.166)			(0.203)
Intercept	−3.037†	−3.294†	−2.722†	−1.581†	−2.817†	−4.391†
	(0.392)	(0.363)	(0.423)	(0.388)	(0.432)	(0.558)
Log likelihood	−985.372	−1,039.372	−794.693	−1,058.694	−907.166	−547.720
AIC	1,988.744	2,098.743	1,611.386	2,135.389	1,834.332	1,117.440

Note. Cell entries report results from logistic regression models predicting whether (a) a bill receives a hearing in committee, (b) whether a bill is reported out by the relevant committee, and (c) whether a bill passes on the floor of the House. Standard errors are reported in parentheses. AIC = Akaike information criterion.

†

p < .10.

We see from the models reported in Table 4 that House bills with Senate companions are more likely to receive committee hearings than those without Senate companions, and that the existence of a committee hearing increases the odds a bill will be reported out by committee and passed in the House. Thus, using this alternative signal of support in the House, we again see that the presence of a companion bill is associated with an increased probability of bill passage through expressions of support early in the legislative process for the bill. Whether that signal of early support is via a committee hearing, or increased cosponsorship coalitions, in either case, House bills with Senate companions do seem to have a greater chance of surviving the legislative process.

A Matching Analysis of Bill Survival

The prior analyses suggest that the direct effect of the introduction of a bill in the House of Representatives paired with a Senate companion is that the said bill is more likely to be supported by a large coalition of cosponsors (and a greater probability of a hearing in the House), an effect that has considerable downstream benefits for the bill’s likelihood of survival (Kessler & Krehbiel, 1996; Kirkland, 2011). However, our regression analysis is not particularly well suited to drawing causal inferences about the effects of companion bills (Ho, Imai, King, & Stuart, 2007; Morgan & Winship, 2014; Rubin, 2006). To try and push our inferences closer to assertions of causality, we now leverage matching analyses to evaluate the average treatment effect of having a Senate companion on the number of cosponsors a bill receives. Companion bills may differ from bills without a companion on a number of variables that are importantly related to the number of cosponsors who sign onto a bill. Thus, we turn to matching to compare similar types of bills with one another, differing only on the presence of a Senate companion. ²⁵

To match bills with companion to those without, we utilize genetic matching (Diamond & Sekhon, 2013; Mebane & Sekhon, 2011). We match bills in the treatment group (those with Senate companions) to the control group (those without Senate companions) on all of the covariates listed in Table 2. We match each bill in the treatment group to one bill in the control group. Thus, at the conclusion of the genetic matching routine, we have a balance-optimized set of treatment and control bills, and each bill with a Senate companion has been matched as nearly as possible on each of the covariates to a House bill without a Senate companion. ²⁶ This exercise allows us to approximate causal inferences with our observational data. We then evaluate the effects of the treatment (the presence of a Senate companion) on the number of cosponsors a bill receives.

To ensure that the treatment and control groups look more similar after matching, we calculate a few balance statistics. ²⁷ Table 5 reports the increased balance between the treatment and control groups that results from our genetic matching efforts for both the 111th and 112th Houses. These balance results report the difference between the treatment and control groups of bills before and after matching. We include the standardized mean differences and differences in empirical QQ values. Specifically, the standardized mean difference, “the difference in means of each covariate, divided by the standard deviation in the full treated group,” is calculated for each covariate before and after matching (Stuart, 2010). Researchers recommend different cutoff points of the standardized mean difference to gauge imbalance, Rubin (2001) suggesting 0.25 and Normand et al. (2001) suggesting 0.10. After matching, all of our standardized difference in means are less than 0.10. The mean empirical QQ difference balance statistic represents the mean difference between the empirical quantile functions of the treated versus control groups. Overall, Table 5 shows notable improvements in balance after matching. On all of the covariates on which we match bills, the reported statistics show decreases in difference between treatment and control groups after matching.

OPEN IN VIEWER

Table 5.

Balance Statistics From Genetic Matching.

Variable	111th House				112th House
	Before		After		Before		After
	(a)	(b)	(a)	(b)	(a)	(b)	(a)	(b)
Sponsor is Democrat	21.558	0.091	0	0	3.411	0.02	0	0
Sponsor is leader	10.031	0.017	0	0	8.362	0.007	0	0
Ideological extremity	−16.995	0.029	0.041	0	−25.899	0.039	0	0
Sponsor is committee chair	−0.987	0.003	0	0	−5.103	0.011	0	0
Sponsor is female	2.743	0.012	0	0	−14.436	0.052	0	0
Sponsor’s tenure	5.681	0.958	0.063	0.007	−0.572	0.725	0	0
Sponsor’s vote share	13.388	1.859	0.019	0.033	10.156	1.449	0	0

Note. Cell entries report balance statistics before and after genetic matching for the number of cosponsors a bill receives. (a) = Standardized difference in means between treatment and control groups; (b) = Mean empirical QQ difference between treatment and control groups.

Given the considerable gains in balance across the treatment and control groups, this matching exercise is likely to produce considerably better inferences than regression alone. We now turn to the estimation of the average treatment effect. Table 6 presents the genetic matching routines estimate of the average treatment effect on the treated (ATT). As the table indicates, in both the 111th and 112th Houses, the treatment effect of a Senate companion added to a House bill increases the number of cosponsors that House bill receives by roughly just over 11 cosponsors in the 111th House and just over 13 cosponsors in the 112th House relative to House bills without Senate companions. Both of these effects are significantly different from zero and are quite similar to the results reported on our regression analyses.

OPEN IN VIEWER

Table 6.

Estimated Treatment Effects After Genetic Matching.

Session	Estimate (ATT)	SE	t statistic	p value
111th House	11.63	2.79	4.22	.00
112th House	13.61	2.74	4.96	.00

Note. Cell entries report the estimated average treatment effects on the treated after matching. The dependent variable is the number of cosponsors a bill receives. The treatment is the presence of a Senate companion to a House bill. ATT = average treatment effect on the treated.

Perhaps more important than the estimated treatment effects themselves, our matching framework allows us to evaluate the impact of hidden confounders (unobserved factors that predict both treatment assignment and the outcome measure) on the inferences we draw. Rosenbaum (2002) provides a means of assessing the sensitivity of effects estimated from matched data to this type of confounder. Briefly, this method estimates lower and upper bounds indicating the degree to which an estimated causal effect (and its p value) would change if the odds of selection into treatment increased due to some hidden confounder. Table 7 reports Rosenbaum bounds for the estimates in Table 6. We vary the parameter Γ, which indicates the increase in the odds of selection into treatment due to a hidden confounder, from 1 to 2 in increments of 0.1. This is a large range compared with other applications in social sciences (Keele, 2014).

OPEN IN VIEWER

Table 7.

Rosenbaum Bounds for the Estimated Treatment Effects After Genetic Matching.

Γ	Effect (lower)	Effect (upper)	p value (lower)	p value (upper)
111th House
1.0	5	5	0	0
1.1	2.9	5.1	0	0
1.2	2.4	6	0	0
1.3	1.4	7	0	0
1.4	0.9	8.5	0	0
1.5	0.4	9.5	0	0
1.6	−0.09	10.5	0	0.10
1.7	−0.59	11.5	0	0.89
1.8	−1.09	12.5	0	0.99
1.9	−1.09	13.5	0	1
2.0	−1.59	14.5	0	1
112th House
1.0	10	10	0	0
1.1	7	10.1	0	0
1.2	6	10.6	0	0
1.3	5	11.6	0	0
1.4	4	12.6	0	0
1.5	3.5	14.1	0	0
1.6	3	15.1	0	0
1.7	2	16.1	0	0
1.8	1.5	17.6	0	0
1.9	1.5	18.6	0	0
1.0	0.99	19.6	0	0

Note. Cell entries report upper and lower Rosenbaum bounds for the estimated treatment effects and p values.

The results of our sensitivity analysis using Rosenbaum bounds indicate several important points about the strength of our inferences. First, for both the 111th and 112th Houses, some unobserved confounder would have to make bills that received a Senate companion at least 1.5 times as likely to opt into the “treatment” group as “control” bills before those confounding variables would pose a threat to our inferences. Thus, our inferences are rather robust to hidden confounding variables. Second, the threat of hidden confounding is much steeper in the 111th House than in the 112th House, where there is virtually no threat of confounding. As such, there may be some important unexplored differences in the motivation for companion bill introductions across session of the House. Nonetheless, our matching analysis paired with our sensitivity analysis using Rosenbaum bounds strongly suggests that there is a causal relationship between the contemporaneous introduction of a Senate companion to a House bill and that House bill’s number of cosponsors in its home chamber.

The Content of Companion Bills: National and Local Orientation and Topic Area

Our theory and analysis concentrate on two questions: (a) Who works together on companion bills, and (b) what impact do those bills have on legislating? Given that we find House bills with Senate companions to be more likely to pass in the House, this begs the question, “What are the topics of companion bills?” We begin by differentiating between local and national impacts of companion bills. This is our initial focus given the predictive power of the same-state covariate in our models of companion bill partnerships. Delineating between local and nationally oriented legislation is a labor-intensive task (Gamm & Kousser, 2010), particularly given that more than 13,000 bills were introduced between the 111th and 112th sessions of just the U.S. House. In addition, the threshold of geographic impact for bill to be considered local rather than national is open to interpretation. To both reduce the time necessary to identify local legislation and to provide an objective measure of the geographic scope of legislation, we compile a list of every U.S. city with a population more than 40,000 people and append to this list the name of every U.S. state. ²⁸ We identify a bill as being locally focused if the bill’s title mentions any of these places by name. We focus on bills introduced in the House in either the 111th or 112th sessions of Congress, which ultimately leaves us with 539 local bills in the 111th session of the House and 433 local bills in the 112th session. ²⁹ Using this simple coding of local legislation, we predict the probability that a bill introduced in the House has a Senate companion as a function of the bill being locally oriented using standard logistic regression models. The results of this simple model appear in Table 8.

OPEN IN VIEWER

Table 8.

Logistic Regression Predicting Whether House Bill Has a Senate Companion in the 111th and 112th U.S. Congress.

Variable name	111th Congress	112th Congress
Local legislation	0.542*	0.552*
	(0.119)	(0.137)
Intercept	−2.071*	−2.198*
	(0.041)	(0.042)
AIC	4,730	4,406
BIC	4,744	4,420
N	6,562	6,618

Note. Cell entries report coefficient values from a logistic regression model predicting the existence of a Senate companion for a House Bill 111th and 112th U.S. Congress. Standard errors are reported in parentheses. AIC = Akaike information criterion; BIC = Bayesian information criterion.

*

p < .05.

As the table shows, the probability that a House bill has a Senate companion is significantly higher for local legislation than for nonlocal legislation. The positive and significant coefficient on the local legislation dummy variable indicates that House bills that included city or state names in the title were particularly likely to have Senate companions associated with them. Indeed, using each models’ coefficients to calculate the probability of a House bill having a Senate companion, nonlocal House bills had a 0.112 and 0.099 probability of having a Senate companion in the 111th and 112th Congresses, respectively. Local House bills, however, had a 0.178 and 0.161 probability of having a Senate companion associated with them. This strongly supports the notion that companion bills are unusually likely to be local in orientation. This is, of course, a simple specification, but does provide descriptive evidence that companion bills are likely to be local in nature.

Now, we examine the substantive content of bills. To do so, Table 9 indicates the percentage of bills introduced in both Congresses falling into each of the 20 major topic categories of the Policy Agendas Project (2008). The largest difference between companion and noncompanion legislation in their topics of focus is in public lands legislation. In both sessions, only just over 6% of House bills without Senate companions focus on public lands. However, in the 111th House, more than 14% of House bills with Senate companions focus on public lands and 10.4% of House bills with Senate companions focus on public lands in the 112th House. This would seem to echo our earlier point that most legislation that is paired with an opposing chamber companion is local in orientation, concentrating benefits to particular geographies. House bills with Senate companions also seem to have a slightly elevated likelihood of focusing on health care issues. Otherwise, the topics of companion bill legislation look rather similar to bills introduced without companions, with one notable exception. The 112th House was unusually focused on the introduction of bills focused on foreign trade, such that more than 23% of bills introduced without Senate companions concentrated on this issue, while House bills with Senate companions were fairly unlikely to concentrate on this topic. Indeed, the raw data from the Policy Agendas Project suggest the 112th House had more than 1,300 pieces of legislation introduced focused on foreign trade, while the 111th session of the House had only 147 such bills. The 112th session saw a number of free trade agreements passed that may account for this large spike in the topic of legislation (i.e., The U.S.-Korea Free Trade Agreement among many others).

OPEN IN VIEWER

Table 9.

Percentage of Bills in Each Major Topic Category for Companion and Noncompanion Legislation.

	111th House		112th House
	% noncompanion bills	% companion bills	% noncompanion bills	% companion bills
1. Macroeconomics	2.1	4.5	4.1	5.1
2. Civil Rights	1.6	0.9	1.5	2.2
3. Health	11.9	17.1	8.4	11.9
4. Agriculture	1.1	1.4	1.3	3.0
5. Labor	5.5	3.9	5.1	6.9
6. Education	5.3	4.8	3.7	5.2
7. Environment	4.3	4.6	3.5	4.3
8. Energy	5.5	5.3	4.1	5.2
9. Immigration	0.0	0.0	0.0	0.0
10. Transportation	3.3	3.7	3.7	5.9
12. Law and Crime	5.4	5.3	4.5	4.5
13. Social Welfare	2.6	2.4	2.0	2.6
14. Housing	3.3	1.9	2.3	2.3
15. Domestic Commerce	9.7	5.4	6.2	6.4
16. Defense	8.1	7.2	6.0	7.2
17. Technology	1.9	1.9	1.4	1.1
18. Foreign Trade	2.3	1.6	23.1	2.7
19. International Affairs	2.5	1.5	2.4	1.6
20. Government Operations	11.6	10.9	9.2	10.9
21. Public Lands	6.8	14.4	6.2	10.4
23. Culture	1.4	0.5	0.6	0.0

Senate Bill Passage

Our initial results in the main body of our article demonstrate that House bills with Senate companions are unusually likely to succeed in their home chamber. This naturally begs the questions of whether Senate bills with House companions experience similar increases in the probability of passage. ³⁰ To test this hypothesis, Table 10 replicates our models in Table 2, but the covariates all now related to Senate bills and their sponsors. Much as we observed for our House bill analysis, Senate bills with House companions are more likely to have large cosponsorship coalitions, which in turn appears to be associated with a greater chance that a Senate bill will be reported out by committee, and ultimately pass in its chamber. Interestingly, we also see that in the 111th Senate, companion bills also have a direct effect on committee reports and final passage, as opposed to the exclusively indirect effects of companion bills we observe in both sessions of the House and in the 112th Senate.

OPEN IN VIEWER

Table 10.

The Effect of Companion Bills on Several Stages of the Legislative Process (111th and 112th U.S. Senate).

Variable name	111th Senate			112th Senate
	No. of cosponsors	Committee report	Passage	No. of cosponsors	Committee report	Passage
Senate companion	0.490*	0.304*	0.592*	0.450*	−0.100	0.350
	(0.064)	(0.133)	(0.194)	(0.066)	(0.163)	(0.225)
Sponsor is Democrat	−0.058	−0.482*	−0.301	0.291*	−0.360	0.570*
	(0.079)	(0.177)	(0.282)	(0.075)	(0.196)	(0.272)
Sponsor is leader	0.633*	0.179	0.929*	0.313	−0.097	−0.201
	(0.131)	(0.299)	(0.396)	(0.167)	(0.382)	(0.655)
Sponsor’s ideological extremity	0.166	−0.616	−0.144	0.254	−0.579	0.250
	(0.221)	(0.456)	(0.750)	(0.188)	(0.520)	(0.696)
Sponsor is committee chair	0.366*	0.502*	0.368	−0.184*	0.207	0.390
	(0.080)	(0.151)	(0.242)	(0.084)	(0.169)	(0.273)
Sponsor is female	−0.139	−0.043	0.334	0.021	0.045	0.047
	(0.075)	(0.145)	(0.217)	(0.080)	(0.162)	(0.248)
Sponsor’s tenure	−0.000	0.001*	0.000	0.000	0.001*	0.000
	(0.001)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Sponsor’s two-party vote share	−0.003	0.001	−0.010*	0.006*	−0.012*	−0.011
	(0.002)	(0.005)	(0.007)	(0.003)	(0.006)	(0.008)
No. of cosponsors on bill	—	0.022*	0.031*	—	0.022*	0.021*
	(—)	(0.005)	(0.007)	(—)	(0.006)	(0.008)
Bill reported out by committee	—	—	2.723*	—	—	2.985*
	(—)	(—)	(0.182)	(—)	(—)	(0.197)
Intercept	1.434*	−2.928*	−4.118*	0.832*	−2.752*	−4.230*
	(0.174)	(0.345)	(0.531)	(0.183)	(0.433)	(0.578)
AIC	36,722	2,254.7	1,011.8	37,389	1,958.3	939.4
BIC	36,780	2,316.9	1,080.2	37,444	2,020.4	1,007.7
Number of bills	4,057	4,057	4,057	3,713	3,713	3,713

Note. Cell entries report coefficient values from three models of legislative support in the 111th and 112th U.S. Senate. Columns 1 and 4 report results from a quasi-Poisson model of the number of cosponsors appearing on a bill. Columns 2 and 5 report logistic regression coefficients predicting the probability that a Senate bill will be reported out by committee. Columns 3 and 6 report logistic regression coefficients predicting the probability that a bill will pass in the Senate. Standard errors are reported in parentheses. AIC = Akaike information criterion; BIC = Bayesian information criterion.

*

p < .05.

This difference may be a result of repeated sampling (over a repeated set of random draws, some coefficients will appear to be statistically significant, when there is no underlying data generating relationships), or may reflect something unique about the Senate’s use of companion bills relative to the House. Nevertheless, we see across both the House and the Senate, across both sessions we examine, opposing chamber companion bills are strongly associated with greater within-chamber support, which in turn translates into an elevated probability of bill passage. Thus, our empirical consistently suggests that having support in the opposing chamber of Congress increases support for a bill within a chamber of Congress.

Discussion

The introduction of companion bills in Congress provides us with an opportunity to understand how members in typically antagonistic chambers occasionally find room to collaborate in the policy process. We offer a theory of this collaborative environment that suggests House members and Senators collaborate on legislation when they (a) wish to improve the odds that their legislation will be successful and (b) are comfortable paying the costs associated with finding companion bill partners. Using a novel data set of companion bill introductions in the 111th and 112th U.S. Congresses, our results indicate that members with higher vote shares and who are less ideologically extreme are increasingly likely to introduce companion legislation. These are exactly the sorts of members who are likely to be in a position to sacrifice a variety of re-election-oriented goals for legislative success. We also observe that the presence of a Senate companion to a House bill is associated with that House bill having greater cosponsorship support in the House, a higher probability of being reported out by committee, and ultimately a higher probability of passage in the House. These changes are small, but noticeable and statistically significant.

Future research on companion bills would do well to continue asking why there are so few companion bills introduced. We suggested that this is because the gains in legislative success by introducing companion bills are small relative to the potential costs, but there may be other reasons for the paucity of companion bills. It may also be the case that there are simply few members in each chamber with partners in the opposing chamber with whom they wish to work. Alternatively, legislators may simply lack the available time to reach across chamber lines. During times of split legislative control, there may be negative consequences in one’s own chamber for working with the opposing chamber. However, the value of working across the chamber in times of divided chamber control may see an increase in the effects of companion bill introduction on bill passage as a bridge across the divide. Future work should consider these partisan combination to address when and why legislators turn to these partnerships. Most legislators may view conference committees as the only real opportunity for cross-chamber work, and thus, may consider companion bill introductions unhelpful. Legislators themselves clearly value companion bill introductions and commonly tout their cross-chamber activities, but ultimately, many may actually regard the opposing chamber as the enemy our introductory anecdote suggested.

Our research also leaves open questions regarding the partner selection process in companion bill introduction. We observed that companion bill dyads were likely to form within parties between members of Congress with similar constituencies and within state delegations, but there are undoubtedly more nuances to how legislators choose their cross-chamber partners. Time spent together on joint committees may help facilitate such partnerships. Party leaders or national party committees may facilitate these sorts of cross-chamber partnerships. Interest groups may also play an important role in facilitating cross-chamber collaborations, as prior research has supported the notion of organized interests as “bridges” between legislators during agenda formation (Grossmann & Dominguez, 2009). In addition, we observe quite a bit of temporal heterogeneity in our partner selection results. We observe much clearer patterns in the 112th session of Congress than in the 111th. This may be a function of the divided nature of the 112th session of Congress with Republicans in control of the House and Democrats in control of the Senate, or the other features that distinguish these two sessions. A dynamic analysis of companion bill introductions could shed light on precisely why we see these differences.

Determining which legislator of the companion bill pair initiates the partnership would aid researchers seeking to answer questions about the rationale behind the formation of this interaction. Unfortunately, our data do not allow us to explore which legislator approached the other legislator to form the partnership. Leveraging other sources of data beyond those used in this article may help legislators solve this limitation of the work. For example, perhaps legislators state in floor speeches or press releases how the partnership was formed and in which direction the relationship flows.

Despite their intentionally designed intercameral opposition, House members and senators collaborate on legislation and choose their partners for these collaborations in systematic and predictable ways. This micro-level collaboration helps shed light on when and why the two normally hostile chambers might coordinate their efforts on legislation, and thus, helps illuminate how these chambers occasionally manage to overcome increasingly common congressional gridlock. Cross-chamber legislative entrepreneurs are driven at least in part by their constituencies and their desire for legislative success. These seem to be key elements in understanding how the individual incentives of members of Congress can both create and potentially alleviate stalemate between the chambers of Congress.