Issue Attention and Legislative Proposals in the U.S. Senate

Issue Attention 29 JONATHAN WOON University of Pittsburgh Issue Attention and Legislative Proposals in the U.S. Senate This analysis of bill sponsorship across a variety of issues and Congresses shows that committee membership is the single most important factor shaping a senator s level of issue attention. Constituency demand is of secondary importance. Ideology, partisanship, and national conditions play little or no role. Consistent with a theoretical cost-benefit framework, the results suggest that senators are motivated by the prospect of electoral and policy rewards from successful legislation rather than from mere position taking. The findings attest to the enduring importance of the committee system in a highly individualistic and increasingly partisan Senate. Legislative institutions play a central role in establishing the relative importance of numerous public concerns. Every two years, members of Congress introduce several thousand bills and collectively whittle these down to a few hundred public laws, elevating some issues while lowering others. The earliest stages of this process require legislators to shape policy alternatives through effort-intensive activities: gathering information, drafting bills, building coalitions, and keeping pace with the actions of various interests. Bauer, de Sola Pool, and Dexter emphasize the importance of the early stages of legislating compared to the final roll-call stage: The decisions most constantly on [a member s] mind are not how to vote, but what to do with his time, how to allocate his resources, and where to put his energy. There are far more issues before Congress than he can possibly cope with. (1972, 405) Thus, a fundamental aspect of lawmaking and representation is issue attention determining which issues to pursue and how much to pursue them. But how do legislators decide which issues are worth their valuable time and effort? How much is issue attention influenced by policy preferences, constituency and electoral concerns, or organizational factors, such as committees and parties? Does any one factor dominate, or do pressures vary from issue to issue? LEGISLATIVE STUDIES QUARTERLY, XXXIV, 1, February 2009 29

30 Jonathan Woon To answer these questions, I analyzed bill sponsorship in the Senate. My data and research design have important advantages over those used in previous research, providing new insights about the role that institutions play in shaping legislative behavior. First, my data have a panel structure that helps disentangle alternative influences on issue attention, distinguishing, for example, constituency interests from institutional positions and partisanship from majority status. Previous analysis of Senate behavior has tended to look only at overall levels of specialization and has concluded that committees are important (Matthews 1960) or that the importance of committees and specialization has diminished over time (Sinclair 1989), without accounting for alternative explanations of committee effects, such as self-selection. Second, because the bills in my data are classified by issue, I can account for senators issue-specific investments of time and effort. Previous research on legislative entrepreneurship failed to take into account the diversity of issues (Schiller 1995; Wawro 2000). Aggregation may mask the possibility that different factors influence attention to different issues. For example, attention to agriculture might be highly influenced by constituency interests, attention to tax policy might be driven more by ideological concerns, and attention to low-salience issues like science policy might be induced by committee membership. More important, by studying bill sponsorship in an issue-specific way and making comparisons across issues, I can offer insights into why legislators engage in the earliest stages of the legislative process. Third, although Matthews s (1960) classic work on legislative activism pertained to the Senate, more recent and sophisticated analyses have focused primarily on the House (Frantzich 1979; Hall 1987, 1996; Wawro 2000). Thus, theoretical developments satisfactorily explain House but not Senate behavior. In the House, institutional rules, procedures, and organizational rewards can be used to encourage representatives legislative efforts. For instance, regardless of whether one believes that restrictive rules enforce gains from trade (Weingast and Marshall 1988) or induce committees to acquire information (Gilligan and Krehbiel 1987; Krehbiel 1991), such rules protect the work of committees, and this power, in turn, motivates committee members to participate in legislative work. Wawro (2000) has also shown that representatives entrepreneurial efforts are rewarded neither by their constituents nor by interest groups but instead by the institution, which promotes legislators to prestigious and powerful committee and party positions. These incentives are weak, at best, in the Senate, which does not use restrictive rules to protect committee work, lacks a germaneness rule for non-appropriations legislation, and offers most

Issue Attention 31 majority party senators some sort of leadership position, say, as a subcommittee chair. 1 Investigating bill sponsorship in the Senate therefore provides an important contrast to previous work on the House. For nine different issues during the 101st through 106th Senates (1989 2000), my central finding is that a senator s committee position is the single most important factor shaping the senator s issue-specific attention. Across issues, committee members pay more attention to issues than do non-committee members. Somewhat less consistently across issues, committee leaders tend to pay more attention than rankand-file members. Constituency interest plays a secondary role. The importance of the committee system is quite surprising in light of the contemporary Senate s reputation as a highly decentralized institution in which individualism runs rampant and policy specialization is no longer the norm (Sinclair 1989). The results suggest that the advantages of committee positions in terms of agenda influence and specialization are significant enough that they shape senators attention to issues and outweigh the potential benefits of circumventing the regular committee process on the floor. Theoretical Framework Legislators possess finite time, energy, and resources with which to carry out their jobs and pursue their goals. A very simple, but useful, point of departure for understanding issue attention is to assume that legislators are rational, goal-oriented actors and to think of the legislator s decision problem as one of constrained optimization. That is, a legislator chooses to pay attention to the set of issues that yield the highest net benefits subject to institutional and strategic constraints. 2 This conceptualization is consistent with the rational-choice, costbenefit frameworks employed by Hall (1996), Schiller (1995), Wawro (2000), and others. To understand issue attention and bill sponsorship, we can partition the benefits of introducing a bill into two broad categories defined by whether or not obtaining the benefits depends on the bill s passage. According to Mayhew, the electoral payment is for positions rather than for effects (1974, 132), so there may be position-taking benefits that do not depend on a bill s passage. Simply by introducing bills on issues that concern their constituents, demonstrating that they share the same issue priorities, legislators may enhance their reelection prospects. When legislation successfully passes all the requisite procedural hurdles and obtains sufficient approval to become public law, it changes public policy. Legislators might also have personal

32 Jonathan Woon preferences about good public policy (Fenno 1973) and therefore benefit when policies are brought closer to what they believe to be ideal. This is a standard assumption in prevailing theoretical explanations of lawmaking (see, for instance, Krehbiel 1998). If constituents and campaign contributors also care about policy effects (rather than merely positions), then successful legislation has electoral benefits beyond mere position taking. 3 The attainment of benefits from policy outcomes is subject to two types of interrelated constraints that affect a bill s passage: institutional (rules and procedures) and strategic (the actions of other legislators). A legislator deciding how to allocate issue attention at the bill sponsorship stage of the legislative process therefore faces uncertainty about final policy outcomes and, ex ante, will take into account a bill s expected policy benefits. Because the outcome is uncertain, the value of policy benefits in the cost-benefit calculus is less than the value of benefits that would actually be received upon the bill s passage. An important component of my theoretical argument is that legislators expectations about the likelihood of a bill s eventual passage depend critically on their agenda influence, their access to and influence over the set of bills that will be considered in committee and on the floor. 4 Although party and committee hierarchies are much weaker in the Senate than in the House, they nevertheless play critical roles in shaping agenda influence. For example, theories of strong parties in Congress emphasize that the majority party controls the legislative agenda through its scheduling power (Cox and McCubbins 1993, 2005). Thus, legislative issue attention should yield greater expected benefits for members of the majority party than for members of the minority party. 5 Committee positions also increase the probability of a bill s consideration and passage, in at least two ways. First, a committee may recognize that its own members possess greater issue-specific expertise than non-members and may therefore be more likely to consider legislation sponsored or cosponsored by committee members (Wilson and Young 1997). Second, a committee member possesses a vote that may affect whether or not a bill is sent to the floor once it is considered by the committee. Although the probability that the vote will be pivotal may be small, that voting clout is nonetheless greater than the power wielded by a non-committee member. Committee leadership especially holding the full-committee chair, but also maintaining subcommittee chairs and ranking minority positions raises the probability of bill consideration further, since committee leaders possess procedural prerogatives that allow them to establish the committee s legislative priorities (Evans 1991; Oleszek 2001).

Issue Attention 33 To complete the theoretical framework, we must weigh positiontaking benefits and expected policy benefits against the cost of bill sponsorship. These costs will depend on whether the primary motivations for bill sponsorship are driven by policy outcomes or position taking. If bills are sponsored primarily for their position-taking benefits, then the costs will be minimal (although nonzero) and should not vary much from member to member or issue to issue; bills will be introduced with the expectation that they will fail and the exact nature of their policy consequences will not matter. On the other hand, if policy outcomes are the primary motivations for bill sponsorship, then the costs will be much more significant; legislators will likely expend sufficient effort to ensure that they introduce quality bills that, if enacted into law, lead to desirable policies. In other words, although drafting bills involves some costs regardless of legislator motivation, policy seeking implies that cost variations will be significant enough to affect sponsorship behavior. Institutional positions should substantially lower the marginal costs of such efforts. Committee members may be low-cost specialists because of their backgrounds or because they have developed expertise from committee experience (Gilligan and Krehbiel 1987; Krehbiel 1991). According to Hall, committee members also have transactional advantages (1996, 91); they gain political information about each other as a byproduct of frequent interaction. In addition, committee leaders control additional resources with which to subsidize bill sponsorship activities. Finally, theories of party in government suggest that strong majority parties have additional power and resources that can be used to subsidize attention to issues that are important to the party (Aldrich 1995; Aldrich and Rohde 1997 98). Hypotheses I grouped specific hypotheses about the factors that affect issue attention into three sets according to how the factors would affect senators cost-benefit decisions (that is, according to position-taking benefits, policy benefits, agenda influence, or costs related to institutional positions). Constituency Demand should, in general, influence issue attention through both position-taking benefits and the electoral benefits from policy outcomes. More specifically, constituency demand, either in the form of an issue s Economic Importance (Adler and Lapinski 1997) or Problem Severity (Kingdon 1995), should be positively related to issue attention.

34 Jonathan Woon Petrocik s (1996) theory of Party Issue Ownership suggests that the relevant constituency may be a partisan one. If each party s core constituency cares about a different set of issues, then legislators will receive greater benefits if they pay more attention to their party s issues. For example, employment and health care are often considered to be owned by Democrats, while crime and trade are owned by Republicans. An alternative hypothesis is that senators follow an inoculation strategy. Recognizing the issues on which their party is weak, legislators seek to remedy such deficiencies by introducing bills and being effective in those areas. For example, members of both parties may devote attention to health care but for different reasons the Democrats because their core constituency cares about this issue (consider President Clinton s health care reform), the Republicans because they seek to neutralize any electoral advantage that issue ownership conveys (consider the GOP s Medicare prescription drug plan). Such inoculation tactics are consistent with evidence regarding issue uptake (Sulkin 2005) and the use of defense bills by women senators to counteract gender stereotypes and demonstrate their credibility on military affairs (Swers 2007). Policy Preferences influence issue attention primarily through their effects on the benefits obtained from new policy outcomes. These preferences are ideological, based on legislators beliefs about what constitutes good public policy. I tested a liberalism hypothesis: because liberal senators prefer government solutions, they should, in general, sponsor more bills than do conservative senators (Schiller 1995). Alternatively, the logic of spatial models implies an extremism hypothesis: since extremists generally have more to gain from policy changes than do moderates, extremists will pay more attention to issues and sponsor more bills. Changes in policy are also more valuable when existing National Conditions are worse and aggregate demand for policy change is greater. This hypothesis dovetails with both the empirical macropolitical agenda-setting literature (Baumgartner and Jones 1993; Kingdon 1995; Walker 1977) and the spatial theories of gridlock (Krehbiel 1998). Institutional Positions either raise ex ante expectations of legislative success (and therefore of expected benefits) or lower the costs of bill sponsorship. First, in terms of committee hierarchies, I expected Committee Members to pay more attention to issues under their committee jurisdictions than non-committee members would, Committee Leaders (both chairs and ranking members) to sponsor more bills than rank-and-file committee members would, and full-committee leaders to sponsor more bills than subcommittee leaders.

Issue Attention 35 Although I would argue that committee positions matter because of strategic and resource advantages, the literature strongly suggests another possibility. If committees are composed primarily of highdemand preference outliers (Adler and Lapinski 1997; Weingast and Marshall 1988), then committee membership might matter only because it is the best proxy for constituency preference. The cost-benefit and self-selection arguments would seem to be observationally equivalent, and the appropriate interpretation of any committee effects would require additional analysis. I will address this inferential problem later in the article, in part by exploiting the panel structure of the data to account for unobserved constituency interest. To continue with institutional positions, I would expect several effects if the Majority Party wields significant agenda influence. If the majority party s agenda influence is greatest in terms of the floor agenda (rather than at the committee level), then majority party status should be positively related to bill sponsorship across all issues and all forms of committee membership. There should also be a positive and significant interaction between majority party status and committee positions. Furthermore, if both the issue-ownership and majority-status hypotheses hold, then there should be an interaction such that Democratic senators pay more attention to employment and health care when in the majority, whereas Republican senators pay more attention to crime and trade while in the majority. A key assumption underlying the policy-preference and institutionalposition hypotheses is that the benefits from policy outcomes (possibly including electoral rewards) are a significant factor motivating behavior. Because the probability of any one bill s passage is typically very low, however, it is quite reasonable to doubt the claim that expected policy benefits are significant enough to affect a legislator s issue-attention decisions. This skepticism suggests an alternative pure position-taking hypothesis: if the benefits of position taking outweigh expected policy benefits, then only the constituency-demand factors should affect issue attention and the policy preferences and institutional positions should have no effect on issue attention. Similarly, although the theoretical framework describes how various factors may affect issue attention and implies a number of testable hypotheses, I do not have strong a priori expectations that every factor will matter for every issue. For this study, I assumed that legislators pursue multiple goals (Fenno 1973), but the exact influence of each factor on legislators issue-attention decisions is ultimately an empirical question.

36 Jonathan Woon Data and Methods The data consist of public bills introduced by senators during the 101st through 106th Congresses (1989 2000) and were obtained from the Library of Congress s THOMAS online database. Only senators who served nearly complete terms are included in the dataset, with 594 senator-congress cases resulting. 6 My analysis covers periods of both Democratic and Republican control of the Senate. An important benefit of the panel structure is that it potentially allows one to distinguish between party and majority-status effects in ways that previous studies have not been able to do. 7 Bill Sponsorship The dependent variable is the number of bills introduced for a particular issue, so the unit of analysis is a senator-congress-issue triple. To classify bills by issue, I used the top term assigned by the Congressional Research Service, which describes the primary topic of a bill. The empirical analysis focuses on nine issues. Three are matters of broad public concern: crime, employment, and health care. The other six issues, although they episodically capture the attention of the broader public, are more typically of special concern to narrower economic interests: agriculture, banking, communications, energy, trade, and transportation. This set of issues represents a mix of broad and narrow concerns. Table 1 presents summary statistics for bill sponsorship. For each issue, the average senator introduces very few bills, and many senators sponsor no bills at all. Nevertheless, a nontrivial fraction of senators does pay attention to each issue and, occasionally, a senator introduces up to ten times more bills than the average senator. The disaggregation of bill counts by issue is an important advantage of my data, for at least two reasons. First, disaggregration by issue allows me to determine whether there are issue-specific motivations for bill sponsorship or if certain factors affect behavior across the board. Previous studies have only looked at overall levels of activity (Anderson, Box-Steffensmeier, and Sinclair-Chapman 2003; Koger 2003; Schiller 1995). Second, if the average complexity of bills differs across issues (for instance, if the average trade bill is relatively easier to write and introduce than the average health care bill), then it would be inappropriate to combine or compare bill counts for different issues. In contrast, examining issue-specific bill counts does not require any assumptions about cross-issue comparability.

Issue Attention 37 TABLE 1 Summary Statistics for Bill Sponsorship Percent of Cases Issue Mean Std. Dev. Max with No Bills N Agriculture 0.78 1.59 12 66 594 Banking 0.62 1.53 16 71 594 Communications 0.46 1.23 12 74 594 Crime 1.36 2.36 17 51 594 Employment 0.79 1.59 17 64 594 Energy 0.84 1.68 18 59 594 Health 1.75 2.93 23 43 594 Trade 2.42 4.84 48 45 594 Transportation 1.04 1.70 13 54 594 Total 1.12 2.48 48 59 5,346 Independent Variables Constituency demand may have one of three possible sources: economic importance, problem severity, and party issue ownership. I measured Economic Importance using several variables calculated from data on gross state product (GSP). First, I calculated issue-specific measures by computing the proportion of GSP for industries relevant to each issue. For example, I used the percentage of GSP from farming for agriculture and the percentage of GSP from natural resources and utilities for energy. Second, I calculated measures that capture general characteristics of a state s economy. I used Economy Size (in terms of the log of GSP) as a measure of overall demand. I also measured the Economy Concentration by calculating a Herfindahl Index from the GSP data. The index ranges from 0 to 1, with higher values indicating greater concentration in fewer industries; this measure is inversely related to demand. I used state-level policy indicators to measure Problem Severity. The relevant measures are issue specific, and problem indicators are available for the broad issues but not the narrow economic ones. The data come from various annual government publications, and I computed the average for each two-year Congress. For State Crime Rate, I used the crime rate (specifically, the incidence of serious violent and property crimes) from the Federal Bureau of Investigation s Crime

38 Jonathan Woon in the United States. For Unemployment Rate, I referred to unemployment rates calculated by the Bureau of Labor Statistics. For health policy, I used data from the Current Population Survey on the percentage of the state population Uninsured (not covered by either private or public health plans, which should be positively related to attention) and the percentage covered by Government Health Care programs (which should decrease issue attention, because coverage implies lower demand for new programs). 8 To test the role of Party Issue Ownership, I used a simple party dummy variable. I hypothesized that Democratic legislators would pay more attention to employment, because of the importance of labor in their core constituency, and to health issues, because of their association with Medicare. Republican legislators should pay more attention to crime, having traditionally been seen as more successful on law and order, and to trade, since they have been associated with both free trade and greater foreign policy expertise. Following Petrocik (1996), I support these issue-ownership characterizations with survey data. 9 To measure policy preferences, I created two dummy variables: Extreme Liberal is coded as 1 if a senator is in the most liberal quartile of senators within a Congress, as measured by Poole and Rosenthal s DW-NOMINATE scores; Extreme Conservative is coded as 1 if a senator is in the most conservative quartile. 10 This coding allows me to distinguish between the extremism and liberalism hypotheses, because it allows for the possibility (but does not assume) that the relationship between ideology and bill sponsorship is linear or nonlinear. For national conditions, I used national-level versions of the state-level problem indicators for crime, employment, and health care. To test the institutional-position hypotheses, I included a set of dummy variables concerning committee positions and majority party status. To test the committee-hierarchy hypotheses, I included dummy variables for committee membership, subcommittee leadership, and full-committee leadership. These variables are not mutually exclusive; the full-committee chair is coded as both a member of the committee and a full-committee leader. The leadership variables are coded as 1 for both chairs and ranking members. To code these variables, I used data from the Congressional Directory, and I determined the relevance of a committee for an issue according to the pattern of bill referrals. 11 I included separate Democratic Majority and Republican Majority variables to test the majority party hypothesis while allowing for a possible interaction with issue ownership. For example, if there is an interaction for health policy, then Democratic Majority should be positive and significant and Republican Majority should have no effect. If there is

Issue Attention 39 a majority-status effect without the issue-ownership interaction, then both Democratic and Republican Majority should be positive and significant. Similarly, I interacted the separate majority-status variables with each of the committee-position variables to allow for a three-way interaction between majority status, committee position, and issue ownership. Statistical Model Since the dependent variable is the number of bills introduced by a senator in an issue area, a count model is more appropriate than a linear regression model. Count models take into account the discrete and non-negative characteristics of the dependent variable in order to obtain unbiased estimates of the effects of each independent variable. There are many zero observations in the data, reflecting senators who did not introduce any bills on an issue, so there may be greater dispersion than in a Poisson count model. I used a negative binomial regression model because it allows for the possibility of overdispersion through an additional parameter in the model. Results I estimated the negative binomial regression model separately for each issue by maximum likelihood. Table 2 summarizes the hypotheses that are supported for each issue. Tables 3, 4, and 5 present the full results in terms of baseline expected values and first differences for all of the independent variables, which are more substantively meaningful than the coefficient estimates. 12 I calculated the baseline by holding all continuous variables at their means and the dummy variables at 0. Substantively, the baseline is an ideologically moderate Republican senator who is in the minority, is not a member of a relevant committee, and hails from an average state. Each first difference describes a marginal effect: how a change in one variable affects the number of expected bill introductions relative to the baseline. For continuous variables, the change is an increase of one standard deviation, and for dummy variables, the change is from 0 to 1. Note that the estimated dispersion parameter for every issue is significantly greater than 0, which suggests that the negative binomial model is, in fact, more appropriate than a Poisson model. 13 The central finding of my analysis is that committee membership plays a consistently important role in increasing attention. For all but one of the nine issues (health policy), the effect of committee membership is positive and statistically significant. The effects also

40 Jonathan Woon TABLE 2 Summary of Supported Hypotheses Variables Crime Employment Health Agriculture Banking Commerce Energy Trade Transportation Constituency Demand Economic Importance GSP GSP Industry Industry Industry GSP HHI GSP GSP HHI Problem Severity n/a n/a n/a n/a n/a n/a Party Issue Ownership n/a n/a n/a n/a n/a Policy Preferences Ideology Liberal National Conditions % Government n/a n/a n/a n/a n/a n/a Insured Institutional Positions Committee Membership Y Y Y Y Y Y Y Y Committee Leadership Full Full Full Full Majority Status Majority Status Interactions Democratic Democratic Subcommittee Republican Subcommittee Subcommittee Chair Chair Chair Chair Notes: A dash indicates the hypothesis is not supported. If a hypothesis was tested using more than one variable, then only the significant variables are listed. If there were alternative versions of a hypothesis (e.g., for ideology, leadership, or majority status), then only the supported version of the hypothesis is listed.

Issue Attention 41 TABLE 3 Negative Binomial Regression Results for Bill Sponsorship on Crime, Employment, and Health Policy Crime Employment Health Policy Variables Estimate 95% C.I. Estimate 95% C.I. Estimate 95% C.I. Baseline Expected Bills 0.96 [ 0.68, 1.32] 0.39 [ 0.26, 0.57] 1.47 [ 1.04, 2.02] First Differences Constituency Demand State Crime Rate 0.10 [ 0.22, 0.02] State Unemployment Rate 0.01 [ 0.05, 0.08] State Uninsured 0.05 [ 0.15, 0.28] State Government Health Care 0.05 [ 0.13, 0.25] Economy, % Health Care 0.37** [ 0.12, 0.69] Economy Size 0.57** [ 0.34, 0.87] 0.10** [ 0.03, 0.19] 0.11 [ 0.07, 0.32] Economy Concentration 0.01 [ 0.11, 0.13] 0.11** [ 0.18, 0.06] 0.07 [ 0.26, 0.11] Democrat 0.20 [ 0.66, 0.25] 0.06 [ 0.28, 0.16] 0.14 [ 0.83, 0.53] Policy Preferences Extreme Liberal 0.18 [ 0.15, 0.62] 0.44** [ 0.14, 0.92] 0.87** [ 0.16, 1.89] Extreme Conservative 0.00 [ 0.31, 0.30] 0.05 [ 0.19, 0.09] 0.88** [ 1.35, 0.51] U.S. Crime Rate 0.13 [ 0.33, 0.05] U.S. Unemployment Rate 0.04 [ 0.05, 0.14] U.S. Uninsured 0.16 [ 0.09, 0.51] U.S. Government Health Care 0.31** [ 0.56, 0.11] Institutional Positions Democratic Majority 0.14 [ 0.47, 0.33] 0.00 [ 0.16, 0.25] 0.85** [ 1.27, 0.48] Republican Majority 0.28 [ 0.69, 0.09] 0.08 [ 0.15, 0.31] 0.42 [ 0.32, 1.18] Committee Member 1.72** [ 0.44, 3.77] 1.01** [ 0.47, 1.77] 0.49 [ 0.14, 1.30] x Democratic Majority 1.09 [ 0.20, 3.64] 0.10 [ 0.28, 0.18] 1.32 [ 0.06, 3.76] x Republican Majority 0.29 [ 0.49, 1.77] 0.01 [ 0.21, 0.42] 0.23 [ 0.54, 1.50] Sub Com Leader 0.05 [ 0.51, 0.96] 0.21 [ 0.15, 0.89] 0.46 [ 1.02, 0.25] x Democratic Majority 0.55 [ 0.99, 0.14] 1.26* [ 0.02, 4.35] 3.12* [ 0.22, 9.00] x Republican Majority 0.48 [ 0.53, 2.64] 0.45 [ 0.29, 2.74] 0.50 [ 0.84, 3.28] Full Com Leader 4.78** [1.09, 12.13] 0.55 [ 0.02, 1.70] 3.00** [ 0.65, 7.07] x Democratic Majority 0.41 [ 0.75, 3.61] 0.67 [ 0.19, 3.00] 1.55 [ 0.63, 6.52] x Republican Majority 0.61 [ 0.72, 4.18] 0.22 [ 0.31, 1.66] 0.36 [ 1.04, 3.58] Dispersion Parameter 0.69** (SE =.11) 0.59** (SE =.15) 0.85** (SE =.10) Log Likelihood 824.51 620.70 953.83 N 594 594 594 *p <.05; **p <.01 (for coefficient estimates); expected values, first differences, and confidence intervals computed using Clarify (King, Tomz, and Wittenberg 2000).

42 Jonathan Woon TABLE 4 Negative Binomial Regression Results for Bill Sponsorship on Agriculture, Banking, and Communications Agriculture Banking Communications Variables Estimate 95% C.I. Estimate 95% C.I. Estimate 95% C.I. Baseline expected bills 0.37 [ 0.25, 0.54] 0.41 [ 0.26, 0.61] 0.19 [ 0.11, 0.31] First Differences Constituency Demand Economy, % Farming 0.27** [ 0.15, 0.45] Economy, % Financial Services 0.01 [ 0.07, 0.07] Economy, % Communications 0.02 [ 0.01, 0.06] Economy Size 0.09* [ 0.02, 0.19] 0.03 [ 0.04, 0.11] 0.03* [ 0.07, 0.00] Economy Concentration 0.06 [ 0.01, 0.15] 0.01 [ 0.06, 0.08] 0.00 [ 0.04, 0.03] Democrat 0.19 [ 0.05, 0.44] 0.24** [ 0.44, 0.07] 0.06 [ 0.07, 0.20] Policy Preferences Extreme Liberal 0.06 [ 0.08, 0.27] 0.16 [ 0.04, 0.48] 0.06 [ 0.04, 0.21] Extreme Conservative 0.05 [ 0.12, 0.21] 0.20** [ 0.38, 0.06] 0.01 [ 0.11, 0.08] Institutional Positions Democratic Majority 0.13* [ 0.26, 0.01] 0.58** [ 0.16, 1.23] 0.02 [ 0.07, 0.15] Republican Majority 0.07 [ 0.24, 0.10] 0.06 [ 0.28, 0.14] 0.07 [ 0.06, 0.20] Committee Member 0.46* [ 0.06, 1.13] 1.37** [ 0.50, 2.72] 0.51** [ 0.23, 0.94] x Democratic Majority 0.74* [ 0.02, 2.30] 0.03 [ 0.27, 0.48] 0.07 [ 0.17, 0.08] x Republican Majority 0.34 [ 0.12, 1.31] 0.04 [ 0.27, 0.77] 0.06 [ 0.16, 0.11] Sub Com Leader 0.00 [ 0.23, 0.38] 0.13 [ 0.35, 0.23] 0.04 [ 0.19, 0.33] x Democratic Majority 0.03 [ 0.31, 0.55] 1.71* [ 0.06, 5.99] 4.44** [0.18, 20.30] x Republican Majority 0.35 [ 0.21, 1.74] 1.52 [ 0.02, 5.65] 4.48* [0.13, 21.00] Full Com Leader 0.21 [ 0.22, 1.15] 0.43 [ 0.18, 1.88] 0.50 [ 0.02, 1.82] x Democratic Majority 2.12 [ 0.06, 9.23] 1.69 [ 0.14, 7.67] 0.57 [ 0.11, 3.07] x Republican Majority 1.99 [ 0.07, 8.43] 1.58 [ 0.22, 7.70] 1.37 [ 0.02, 6.06] Dispersion Parameter 0.82** (SE =.17) 0.91** (SE =.21) 0.82** (SE =.24) Log Likelihood 604.28 524.00 459.97 N 594 594 594 *p <.05; **p <.01 (for coefficient estimates); expected values, first differences, and confidence intervals computed using Clarify (King, Tomz, and Wittenberg 2000).

Issue Attention 43 tend to be substantively large: the increase in the expected number of bills is always at least as large as the baseline itself, and for four issues, the increase is more than twice that amount. For example, membership on the Judiciary Committee leads to an increase of 1.72 more crime bills than the baseline of 0.96, membership on the Labor and Human Resources Committee increases employment-related bills by 1.01 over the baseline of 0.39, and membership on the Finance Committee leads to 5.16 more trade-related bills than the baseline of 1.37. In addition, these effects are larger than the effects of any of the statistically significant constituency variables yet to be discussed. The data support several of the other hypotheses, but less consistently across issues. The second most consistently important factor affecting issue attention is constituency demand. Specifically, the Economic Importance hypothesis is supported for six issues but takes on a slightly different form for each one. Industry variables are statistically significant and positive for three of these (health policy, agriculture, and trade), state Economy Size is significant and positive for five issues (crime, employment, agriculture, trade, and transportation), and Economic Concentration is significant and in the expected direction for two (employment and transportation). In substantive terms, however, the effect of Economic Importance is a distant second when compared to Committee Membership. For example, a one standard deviation increase in the size of a state s economy only leads to an increase of 0.57 crime bills, 0.10 employment bills, 0.77 trade bills, and 0.14 transportation bills. In terms of other institutional positions, the effects of fullcommittee leadership positions are positive and statistically significant for four issues (crime, health policy, energy, and transportation). Although these leadership effects are less consistent across issues, their sizes tend to be as large as, or slightly larger than, the committee effects. A full-committee leadership position increases expected bill sponsorship by 4.78 bills for crime, 3.00 bills for health policy, 1.69 bills for energy, and 1.30 bills for transportation. None of the basic subcommittee leader dummy variable effects are statistically significant, but a few of the subcommittee and majority party interaction variables do have positive and statistically significant effects. There is, however, no consistent pattern across issues to support specific hypotheses. 14 Taken together, the results for the various leadership variables suggest that leadership positions increase issue attention, but in more idiosyncratic ways than committee membership. A reasonable interpretation of this finding is that even though leadership positions provide additional agenda influence and resources, senators may not always

44 Jonathan Woon TABLE 5 Negative Binomial Regression Results for Bill Sponsorship on Energy, Trade, and Transportation Energy Trade Transportation Variables Estimate 95% C.I. Estimate 95% C.I. Estimate 95% C.I. Baseline expected bills 0.54 [ 0.38, 0.73] 1.37 [ 0.92, 1.95] 0.83 [ 0.62, 1.10] First Differences Constituency Demand Economy, % Resources 0.01 [ 0.06, 0.09] Economy, % Utilities 0.03 [ 0.10, 0.04] Economy, % Farming 0.29** [ 0.52, 0.08] Economy, % Manufacturing 0.35** [ 0.11, 0.66] Economy, % Transportation 0.03 [ 0.12, 0.08] Economy Size 0.02 [ 0.09, 0.05] 0.77** [ 0.40, 1.28] 0.14* [ 0.03, 0.28] Economy Concentration 0.05 [ 0.11, 0.02] 0.06 [ 0.29, 0.17] 0.10* [ 0.19, 0.00] Democrat 0.13 [ 0.35, 0.07] 0.21 [ 0.86, 0.40] 0.26 [ 0.55, 0.02] Policy Preferences Extreme Liberal 0.18 [ 0.02, 0.48] 0.36 [ 0.77, 0.07] 0.09 [ 0.18, 0.45] Extreme Conservative 0.18* [ 0.36, 0.03] 0.75* [ 0.11, 1.46] 0.27* [ 0.52, 0.05] Institutional Positions Democratic Majority 0.16 [ 0.07, 0.46] 0.90* [ 0.13, 2.01] 0.13 [ 0.17, 0.51] Republican Majority 0.06 [ 0.28, 0.15] 0.51* [ 1.04, 0.06] 0.04 [ 0.34, 0.25] Committee Member 0.87** [ 0.30, 1.70] 5.16** [ 2.75, 8.51] 1.20** [ 0.47, 2.20] x Democratic Majority 0.02 [ 0.31, 0.63] 0.03 [ 0.79, 1.44] 0.23 [ 0.33, 1.23] x Republican Majority 0.11 [ 0.25, 0.78] 0.42 [ 1.04, 0.57] 0.05 [ 0.46, 0.65] Sub Com Leader 0.18 [ 0.17, 0.78] 1.79 [ 0.71, 8.46] 0.61 [ 0.12, 1.85] x Democratic Majority 0.76 [ 0.09, 2.55] 1.61 [ 1.28, 12.85] 1.11 [ 0.22, 3.92] x Republican Majority 0.70 [ 0.13, 2.53] 0.15 [ 1.54, 4.67] 0.21 [ 0.57, 1.92] Full Com Leader 1.69** [ 0.35, 4.33] 0.65 [ 1.43, 0.90] 1.30* [ 0.02, 3.74] x Democratic Majority 1.25 [ 0.12, 4.68] 1.37 [ 1.30, 11.36] 0.79 [ 0.51, 4.21] x Republican Majority 0.46 [ 0.34, 2.47] 70.29** [1.65, 373.64] 2.38 [ 0.16, 9.06] Dispersion Parameter 0.40** (SE =.12) 1.66** (SE =.15) 0.56** (SE =.12) Log Likelihood 646.78 1,073.54 755.60 N 594 594 594 *p <.05; **p <.01 (for coefficient estimates); expected values, first differences, and confidence intervals computed using Clarify (King, Tomz, and Wittenberg 2000).

Issue Attention 45 utilize these advantages if the increase in net expected benefits is not sufficiently large to motivate a change in their behavior. The results show that disaggregating bills by issue and estimating separate models for each (in contrast to Schiller s 1995 and Wawro s 2000 aggregate analyses) is methodologically consequential. One of the more obvious consequences is that health policy very clearly stands apart from the other issues. It is the only issue for which committee membership itself does not matter (although full-committee leadership is important), and it is the only issue for which the policy-preference hypotheses receive any support. The liberalism hypothesis is supported: Extreme Liberal has a positive effect and Extreme Conservative has a negative effect, so that the total difference in bill sponsorship between an extremely liberal senator and an extremely conservative one is 1.75 bills, which is more than the baseline of 1.47. 15 Similarly, the national policy conditions hypothesis receives some support, since a one standard deviation increase in national Government Health Care leads to a decrease of 0.31 expected bills, but this change is substantively rather small. The subtler consequence of disaggregation is that, although the basic qualitative findings (that committee membership is the primary influence and that committee leadership and constituency demand are secondary influences) generalize across the other issues, the quantitative results indicate significant parameter heterogeneity. For example, baseline expected values and the effects of Committee Membership range from 0.37 and 0.46 for agriculture, respectively, to 1.37 and 5.16 for trade. As noted earlier, committee leadership positions do not affect bill sponsorship for every issue, and constituency demand does not matter for banking, communications, or energy. For those relatively narrow issues, committee positions are the only significant influences on bill sponsorship. A pooled analysis would lead to substantively different conclusions about the importance of factors such as ideology and majority status but such conclusions would be inappropriate, since a likelihood ratio test shows that a pooled model can be rejected. 16 The data provide very little or no support for the remaining hypotheses across all issues. The importance of policy preferences (ideology and national conditions) is limited to health policy, and none of the party-related hypotheses receive support from these issue analyses. The lack of a majority party effect is consistent with the view of the Senate as a much less partisan institution than the House, and the lack of support for the issue-ownership hypothesis for any of the issues suggests that the inoculation strategy may be at play (although we cannot make a conclusive determination from a null finding).

46 Jonathan Woon Overall, the results support the cost-benefit model of issue attention while providing a more refined picture of the specific factors that affect those costs and benefits. Not only do committees matter, a finding consistent with the literature on congressional behavior (Fenno 1973; Frantzich 1979; Hall 1996; Matthews 1960; Schiller 1995), but committee membership matters the most. In addition to committee membership, committee leadership also lowers the costs of bill sponsorship and raises expected benefits by increasing the probability of bill passage, even if committee leaders do not always exercise their influence. Constituency demand is the primary source of benefits for most issues, and the benefits of policy-based preferences are limited to a single issue (health policy). Although the results do not provide direct numerical estimates of the mix of benefits due to position taking versus policy outcomes, the importance of committee factors implies that we can rule out the pure position-taking hypothesis and strongly suggests that policy outcomes (in terms of the electoral benefits of successful legislation) are quite significant. Committee Effects: Institutional Advantages or Self-selection? To the extent that committee assignments are governed by selfselection and senators preferences for committee assignments are driven primarily by constituency interests, the presence of committee effects might not be due to greater agenda influence and lower costs of attention, as I claim, but might arise because committee membership is a better indicator of constituency interest than the independent variables used in this analysis. There are several reasons, however, for discounting this interpretation. Theoretically, even if preferences (that is, demand) for committee assignments are perfectly determined by constituency interests, the assignment process itself (supply) is constrained by committee size and party rules (for example, rules that preclude two senators from the same state of the same party from serving on the same committee). In other words, senators might want to be on committees for constituency reasons, but they can t always get what they want. The fact that senators regularly transfer to new committees attests to the strength of the supply constraint and to the reality that senators committee preferences are never entirely satisfied. Moreover, there are good reasons (including the theory of progressive ambition) to believe that senators may leave constituency committees for more powerful and policyrelevant committees, such as Judiciary and Finance. In fact, Stewart

Issue Attention 47 and Groseclose (1999) have shown that constituency-oriented committees are generally ranked last in terms of desirability, compared to prestige or policy committees. Several additional empirical analyses more convincingly rule out the possibility that committee assignments simply mask underlying or imperfectly measured constituency interest. First, I reestimated the models for agriculture and trade after omitting the industry-specific Economic Importance variables. These issues represent the best-case scenarios for the face validity of the industry variables as measures of constituency demand. By omitting them and observing how the committee effects changed, I assessed the extent to which omittedvariables bias affects the committee coefficients for the other issues. For agriculture, the effect of committee membership (first difference) increased from 0.46 additional bills to 0.66; for trade, the effect increased from 5.16 additional bills to 5.91. In both cases, the difference is minor, which suggests that any potential inflation of the coefficients due to omitted-variables bias is small for other issues. Second, I examined changes in bill sponsorship before and after committee transfers. This exercise takes advantage of the panel structure of the data and is essentially a natural experiment in which constituency demand is held constant and committee membership is the treatment variable. The self-selection interpretation implies that there should be no difference in bill sponsorship before and after committee transfers. Table 6 presents the effects of committee transfers in terms of estimated first differences from negative binomial regression models. There are relatively few transfers for several issues, so I pooled the data and controlled for issue-specific fixed effects. For transfers to a committee, we can reject the hypothesis that the new committee assignment has no effect, for issues pooled together as well as disaggregated into broad and narrow concerns. For transfers off of a committee, the transfer has no effect when all issues are pooled together and when only broad issues are considered. The lack of a transfer-off effect for broad issues is consistent with the notion that senators (or their personal staffs) acquire expertise, thus permanently lowering the cost of bill sponsorship on those issues. For narrow issues, however, there is a statistically significant decrease in bill sponsorship after a senator transfers off a committee. This decrease is consistent with the notion that the cost of acquiring technical information relative to the small payoffs of success on these issues leads members to rely more heavily on the expertise of committee staff. When senators leave these committees, they lose this resource advantage and their bill sponsorship declines accordingly. Overall, the transfer analysis favors the cost-benefit interpretation of committee effects.

48 Jonathan Woon TABLE 6 Effects of Committee Transfers on Bill Sponsorship Transfer to Transfer from Issue Range Estimate 95% C.I. Estimate 95% C.I. All issues 0.67** [0.25, 1.45] 0.34 [ 1.14, 0.28] Broad issues 2.52** [1.07, 4.99] 0.77 [ 0.27, 3.12] Narrow issues 0.32* [0.05, 0.81] 1.24* [ 2.97, 0.13] *p <.05; **p <.01 (for coefficient estimates). Notes: Each estimate is a first difference from a separate negative binomial regression model that includes a set of issue dummy variables. Estimated using Clarify (King, Tomz, and Wittenberg 2000). Third, I estimated alternative specifications of the models presented in Tables 3, 4, and 5 to include additional covariates intended to control for alternative explanations of the committee effects. 17 Since party rules or the supply of committee assignments sometimes preclude two same-state senators from serving on the same committee, the presence of at least one senator on a committee may be a good indicator of state-level constituency demand. I therefore included a dummy variable indicating if at least one senator from a state was a member of the relevant committee. In that case, I interpreted the committee effect as an additional effect above and beyond constituency interest. Alternatively, I could (and did) take advantage of the data s panel structure to control for unobserved constituency interest by including a set of state dummy variables (that is, state fixed effects). Finally, the relevant constituency may not be a geographic one but rather a set of high-demand interest groups. When data were available (from the Center for Responsive Politics), I controlled for campaign contributions from relevant industries (health care, agriculture, banking, communications, energy, and transportation). Although I found some minor differences in the point estimates for the committee membership coefficients across alternative specifications, the main results of my original analysis remained intact: committee membership is the dominant factor influencing senators issue-attention decisions. All of the additional analyses favor the cost-benefit framework over the selfselection interpretation.

Issue Attention 49 Conclusion Most observers would characterize the United States Senate as an institution in which individualism exerts a much stronger force on legislative behavior than do partisan or committee organizations, especially when compared with the House (Binder 1997; Evans and Lapinski 2005; Ripley 1969). Procedures such as unanimous-consent agreements, the filibuster, and the ability to offer nongermane floor amendments give individual senators tools with which to circumvent the normal committee process. According to Sinclair (1989), the level of individualism has increased along with the decline in specialization and the importance of committees since the earlier eras described by Matthews (1960) and Fenno (1973). My findings, in contrast, attest to the enduring importance of committee assignments and leadership positions in shaping senators legislative activities. Since the Senate does not provide the same kind of valuable selective incentives as the House, such as restrictive rules (Krehbiel 1991, 1997) or promotions to powerful institutional positions (Wawro 2000), it may seem puzzling that senators would engage in the highly detailed, dull, and politically unrewarding (Matthews 1960, 64) work of the Congress. Describing such legislative efforts as public goods, Hall (1996, 50) asks, Why would a member participate in their production? Why not free ride? Why not let Representative Schmuck do it? It would seem that Senate individualism would further exacerbate this problem. The production of collective benefits by an activity such as bill sponsorship does not mean, however, that the level of the activity will be zero (Olson 1965). As I have shown, there can be substantial private benefits from bill sponsorship for members of the Senate. Although the costs and benefits of legislative issue attention cannot be observed directly, my theoretical framework implies several testable hypotheses about their sources. The results suggest that many of the benefits of issue attention are related to constituency demand and, occasionally, policy preferences. Early-stage legislative efforts are thus not entirely public goods. My analysis suggests that there may be substantial personal rewards in terms of electoral benefits. Moreover, the rewards cannot be solely due to position taking, since, if that were the case, committee positions would be irrelevant. Thus, even though the probability of bill passage may be very low, the data suggest that the potential benefits of new policy outcomes for committee members are sufficiently large to weigh heavily in bill sponsorship decisions.