Legislative Entrepreneurship and Campaign Finance by
Gregory J. Wawro
y
July 21, 1997
Paper prepared for presentation at the 1997 Political Methodology Summer Conference. The author would like to thank Gary Jacobson, Garrison Nelson, Keith Poole, and David Rohde for providing data. These individuals bear no responsibility for any malfeasance that the author may have committed with their data. y Asst. Professor, Department of Political Science, Columbia University,
[email protected].
Legislative Entrepreneurship and Campaign Finance by
Gregory J. Wawro
Drawing on models of service{induced and investor PAC campaign contributions, I analyze the role that legislative entrepreneurship plays in PACs' contribution decisions. I explore the possibility that PACs use campaign contributions to invest in members of Congress with the expectation that members will reciprocate by engaging in entrepreneurial behavior to the bene t of PACs. To determine whether a relationship exists between legislative entrepreneurship and PAC contributions I compute measures of entrepreneurial behavior for individual members of the U.S. House using detailed data on bill sponsorship and congressional hearings from the 97th through the 101st Congress. In order to cleanly estimate the e ects of legislative entrepreneurship, we need to account for unobservable member{speci c factors that enter into the PAC contribution calculus. To account for such factors I employ panel data methods which require very few assumptions about the data and provide a way to test whether the manipulations of the data that are required for a panel analysis introduce bias.
Introduction An important part of the recent research on campaign nance has focused on the \service" relationship between members of Congress and interest groups (Baron 1989a; 1989b). This work focuses on how members can use the authority that their oce confers on them to deliver services to interest groups in exchange for campaign contributions. These services range from intervening with government agencies on a group's behalf to voting for legislation that the group desires to more general legislative activities that bene t the group. The work on service{induced campaign contributions is closely related with work on contributions as \investments" in members of Congress (Snyder 1990; 1992; Grier and Munger 1986; 1991). Political organizations invest in members by giving them campaign contributions with the expectation of receiving some future return. Snyder (1992) has found some evidence that political action committees (PACs) base their investment decisions on such factors as the age of the member when she enters Congress and the likelihood that she will advance to the Senate. Younger members and members who are more likely to become Senators are more attractive investments because in the long run they can provide bigger payo s to interest groups.1 A key feature of models of service{provision and investments is that PACs base their contribution decisions on their beliefs about members' abilities to deliver on promises of services. At the very least members must be in oce if they are to provide services to groups, so the probability of winning the election should be a key determinant of contributions. Once in oce incumbents will vary in their abilities to provide services. For example, a member's position in the institution will provide him with certain resources that he can use to a group's bene t. Some members will simply be better than others at browbeating recalcitrant bureaucracies or manipulating the legislative process to the bene t of groups. This paper extends the work on service and investment relationships by examining the role that legislative entrepreneurship plays in PACs' contribution decisions. Members engage in legislative entrepreneurship when they acquire policy expertise, draft legislation, build coalitions and push their proposals through the legislative process.2 I explore the possibility that PACs use campaign contributions to invest in members of Congress with the expectation that members will reciprocate by engaging in entrepreneurial behavior to the PACs bene t. Drawing on Hall and Wayman 1990, I contend that interest groups who want to in uence policy should be concerned with the degree to which members are involved in the legislative McCarty and Rothenberg 1996, however, present empirical evidence that calls into question the ability of legislators and groups to commit to these kinds of implicit contracts. 2 \Acquiring policy expertise" here means learning about the relationship between policy proposals and outcomes (cf. Gilligan and Krehbiel 1987) as well as learning about the policy preferences of other political actors, including other members of Congress, government ocials, etc. 1
2
process. Those members who engage in legislative entrepreneurship engage in the kinds of activities that are necessary to enact legislation and in uence policy. A member who engages in entrepreneurial activity on a group's behalf will generally further the group's policy goals more than a member who simply votes for legislation that the group favors. Thus interest groups should invest in those individuals whom they think are capable entrepreneurs. An empirical implication of this argument is that we should observe that past entrepreneurial activity will a ect the amount of contributions members receive. To determine whether such a relationship exists between legislative entrepreneurship and campaign contributions I compute measures of entrepreneurial behavior for individual members of the U.S. House of Representatives using detailed data on bill sponsorship and congressional hearings from the 97th through the 101st Congress. From di erent characteristics of bills, such as number of cosponsors and the number and variety of issues that the bills address, I determine the \entrepreneurial quality" of bills and credit their primary sponsors accordingly. The techniques that I use to estimate the e ects of the entrepreneurship measures on PAC contributions take into consideration possible econometric problems raised by treating contributions as investments. PACs will consider a variety of activities other than legislative entrepreneurship when making contribution decisions. It is prohibitively costly to obtain data on all of the possible activities that PACs take into account when deciding to whom they should contribute. Yet not accounting for these other activities will contaminate the inferences we make about legislative entrepreneurship. In order to get around this problem, I use panel data methods, treating other factors in the PAC contribution calculus as e ects that are speci c to individual members. These methods enable us to cleanly estimate the e ects of legislative entrepreneurship, though there are other potential econometric problems| simultaneity bias in particular|which I do not address that may render this analysis suspect. The results that I obtain indicate that legislative entrepreneurship does not play a signi cant part in investor PACs' contribution decisions. The paper is organized as follows. In section 1 I de ne the concept of legislative entrepreneurship. In section 2 I explain why we should nd a relationship between entrepreneurship and campaign contributions. In section 3 I discuss the measures of entrepreneurial activity I use in the empirical analysis. Section 4 presents the methods and empirical analysis. Section 5 discusses the ndings of the paper and mentions plans for future work.
3
1 Legislative Entrepreneurship De ned Members of Congress who assume the role of legislative entrepreneurs (LEs) work to form coalitions that can pass legislation, combining various legislative inputs and issues in order to a ect legislative outcomes.3 LEs invest time and e ort to become aware of existing opportunities for enacting legislation that others have failed to notice and gather information about how to combine various legislative inputs to exploit these opportunities. They use this information to supply a legislative package that has the potential to attract a winning coalition. A legislative package includes the actual text of the proposal as well as information about the proposal's chances for becoming law and the consequences of the proposal if Congress actually passes it. Thus the legislative package reduces uncertainty about the consequences of di erent legislative alternatives and uncertainty about what other members want to enact. Legislative packages are mechanisms for pooling members' resources. The most important resources that LEs pool are members' votes. Legislative packages can pool other resources, such as those associated with a member's position in the institution. For example, LEs might draw on a committee chair's authority to schedule hearings to inform others of their proposals. However, these other resources are useful for LEs only if they can use the resources to attract votes. The central goal of LEs is to gather enough votes to pass legislation|or at least to make credible threats of passing legislation. In order to pass legislation, LEs must convince a majority of the chamber as well as other key players involved in the legislative process that the LEs' proposals will make them better o . In sum, legislative entrepreneurship involves four main activities: acquiring information, bill drafting, coalition building, and pushing legislation.
2 Investment in legislative services Several studies have examined what factors a ect the decisions of PACs to make campaign contributions. Most studies posit that PACs are rational actors that seek to in uence policy through contributions to candidates for Congress. PACs have been separated into two types with each type pursuing di erent strategies to attain their goals. The rst type, which have Though I will often refer to members as legislative entrepreneurs, I am sacri cing conceptual clarity for ease of exposition. We cannot divide the House into entrepreneurs and non-entrepreneurs because members engage in entrepreneurial activity to di erent degrees at di erent times. \Legislative entrepreneur" is one of many roles that members can assume. A more thorough development of the concept of legislative entrepreneur can be found in Wawro 1997. 3
4
been labeled as as \ideological PACs" (Wright 1985), attempt to in uence policy by making contributions in ways that a ect the composition of Congress.4 An ideological PAC tends to give money to candidates who are sympathetic to its views in the hope that the candidates will win oce and then shape policy in ways the PAC favors. The contributions of ideological PACs should mainly be driven by the degree to which candidates|both incumbents and challengers|are aligned with the PACs' interests. PACs in the other category|which Snyder (1992) labels as \investor PACs"|are less concerned with shaping the composition of Congress than they are with in uencing the behavior of those individuals who are already members and are likely to remain so.5 These PACs contribute money to incumbents who are not necessarily favorably disposed to the PACs' interests with the expectation that the contributions will encourage incumbents to behave in ways that further the PACs' goals. The contributions of investor PACs should mainly be driven by their beliefs about the degree to which an incumbent can promote the PACs' interests. Members will vary in their ability to serve the interests of PACs based on natural talent, constituency constraints, and the resources they possess by virtue of their position in the institution. Numerous empirical studies have examined the relationship between PAC contributions and members' roll call behavior.6 Many of these studies have failed to nd a link between PAC money and votes. Hall and Wayman (1990) argue that this is at least partly due to the fact that legislators face severe restrictions on the way they cast their votes on the oor. Because of the visibility of roll call votes, constituency constraints prevent legislators from voting one way or another at the behest of an interest group. Thus campaign contributions aimed at in uencing members' votes are relatively poor investments. Hall and Wayman argue that members have more exibility in how they can spend their time and allocate their legislative resources than they have when they cast oor votes. This is especially true at the committee stage of the legislative process where members feel less constrained by their constituents than they do when they cast their votes on the oor (Hall 1996). The profound impact that committee deliberations can have on nal legislative outcomes means that interest groups who give to members involved in these deliberations can receive large returns on their investments. Thus we should observe a positive relationship between campaign contributions and members' involvement at early stages of the process. In their empirical analysis Hall and Wayman do nd that for three separate pieces of legislation, These PACs have been closely identi ed with what the Federal Election Commission (FEC) ocially designates as nonconnected PACs |that is, they are not formally connected with a parent organization. 5 Investor PACs are those that are aliated with corporations, labor unions, trade associations, or cooperatives. 6 See Snyder 1992, n. 1, for a thorough list of these studies. 4
5
contributions from concerned PACs in the previous election cycle had a positive impact on members' involvement during subsequent committee deliberations. However, the connection between contributions and legislative activity does not seem to extend beyond the committee room. Ragsdale and Cook (1987) examined the e ect of bill sponsorship and cosponsorship on contributions to incumbents. They did not nd a statistically signi cant relationship between the two. But, as with roll calls and contributions, does the lack of a relationship between general legislative activity (as measured by Ragsdale and Cook) imply that contributions related to this type of activity are poor investments? For a PAC to invest in members with the expectation that they will engage in legislative entrepreneurship on the PAC's behalf, members must have some leeway in deciding how to allocate their resources. Morton and Cameron (1992) raise the possibility that members' accountability to their constituents imposes signi cant constraints that can prevent them from providing legislative services to groups. Morton and Cameron argue that providing services to interest groups can hurt members reelection chances because their constituents might react negatively (cf. Denzau and Munger 1986). Since the provision of services to interest groups might decrease members' e ectiveness in taking care of their constituents, voters might punish members who \sell out" in exchange for campaign funds. If this is true, it would be unwise for groups to invest in members with the expectation they will engage in activities like legislative entrepreneurship to the bene t of the groups. But for such constituency constraints to be binding, voters must have access to reliable information to determine whether or not their member has sold out. Essentially, there are two ways constituents can assess whether their member has sold out: through personal experience and through information provided by a challenger or by groups aligned with a challenger. Constituents can personally determine if their member has sold out if they seek help from their member's oce. If members devote a substantial amount of their resources to providing services to interest groups, this can detract from their capacity to take care of their constituents' needs relating to casework and constituency service. A poor constituency service operation is something that constituents can directly observe and have personal experience with. Given the important role that constituency service plays in members' reelection strategies (Fiorina 1989; Cain, Ferejohn, and Fiorina 1987), providing services to interest groups can hurt members' reelection chances if it prevents them from taking care of their constituents' needs. Constituents could also nd out about their incumbent's fundraising practices by looking at FEC reports to determine how much money he has raised. Constituents can infer that an incumbent who has raised substantial amounts of money has sold out and they should punish him accordingly. The FEC's reporting requirements ensure the availability of a wealth of 6
information about the amount and sources of campaign contributions. However, it seems highly unlikely that constituents would take the time and e ort necessary to collect this information. One might argue that that constituents can infer whether their incumbent has sold out by observing campaign expenditures (cf. Morton and Cameron 1992, 84). For example, if constituents observe a large number of television or radio commercials in favor of the incumbent, then they can conclude that the incumbent must have sold out in order to acquire the funds necessary to saturate the airwaves. But this can be a very poor indicator to base one's voting decision on. Incumbents who spend a lot of money may be doing so in response to a strong challenger (Jacobson 1980). Unless a challenger has substantial personal wealth and spends her own money on the campaign, she must have raised considerable amounts of money herself in order to mount a serious challenge.7 Voters should be concerned that they would be throwing one sell out out of oce in exchange for another sell out. Constituents should be skeptical about any signals from a challenger about the incumbent's fundraising practices, since the challenger might have engaged in similar practices in order to raise enough money to mount a serious challenge. Generally, the size of an incumbent's war chest will be an unreliable indicator of whether he has sold out in the sense that he is providing services to interest groups and forsaking his constituents. In terms of policy services, an incumbent can work in policy areas that concern his constituents and in ways that bene t constituents as well as interest groups in order to amass campaign funds. Interestingly, the amount of money an incumbent raises can determine whether or not she even faces a challenger who can inform constituents about the incumbent's behavior. Epstein and Zemsky (1995) show formally how war chests can discourage quality challengers from entering races. Box-Ste ensmeier (1996) presents empirical evidence that war chests deter the timely entry of challengers into congressional races. Thus the amount of money an incumbent has raised can a ect election outcomes without the incumbent ever having to spend a dime to in uence a constituent's vote. Though constituents might be upset if their member provides services to interest groups, the amount of money the incumbent gets in return for such services can prevent a challenger from emerging to inform voters that they should be upset and to give them a viable alternative to vote for in the election. Though incumbents are ill-advised to ignore the demands of their constituents when making decisions about how to spend their resources and allocate their time, the barriers that voters face in acquiring reliable information about their incumbents' relationship with interest groups grant incumbents substantial leeway to provide services to groups. Furthermore constituency constraints are less binding if groups share the same policy interests as Jacobson (1997, 40) estimates that for recent elections challengers must raise and spend close to $600,000 in order to run a competitive campaign for the House. 7
7
members' constituents. If by looking after groups' policy interests a member is simultaneously looking after the policy interests of her constituents, it will be dicult to make the case that the member has sold out. Thus a group that invests in a member whose constituents have similar policy interests as the group is more likely to realize a return on its investment. I would argue that the failure to nd a relationship between general legislative activity and campaign contributions is not due to the fact that interest groups are reluctant to invest in members because they face constraints that prevent them from engaging in legislative activities that bene t groups. I suspect that the past e orts have not found a connection between campaign contributions and legislative activity because they have poorly measured the kinds of activities that groups care about. The number of bills sponsored or cosponsored is not a very accurate measure of the level of legislative services that members can provide to interest groups. Members introduce many bills but then do nothing about them. An interest groups seems to get little in return for campaign contributions if a member simply introduces a bill on a subject the group cares about. Bill introduction without any e ort to pass the legislation is not likely to provide the group with any substantive policy bene ts. It seems reasonable that groups should want to see some sort of policy return on their investment which typically requires at least a credible threat of passing a bill (Ferejohn and Shipan 1989).8 We can use the bills that members sponsor to come up with measures of legislative activity that are better than simply the total number of bills sponsored. Bill sponsorship provides a way for members to formally associate themselves with particular legislative causes. As lead sponsor of a bill members can claim some credit for what happens to the bill (Schneier and Gross 1993). But for bill sponsorship to be much of a measure of legislative activity it is necessary to look at the characteristics of bills to get a sense of the amount of e ort that members are putting into their legislation. I contend that the kinds of e ort that interest groups are interested in are the activities that I de ne as legislative entrepreneurship. The next section discusses how we can get a better sense of the entrepreneurial e ort members exert by looking at di erent characteristics of the bills that they sponsor. Some interest groups may want to maintain the status quo. This may require members to build coalitions against passing legislation. For example, Thomas Bliley, who currently chairs the House Commerce Committee, has been entrepreneurial against others' legislative e orts to regulate the tobacco industry. Bliley has attempted to pick apart majority coalitions that are bent on imposing more regulations on tobacco usage. Bliley has engaged in entrepreneurial activity in defense of the tobacco industry by o ering amendments that would weaken others' anti-tobacco proposals (Duncan 1993). 8
8
3 Measures of Legislative Entrepreneurship To compute measures of legislative entrepreneurship I collected data on bill characteristics from the Congressional Research Service's Legislative Information Files. The CRS les contain detailed information on every piece of federal legislation introduced from the 93rd through the current Congress. Members of Congress and their sta rely on this data base for information about the content and progress of legislation. Therefore the CRS les are a reliable source for the kinds of data required for an empirical analysis of legislative entrepreneurship.9 I develop several indicators of legislative entrepreneurship from the CRS les. For each of the indicators I compute the average of the measure per bill introduced by the member. I then compute an entrepreneurship scale score for each member using each of these measures. I compute the entrepreneurship scores from only those bills that had hearings. Filtering out bills on the basis of hearings eliminates frivolous bills that members introduced and then paid no attention to. Filtering on this criterion also provides some measure of bill progress and the degree to which others are concerned with the legislation. I develop the following measures.
3.1 Cosponsorship as a measure of coalition building One of the most important aspects of legislative entrepreneurship is coalition building. A way to measure a member's coalition building skills is to look at the number of cosponsors a member gets to sign onto her legislation. The number of cosponsors a member can get to sign onto a bill is indicative of the member's ability to convince others that the bill is worthy of their support. Members exert \signi cant e ort to recruit members as cosponsors" and use the number and diversity of cosponsors to make claims about the support for the legislation (Campbell 1982, 415). If many legislators sign on to a proposal, the primary sponsor can argue that the bill has a good chance of passing and encourage others to \jump on the bandwagon" in support of the legislation. Thus one indicator of legislative entrepreneurship is the number of cosponsors of members' proposals. The CRS data contain lists of cosponsors for each bill. For each member I add up the number of cosponsors for each bill sponsored by the member. I then calculate the I obtained access to the CRS les via telnet through the Library of Congress. I used programs written in the TCL language and run by expect to collect the data. expect enables one to automate the search and retrieval process that one can do manually over the Internet. I retrieved the data by searching for bills sponsored by each member of Congress. I then downloaded the raw text les and used text processing programs to cull the relevant information from the les. 9
9
average number of cosponsors per bill for each member.10 As a complement to the general cosponsorship measure, I compute a leadership cosponsorship measure. While a bill must obtain the support of a majority of members in the House in order to pass, a bill that has the support of members in formal leadership positions has a better chance of making it to a nal vote on passage than one that does not. The chair of a committee who supports a bill is more likely to schedule the committee meetings necessary for the bill to advance to the later stages of the process. If the Speaker supports a bill then it is more likely that he will schedule the bill for consideration on the oor. Since it is generally more dicult to convince leaders to cosponsor proposals (see Campbell 1982), leadership cosponsorship is an indication of superior coalition building e orts. To compute this measure I determine the number of times that party and committee leaders signed onto a member's proposals.11
3.2 Issue grouping Another important aspect of legislative entrepreneurship is the grouping of issues together to attract a majority coalition (cf. Riker 1986). The CRS data contain information that I use to measure how members group issues together in the legislation that they introduce. Each bill le in the CRS data contains a list of index terms that indicate the number and variety of issues the bill addresses. The index terms can be used to locate bills that deal with particular subjects of interest. As a measure of issue grouping, I count the number of index terms in the lists for each bill and determine the average number of index terms per bill. This is a very crude measure of the phenomenon I am interested in. When drafting bills legislative entrepreneurs do not necessarily try to maximize the number of issues their bills address. They combine issues in certain ways. Still, it may be that the higher the \dimensionality" of a bill|i.e., the larger the number of issues it addresses|the greater the probability that it will attract more support. In an attempt to broaden a coalition, a member may draft a bill that addresses a Each of these measures accounts for duplicate bills that members introduce. Not accounting for duplicate bills would result in serious measurement error. For the cosponsorship measure I add the number of cosponsors of duplicate bills to the number of cosponsors of the original bill. 11For Democrats, party leaders include the Speaker, the Majority Leader, Majority Whip, Chief Deputy and Deputy Whip, and the Chairman and Secretary of the Democratic Caucus. For Republicans, party leaders include Minority Leader, Minority Whip, Chief Deputy and Deputy Whips, the Chairman, Vice Chairman, and Secretary of the Republican Conference. Committee leaders are full committee chairs for Democrats and ranking minority members on full committees for Republicans. The data on party leaders are from Congressional Quarterly special reports. The data on committee and subcommittee chairs are from various editions of the Congressional Directory and Legi-Slate. 10
10
variety of issues that many members care about so that they will become interested in the legislation.
3.3 Bill titles Another measure of legislative entrepreneurship is the number of titles in a particular bill. A bill with multiple titles is usually a complicated piece of legislation that addresses a broad range of issues. Also, bills that include multiple titles are the kinds of legislation that only members with extensive knowledge and well{developed parliamentary skills can draft and push through the legislative process.12 The CRS data contain digests for each bill. A team of CRS lawyers puts together the digests which concisely describe the contents of each bill. As part of this description the digest lists the number of titles included in the bill, if there are any. To compute this measure I simply count the number of titles in each bill digest and determine the average number of titles in a member's bills.
3.4 Knowledge of Policy To compute a measure of policy knowledge I use data on committee hearings.13 Members often testify before committees regarding legislation that concerns them. They even testify before their own committees. In some instances members testify before committees when they are the primary sponsor or a cosponsor of a bill.14 The fact that members appear as witnesses rather than assuming their usual role of asking questions of witnesses during hearings is a signal about their knowledge of and engagement with a particular topic.15 To compute this measure I count the number of times that members testify before committees in the House in each Congress, excluding testimony before the Appropriations comOne diculty with this measure is that bills with multiple titles are typically omnibus bills which are the result of many di erent members' legislative e orts. Senior members, especially those in the party or committee leadership, usually sponsor such major legislation. This measure might just be another way to measure seniority. However, it is not uncommon for junior members to draft and push these kinds of complicated bills. Junior members who sponsor bills with multiple titles can send a strong signal to others about their entrepreneurial abilities. 13The data on committee hearings are from the Congressional Information Service's Congressional Master le 2 CD-ROM. 14For example, in the 100th Congress members testi ed before committees 1,723 times (not including hearings before Appropriations). In 197 cases the members were the primary sponsors of the bills that were the subject of the hearings. In 320 cases the members who testi ed were cosponsors. 15Legislators do not have to appear as witnesses to participate in hearings held by committees on which they do not sit. Members often sit in on other committees' hearings and participate in the questioning of witnesses. 12
11
mittee. Members typically testify before Appropriations regarding the merits of public works projects that are of interest to their districts. Such testimony does not appear to be signaling the kind of expertise that we are interested in, but rather indicates members' attempts to secure pork for their districts. I divide the sum of the testimonies for each member by the total number of hearings that were held during the Congress.
3.5 Legislative Entrepreneurship Scale Score In order to simplify the analysis I compute a scaled entrepreneurship score. This score combines the di erent entrepreneurship measures into a single score which is measured in terms of the same units as the general cosponsorship measure. For member i I divide each of the ve measures by its standard error in Congress t and then multiply by the standard error of the general cosponsorship measure. I then add these scaled measures together to get the entrepreneurship score for each member.16
4 Empirical analysis The recent theoretical work on which this analysis is based suggests that standard regression techniques are not appropriate for making inferences about the relationship between campaign contributions and entrepreneurial activity. As discussed above this work has modeled campaign contributions as investments where politcal organizations contribute money with the expectation of receiving services from legislators. These services can include private, non{ policy services as well as more public, policy{oriented services.17 Legislative entrepreneurship is only one type of service that PACs would want to consider when deciding in whom they should invest. It would be extremely costly to obtain data on all of the possible activities PACs care about. Yet not accounting for these other activities will contaminate the inferences we make about legislative entrepreneurship. Legislative entrepreneurship might be credited for a ecting campaign contributions when that credit is due to other activities. Panel data methods are useful for accounting for individual{speci c e ects that we do not get to observe. The individual e ect that I contend we need to account for is the (common) belief held by PACs concerning a member's ability to deliver promised services (other than I also computed scale scores derived from a factor analysis of the entrepreneurship measures. The factor scores were highly correlated with the scaled entrepreneurship score, so I suspect that using the former would not change the results of this analysis much. 17Morton and Cameron (1992) contend that the services in these models cannot be policy{oriented because they are nonspatial, and therefore the services must provide private, nonpolicy bene ts. So it is questionable how much guidance for empirical analysis these formal models give other than to indicate the importance of groups' beleifs about members ability to deliver on services. 16
12
entrepreneurial services). A PAC's contribution decisions should be a function of its beliefs about receiving a return on its contributions. These beliefs will depend on factors which are easy to observe|such as a member's position in the institution| as well as on other factors which are not so easily observable. The speci cation that I use tries to separate out these di erent factors by including variables in the model to account for variation in members capacity to provide services.18 The method developed by Chamberlain (1982) is especially attractive for this analysis for several reasons.19 First, it requires that we impose very little structure on the data| we need only assume that the observations on the cross{sectional units are identically and independently distributed and have nite fourth order moments. Second, the method allows for correlation between the explanatory variables and the individual speci c e ects which is likely to exist in this context. Third, it allows for the possibility that the relationship between the explanatory variables and the individual e ects is nonlinear. Finally, this method does not require us to impose a priori restrictions on the variance{covariance matrix and hence we can incorporate serial correlation and heteroskedasticity in the error process. Correlation between the individual e ects and explanatory variables will cause serial correlation and heteroskedasticity. Also the large dollar amounts and substantial variation that characterize campaign contributions are likely to lead to heteroskedastic errors. Suppose we have the following behavioral model y = + x + u ; t = 1; : : : ; T; i = 1; : : : N (1) Assume E (u j ; x ) = 0 (2) where x0 = (x0 1; : : : ; x0 ): We allow to be correlated with not just x but with x : = a0x + ! where ! is orthogonal to x by construction. Stacking the observations for individuals, we can then write the unrestricted reduced form of the model as y = x + v (3) it
i
it
it
i
i
it
i
i
iT
i
it
i
i
i
i
i
i
i
i
i
Note that I am assuming that interest groups' beliefs do not change over time, which may not make sense since other factors in the model which interest groups respond to do vary from period to period. The speci cation could be altered to allow for such variation in the individual{speci c e ects, and I leave this as an extension for future work. 19For a straitfoward description of Chamberlain's method see Hsiao 1986, 57{63. 18
13
where y0 is a 1 T vector with elements y , is a T KT matrix of projection coecients, and v0 is a 1 T vector of disturbances. After estimating the reduced form , we use a minimum distance estimator to back out the s: it
i
i
h
i,
^ = , G0 ^ ,1 G
1
G0 ^ ,1 ^
^ where G is a KT 2(T +1) asymptotic variancep K matrix of zeros0 and ones,0 0 is the0 estimated 0 covariance matrix of N (^ , ), and = vec( ) = [1; : : : ; ]. G is constructed to impose various restrictions on . In the case where the s do not vary over time, = ( 0; a0). h i ,1 The asymptotic variance of ^ = 1=N G0 ^ ,1G . To test the validity of the restrictions we compute T
N [^ , G ]0 ^ ,1 [^ , G ]
which converges to a 2 distribution with KT 2 , K (T + 1) degrees of freedom. A potentially major drawback in using panel data methods is that members exiting the institution will cause our panels to be unbalanced.20 Chamberlain's method requires a balanced panel, which means that we have to exclude from the analysis those members who do not appear in every period covered by the study (namely, the 97th through the 101st Congress). Thus those members who leave Congress during this period must be dropped.21 This introduces a possible sample selection problem and calls into question the inferences we can make from the subpanel. By using panel methods to avoid omitted variable bias, we might be introducing selection bias. We should test to determine whether selection bias exists before applying Chamberlain's method to the balanced panel. A test for selection bias has not been developed speci cally for Chamberlain's estimator, but we can rely on available tests for selection bias for other panel data estimators provided they adopt the same assumptions about the data as Chamberlain's method. Tests and corrections developed for a random e ects framework (see Verbeek and Nijman 1992a, 1992b) are inappropriate because they do not allow for correlation between the individual e ects and the explanatory variables. Verbeek (1990) allows for such correlation in the behavioral equation but not in the selection equation. It seems inconsistent to assume that such correlation exists in one equation and not the other (Zabel 1992). Nijman and Verbeek (1992) and Zabel (1992) develop methods that allow for correlation between unobserved e ects and If we consider new members entering the institution after the rst time period of the study, this will also lead to unbalanced panels. However, I will only worry about the problems introduced by attrition. 21According to the ICPSR Congressional Biographical Data File of the 723 members who served in the House between the 97th and 101st Congresses, only 206 of them served in every one of these Congresses. If we treat the 98th Congress as the starting wave of a panel, we lose approximately 51% of the sample by the 102nd Congress. 20
14
the explanatory variables in both the behavioral and selection equations. However, these methods are quite computationally involved, especially if we allow for serial correlation in the errors of the selection equation. The method developed by Wooldridge (1995) for a xed e ects estimator is the most appropriate for testing for selection bias considering the assumptions adopted for Chamberlain's estimator. Wooldridge's method, like Chamberlain's, imposes minimal distributional assumptions on the data and allows for serial correlation and heteroskedasticity while remaining computationally simple. The conditions that are required for Wooldridge's estimator of the behavioral coecients to be consistent are essentially the same as are required for consistency of the Chamberlain estimator. Suppose we have the behavioral model given by (1) but (y ; x ) are observed only if the indicator variable s = 1. The xed e ects estimator for is consistent for the selected subpanel (i.e., those observations for which s = 1) if it
it
it
it
E (u j ; x ; s ) = 0; it
i
i
t = 1; : : : ; T;
i
(4)
where x = (x 1; : : : ; x ) and s is de ned similarly. A sucient condition for (4) to hold is i
i
iT
i
E (u j ; z ; v ) = 0; it
i
i
t = 1; : : : ; T:
i
(5)
where z contains the explanatory variables from the selection equation (including elements of x ) and v contains the disturbances for the selection equation in each period. The test for selectivity bias amounts to testing whether E (u j ; z ; v ) is nonzero. The two-step method proceeds as follows. We rst estimate P (s = 1jz ) = ( z ) by probit and compute an estimate of the inverse Mills ratio ^ = (^ z ). We then estimate i
i
i
it
i
i
i
it
it
y~ = x~ + ~ + error it
it
it
P
t
i
t
i
i
(6)
it
where y~ = y , T ,1 =1 s y and x~ and ~ are de ned similarly. We estimate (6) by OLS, pooling the observations for which s = 1, and compute asymptotic standard errors that are robust to serial correlation and heteroskedasticity and account for the fact that ~ is a generated regressor. If we do not reject H0 : = 0 using a t-test, then it is valid to use the subpanel to make inferences about the population. Wooldridge's method assumes that even though we do not get to observe the dependent variable for all cross{sectional units in every time period, we always observe the explanatory variables. This is not the case with the current analysis. Once members leave Congress we do not get to observe explanatory variables of interest. However, we do get to observe the explanatory variables up until the period when they leave. In order to get around this problem, we can use the data that we have on members in previous Congresses in order to estimate the selection equation. That is, we estimate T
it
it
i
r
ir
ir
it
it
it
15
P (s = 1jz) where z includes (zit,1 ; zit,2 ; : : :). Lagging the variables like this is consistent it
i
i
with thinking about contributions as investments, because investment decisions should at least in part be based on past behavior. The selection equation has the following variables in common with the behavioral equations.22 In addition to the main variable of interest, the scaled entrepreneurship score,23 I include several other explanatory variables that might have an e ect on campaign contributions and whether a member remains in or leaves Congress. I include two dummy variables that indicate the institutional position of the member. Members in leadership positions possess resources that enable them to have more in uence over bureaucrats as well as the legislative process. A PAC may want to give money to these individuals in an attempt to get them to use some of their in uence on the PAC's behalf. I include these variables because I want to determine if entrepreneurial activity has an e ect on contributions that is independent of the formal positions that members hold in the House. These variables should also a ect selection because members who occupy these positions may be less willing to give up their seats or may be more electorally secure because of these positions (Hall and Van Houweling 1994). The rst dummy indicates whether or not the member occupies a full committee or subcommittee leadership position. This variable equals one for Democrats if they hold a full committee or subcommittee chair and one for Republicans if they are ranking minority members of committees or subcommittees. The second dummy indicates whether the member holds a party leadership position. Party leadership positions here are the same as those identi ed for the leadership cosponsorship score (see n. 11). The increasing ideological polarization in Congress during the period of analysis suggests more moderate members might have a higher probability of exiting the institution (Poole and Rosenthal 1997). To measure this e ect, I include the absolute value of the di erence between a member's NOMINATE score and the median score in the chamber.24 Investor PACs might take into consideration the ideological extremity of members, but it is not clear whether this would have a positive or negative a ect on contributions. Investor PACs should also be concerned with the ideological direction of a member's entrepreneurial e orts. For the speci cation of the selection equation I draw on studies that examine why members leave Congress (e.g., Groseclose and Krehbiel 1994, Hall and Van Houweling 1994, Kiewiet and Zheng 1993). 23In other work, I have argued that a Bayesian updating formulation should be used to model how others assess the entrepreneurial ability of an individual. This requires requires computing a variable that measures a member's entrepreneurial activity over her entire career. While I do not explicitly compute such a variable in this analysis, Chamberlain's method in a sense takes past entrepreneurial behavior into account by allowing correlation between the unobserved e ects and the explanatory variables in past periods. 24To be precise, I use the DW-NOMINATE rst dimension score. These scores range from {1 to +1, with increasing scores indicating increasing conservatism of the member. It may be more appropriate to include the absolute deviation from the median in the member's caucus. 22
16
Though the entrepreneurship scores alone give us no indication of this direction , we can crudely measure this by including a term that interacts a member's NOMINATE score with his entrepreneurship score. The speci cation also includes the percentage of votes the incumbent received in the previous election.25 Potential challengers might view incumbents who barely squeaked by in the previous election as vulnerable, and PACs might want to help vulnerable incumbents build a war chest with the expectation that they will face a quality challenger in the next election. Vulnerable incumbents might feel more indebted to their contributors than those members who are not vulnerable. However, investor PACs might be reluctant to contribute to insecure incumbents because they may not remain in oce to assist the PACs. Vulnerable incumbents might also decide to retire rather than face a grueling reelection campaign. Following Snyder 1992, I include the square of previous vote share as well. I also include variables that measures a member's age and seniority. Older and more senior members should be more likely to retire or die.26 Senescent members will be less attractive investments as a consequence, though more senior members are likely to be in positions where they can further PACs' interests. Snyder (1992) nds that age (seniority) has a statistically signi cant and negative (positive) e ect on investor PAC contributions. Unfortunately, these variables cannot be included in the Chamberlain formulation because they would induce perfect multicollinearity in the regressors. I also exclude some variables that previous empirical analyses of investment models have found to be important predictors of PAC contributions. Some variables, such as party, district{to{state ratio and other district characteristics, do not vary over time for members and so are inappropriate for a xed e ect framework. The selection equation includes one variable that the behavioral equations do not. A member may be more likely to leave the House to run for the Senate if she does not have to face an incumbent Senator. However, this variable should not a ect investor PAC contributions to House incumbents in the Congress following the one in which there was an open Senate seat. Thus I include in the selection equation but not in the behavioral equations a dummy variable indicating whether there is an open seat in the Senate for a member's state. While one could easily make the case for including other variables in the selection equation, I try to keep the speci cation as simple as possible since it is not the main focus of this analysis. The variables that I include should capture enough of what is going in the selection process in order to conduct the test for selection bias.27 It is possible that this variable is endogenous to entrepreneurial activity. Though a quadratic speci cation may be more appropriate with these variables, both the probit and panel data equations became dicult to estimate when these variables were included. 27Wooldridge (1995, 119) notes that \the selection mechanism need not be correctly speci ed in any sense; 25 26
17
Table 1 and Table 2 include descriptive statistics of the variables used in this analysis for the subpanel of members who served in every Congress from the 97th through the 101st and for the panel of members who served at least in the 97th Congress.28 For purposes of comparison Table 3 reports the results of OLS estimation of the behavioral equation for the balanced and unbalanced panels. Note that the coecient on the main variable of interest, the entrepreneurship score, is positive and statistically signi cant as expected. The e ect of this variable is marginally substantively signi cant|a member who increased her entrepreneurship score by one standard deviation would receive approximately $15,000 more in investor PAC contributions in the next election cycle. The results of the estimation of the selection equation for each period appear in Table 4. The variables that consistently have signi cant coecients across di erent Congresses are previous vote share and the square of previous share. The probability of being in the next Congress increases with previous vote share but this e ect decreases as the vote share approaches one. The estimated coecients from the selection equation were then used to generate the regressor ~ to test for selection bias. The results from this test are reported in Table 5. The estimated cocient on ~ is only slightly larger than its standard error, so a standard t test does not lead us to reject the null hypothesis that this coecient is in fact zero. Thus the selection rule is \ignorable" and it is appropriate to proceed to estimate and make inferences from the balanced subpanel of members. I rst estimated the e ects of the explanatory variables on total investor PAC contributions using Chamberlain's method by restricting the coecients on the explanatory variables to be the same across time periods. The results are reported in Table 6. The estimates were obtained using the minimum distance estimator described above. Contrary to previous estimates, Chamberlain's method produces an estimated coecient on the entrepreneurship variable that is negative. However, the 2 statistic for testing the imposed restrictions on the ^ matrix, indicates that we should not make inferences based on these restrictions. The computed statistic, 2152 = 1569:72, is highly statistically signi cant and leads us to reject the null that the s are the same for all time periods. Table 7 reports the minimum distance estimates and standard errors where the s are allowed to vary over time. The 2 statistic here indicates these restrictions are approit is simply a vehicle for obtaining a sensible test." One could argue that the amount of incumbent contributions in the previous election should be included in the selection equation. Some argue that incumbent contributions/spending has an important e ect on elections outcomes (Green and Krasno 1988) while others argue it does not matter relative to challenger spending (Jacobson 1980). Whatever the case, the selection test employed here assumes that the right hand side variables are strictly exogenous, and so this test is inappropriate if we want to include past contributions to incumbents in the selection equation. 28The campaign nance data are from The Federal Election Commission's Campaign Expenditures in the United States, Reports on Financial Activity (RFA) Data, House Spread Data, 1984{1992.
18
priate ( 2152 = 143:273), which makes sense given the variability in the estimated coecients across time periods. In periods 1 and 5, the coecients on the entrepreneurship score are positive, while they are negative in the other periods. In none of the periods are the e ects substantively signi cant, though they are statistically signi cant.29 Legislative entrepreneurship seems to matter little for investor PAC contributions, and this activity certainly does not seem to matter in the way investment models predict. The coecients on the Entrepreneurship/NOMINATE interaction term are consistently negative, indicating that liberal entrepreneurs collect more investor PAC contributions|however, this e ect is also substantively small. The coecient on the NOMINATE score is negative in all periods except the rst. I suspect this e ect is due to the fact that Democrats, who are more likely to have negative NOMINATE scores, controlled the House during the periods under analysis, and hence were more attractive investments for PACs than were Republicans. Similar to the results in Snyder 1992, the vote share in previous election and its square have consistently negative and positive coecients, respectively, though there is substantial variability in their substantive impact. Leadership position variables also vary substantially in their e ects, with the party leadership dummy having the biggest e ect in period 5.
5 Discussion This paper advances existing research in two ways. First it pushes the analysis of campaign nance and legislative activity beyond the committee stage. Though the nascent stages of a bill are important in determining the nal product, the rise in oor activity in the post reform Congress has increased the importance of post{committee deliberations (Bach and Smith 1988; Smith 1992). The measures of entrepreneurial activity account for members' legislative e orts during and after the committee stage. The measures of legislative activity used here are more detailed than those used in previous analyses. Previous studies have relied on number of bills sponsored or cosponsored as measures of legislative activity. The measures I use here go beyond simple sponsorship, tapping into such activities as bill drafting, coalition building, and the communication of policy{relevant information. Second, this analysis takes into consideration that PACs are interested in a wide range of legislators' activities and any analysis of the relationship between a particular activity and campaign contributions needs to account for these other, possibly unobservable e ects. Panel data methods like the ones I employ in this analysis enable us to cleanly estimate the e ects of certain activities, even when these activities might be correlated with other unobservable factors that can a ect contribution decisions. The methods that I use here require very few Nearly all of the coecients are statistically signi cant which raises questions about the computation of the standard errors for the minimum distance estimator. 29
19
assumptions about the data and provide a way to test whether the manipulations of the data that are required for a panel analysis introduce bias. The results obtained using panel methods are di erent from those that are produced by simpler methods snd would lead to di erent inferences. While more standard methods might lead us to conclude that investor PACs incorporate legislative entrepreneurship in their contribution decisions, such e ects appear to be incorrectly attributed to entrepreneurial activity. Once individual speci c e ects|which I contend are PACs beleifs' about members' abilities to engage in other types of activities|are accounted for, there is no consistent pattern that would indicate that legislative entrepreneurship plays an important role in investor PACs' contribution decisions. While investor PACs might care about this activity, commitment problems may make it unattractive for PACs to invest in members with the expectation that they will engage in entrepreneurial activity to the bene t of the PACs (cf. McCarty and Rothenberg 1996). To conclude I mention a few caveats and extensions for future work. Though Wooldridge's test for selection bias is appropriate when the data are generated according to Chamberlain's model, critics might feel more comfortable with a selection test explicitly developed for Chamberlain's set{up. Such a test would involve something like estimating a systems of selection equations where each equation is a probit model corresponding to a particular time period and would require evaluating a T {order multivariate normal integrals. Given the computational complexity involved, it is questionable whether such a test is feasible. Though the methods I have used in this paper address important econometric problems with analyzing campaign nance, some problems remain that cast doubt on the reliability of the inferences I have drawn. The most important problem is that campaign contributions might be simultaneously determined with entrepreneurial activity (cf. Hall and Wayman 1990). PACs should expect a return on their investment, thus past campaign contributions should a ect a legislator's activities at some future point. Also if past activities bring in more campaign contributions then as members need for campaign contributions increases, they should increase the activities that bring in such contributions. It may be more appropriate then to treat entrepreneurship and other activities as endogenous and employ a technique such as instrumental variables to correct for this problem. While it is conceivable that the panel data methods employed in this paper could be extended to account for simultaneity bias, I suspect that they would become prohibitively complicated. There is also the devilishly tricky problem of nding variables that are correlated with entrepreneurial activity but not correlated with campaign contributions. Nevertheless, work continues along these fronts.
20
References Bach, Stanley and Steven S. Smith. 1988. Managing Uncertainty in the House of Representatives. Washington, D.C.: Brookings Institution. Baron, David P. 1989a. \Service{Related Campaign Contributions and the Political Equilibrium." Quarterly Journal of Economics :104:45{72. Baron, David P. 1989b. \Service-induced Campaign Contributions, Incumbent Shirking, and Reelection Opportunities." In Models of Strategic Choice in Politics, ed. Peter C. Ordeshook. Ann Arbor: University of Michigan Press. Box-Ste ensmeier, Janet. 1996. \A Dynamic Analysis of the Role of War Chests in Campaign Strategy." American Journal of Political Science 40:352{371. Cain, Bruce E., John Ferejohn, and Morris Fiorina. 1987. The Personal Vote: Constituency Service and Electoral Independence. Cambridge, MA: Harvard University Press. Chamberlain, Gary. 1982. \Multivariate Regression Models for Panel Data." Journal of Econometrics 18: 5{56. Denzau, Arthur, and Michael C. Munger. 1986. \Legislators and Interest Groups: How Unorganized Interests Get Represented." American Political Science Review 80:98{ 106. Duncan, Phil, ed. 1993. Politics in America 1994. Washington, D.C.: CQ Press. Epstein, David and and Peter Zemsky. 1995. \Money Talks:A Signaling Approach to Campaign Finance." American Political Science Review 89:295{308. Fiorina, Morris. 1989. Congress: Keystone of the Washington Establishment. 2d ed. New Haven: Yale University Press. Gilligan, Thomas W., and Keith Krehbiel. 1987. \Collective Decisionmaking and Standing Committees: An Informational Rationale for Restrictive Amendment Procedures." Journal of Law, Economics, and Organization 3:287-335. Green, Donald P., and Jonathan S. Krasno. 1988. \Salvation for the Spendthrift Incumbent." American Journal of Political Science 32: 844{907. Grier, Kevin B., and Michael C. Munger. 1986. \The Impact of Legislator Attributes on Interest Group Campaign Contributions." Journal of Labor Research 7: 349{61. 21
Grier, Kevin B., and Michael C. Munger. 1991 \Committee Assignments, Constituent Preferences, and Campaign Contributions." Economic Inquiry 29:24{43. Groseclose, Timothy and Keith Krehbiel. 1994. \Golden Parachutes, Rubber Checks, and Strategic Retirements from the 102nd House." American Journal of Political Science 38:75{99. Hall, Richard L. 1996. Participation in Congress. New Haven: Yale University Press. Hall, Richard L. and Robert Van Houweling. 1994. \Avarice and Ambition in Congress: Representatives' Decisions to Run or Retire from the U.S. House." American Political Science Review 89: 121{136. Hall, Richard L. and Frank W. Wayman. 1990. \Buying Time: Moneyed Interests and the Mobilization of Bias in Congressional Committees." American Political Science Review 84:799{820. Hibbing, John R. 1991. Congressional Careers: Contours of Life in the U.S. House of Representatives. Chapel Hill, NC: University of North Carolina Press. Hsiao, Cheng. 1986. Analysis of Panel Data. Cambridge: Cambridge University Press. Kiewiet, D. Roderick and Langche Zeng. 1994. \An Analysis of Congressional Career Decisions, 1947-1986." American Political Science Review 87: 928-941. McCarty, Nolan and Lawrence S. Rothenberg. 1996. \Commitment and the Campaign Contract." American Journal of Political Science 40:872{904. Jacobson, Gary C. 1980. Money in Congressional Elections. New Haven: Yale University Press. Jacobson, Gary C. 1997. The Politics of Congressional Elections. 4th ed. New York: Longman. Jacobson, Gary C. and Samuel Kernell. 1983. Strategy and Choice in Congressional Elections. 2nd ed. New Haven: Yale University Press. Morton, Rebecca and Charles Cameron. 1992. \Elections and the Theory of Campaign Contributions: A Survey and Critical Analysis." Economics and Politics 4:79{108. Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political{Economic History of Roll Call Voting. Oxford: Oxford University Press. 22
Ragsdale, Lynn and Timothy Cook. 1987. \Representatives Actions and Challengers Reactions: Limits to Candidate Connections in the House." American Journal of Political Science 31: 45-81. Schneier, Edward V., and Bertram Gross. 1993. Legislative Strategy : Shaping Public Policy. New York : St. Martin's Press. Smith, Steven S. 1992. \Revolution in the House: Why Don't We Do It on the Floor?" In Robert L. Peabody and Nelson W. Polsby, eds. New Perspectives on the House of Representatives 4th. ed. Baltimore: Johns Hopkins University Press. Snyder, James. 1990. \Campaign Contributions as Investments: The U.S. House of Representatives, 1980{1986." Journal of Political Economy 98: 1195{1226. Snyder, James. 1992. \Long-term Investment in Politicians; Or, Give Early, Give Often." Journal of Law and Economics 35: 15{43. Verbeek, Marno. 1990. \On The Estimation of a Fixed E ects Model With Selectivity Bias." Economics Letters 34: 267{270. Verbeek, Marno and Theo Nijman. 1992. \Incomplete Panels and Selection Bias." In Laszlo Matyas and Patrick Sevestre, eds. The Econometrics of Panel Data. Dordrecht: Kluwer Academic Publishers. Verbeek, Marno and Theo Nijman. 1992. \Testing For Selectivity Bias in Panel Data Models." International Economic Review 33:681{703. Wawro, Gregory J. 1997. Legislative Entrepreneurship in the U.S. House of Representatives. PhD Dissertation, Cornell University. Wooldridge, Je rey M. 1995. \Selection Corrections for Panel Data Models Under Conditional Mean Independence Assumptions." Journal of Econometrics 68: 115{132. Wright, John R. 1985. \PACs, Contributions, and Roll Calls: An Organizational Perspective." American Political Science Review 79:400{14. Zabel, J.E. 1992. \Estimating Fixed and Random E ects With Selectivity." Economics Letters 40: 269{272.
23
Std. Dev. Median
163,153.54 128,496.56 110.702 123.033 .281 .449 .686 .464 .385 .222 .737 .142 .110 .313 51.80 10.28 6.433 3.835 {.156 .445 {32.663 8.345 .564 .224
Mean
Max.
0 1,106,661 0 981.799 0 1 0 1 .001 1.277 .472 1 0 1 27 80 1 25 {.981 .986 {584.67 302.693 .222 1
Min.
Note: The total number of members who served in every Congress from the 97th{101st is 206. Treating each member in each Congress as a unique observation yields a sample size of 1030.
Investor PAC Contributions Entrepreneurship Score Party Leader Committee/Subcommittee Leader NOMINATE Score (Deviation from Chamber Median) Vote Share in Previous Election Open Seat in Senate Age Seniority NOMINATE Score Entrepreneurship/NOMINATE Score Interaction (Vote Share in Previous Election)2
Variable
Table 1: Descriptive Statistics of Variables for Panel of Members, 97th{101st Congress
Std. Dev. Median
Min.
Max.
140,645.60 153,641.83 112,900 0 1,278,949 102.54 119.06 59.90 0 981.799 .239 .426 0 0 1 .654 .475 1 0 1 .388 .226 .389 0 1.277 .736 .163 .701 .374 1 .115 .320 0 0 1 51.88 10.97 52 27 89 6.15 3.91 5 1 25 {.101 .452 {.124 {.981 .986 {26.32 75.59 {1.584 {749.467 302.693 .568 .254 .492 .139 1
Mean
Note: The total number of members who served at least in the 97th Congress is 441. Treating each member in each Congress as a unique observation in the 97th and subsequent Congresses yields a sample size of 1639.
Investor PAC Contributions Entrepreneurship Score Party Leader Committee/Subcommittee Leader NOMINATE Score (Deviation from Chamber Median) Vote Share in Previous Election Open Seat in Senate Age Seniority NOMINATE Score Entrepreneurship/NOMINATE Score Interaction (Vote Share in Previous Election)2
Variable
Table 2: Descriptive Statistics of Variables for All Members of Congress, 97th{101st Congress
Table 3: E ects of legislative entrepreneurship on total PAC contributions{Balanced and unbalanced panels variable Column 1 Column 2 Constant
867,025.458 (114,384.702) Entrepreneurship score 122.350 (39.405) Party leader 77,946.853 (8,401.578) Committee/Subcommittee leader {16,451.732 (8,158.840) NOMINATE Score (Deviation from Chamber Median) {27,960.562 (16,978.995) Vote Share in Previous Election {1,824,205.819 (299,676.571) NOMINATE Score {21,691.769 (11,292.661) Entrepreneurship/NOMINATE Score Interaction 13.200 (75.871) (Vote Share in Previous Election)2 1,108,025.194 (189,786.282) N F ^ 2 R 2
1,030 23.707 118,464.579 .15
30,135.592 (92,030.762) 112.185 (39.442) 70,294.081 (8,762.395) {4,515.417 (7,876.616) {19,654.622 (16,728.634) 216,709.019 (246,655.794) {16,146.396 (11,067.907) 14.624 (77.503) {119,592.031 (158,342.423) 1,639 12.837 149,384.328 .055
Notes: Coecients are OLS estimates. Column 1 contains estimates for balanced subpanel. Column 2 contains estimates for unbalanced panel starting in the 97th.
variable
Table 4: Selection Equation Estimates Period 1 Period 2 Period 3 Period 4 Period 5
Constant Entrepreneurship score Party leader Committee/Subcommittee leader NOMINATE Score (Deviation from Chamber Median) Vote Share in Previous Election Open Seat in Senate Age Seniority NOMINATE Score Entrepreneurship/NOMINATE Score Interaction (Vote Share in Previous Election)2 N
% Correctly Predicted
{12.036 (2.069) .002 ( .001) .096 ( .215) .230 ( .178) .394 ( .344) 35.79 (5.680) { .013 ( .242) { .021 ( .009) { .017 ( .028) .041 ( .240) { .002 ( .003) {22.06 (3.672)
{15.06 {13.05 {12.339 (2.999) (2.753) (3.143) { .001 { .00008 .0008 ( .001) ( .0008) ( .001) .270 { .047 .450 ( .311) ( .240) ( .293) .0070 .338 .270 ( .251) ( .233) ( .273) { .479 .461 .022 ( .485) ( .520) ( .640) 42.77 36.209 34.936 (8.148) (7.103) (8.287) .071 .539 .278 ( .423) ( .402) ( .606) .0006 { .005 { .003 ( .013) ( .013) ( .015) { .068 { .021 { .070 ( .039) ( .038) ( .039) { .092 { .154 { .291 ( .344) ( .339) ( .454) { .002 .0005 .002 ( .002) ( .001) ( .002) {25.18 {21.777 {20.601 (5.344) (4.559) (5.355)
{12.26 (2.678) .002 ( .001) .834 ( .345) .748 ( .268) { .629 ( .615) 35.213 (7.677) .580 ( .321) { .026 ( .016) .071 ( .047) { .208 ( .438) .004 ( .002) {21.114 (5.029)
441 77.9%
356 86.9%
248 86.6%
316 79.8%
278 83.8%
Notes: Coecients are probit MLEs. Standard errors in parentheses.
variable
Table 5: Test for Selection Bias Estimated coecient Asymptotic Standard Error
Entrepreneurship score
7.769
5.210
{4,563.557
1,573.268
Committee/Subcommittee leader
731.224
1,454.321
NOMINATE Score (Deviation from Chamber Median) Vote Share in Previous Election
34,470.262
6,011.332
{323,978.547
55,976.313
Age
{3,293.208
1,514.125
Seniority
{1,837.727
2,964.490
NOMINATE Score
102,327.519
5,000.649
Party leader
Entrepreneurship/NOMINATE Score Interaction (Vote Share in Previous Election)2
{6.6095
10.030
196,836.544
33,986.209
~ (Generated Regressor)
{14,748.493
14,028.97
Notes: N = 361 (i.e., number of cross-sectional units; total number of observations in sample is 1404). Coecients are OLS estimates of and from equation (6). Asymptotic standard errors are computed using the formulas in the Appendix to Wooldridge 1995. R 2 = :577. ^ 2 = 13; 489:515. F11 1393 = 174:563. ;
Table 6: E ects of legislative entrepreneurship on Investor PAC contributions|Panel Data Estimates variable Estimated coecient Standard Error Entrepreneurship score Party leader Committee/Subcommittee leader NOMINATE Score (Deviation from Chamber Median) Vote Share in Previous Election NOMINATE Score Entrepreneurship/NOMINATE Score Interaction (Vote Share in Previous Election)2
{83.55 35,956.54 2,476.26 {45,595.63 {525,190.25 69,589.97 {50.856793 315,405.86
.346 145.19 161.94 620.46 4,167.69 635.51 .558 2,557.47
Notes: N = 206. 2152 = 1569:72 (p < :001). Estimates and standard errors obtained by using the minimum distance estimator described in Chamberlain 1982, where coecients on independent variables are constrained to be constant over time.
Table 7: E ects of legislative entrepreneurship on total PAC contributions|Panel Data Estimates variable Period 1 Period 2 Period 3 Period 4 Period 5 Entrepreneurship score
32.99 (1.30) Party leader {6,351.13 (315.43) Committee/Subcommittee 15,926.68 leader (307.48) NOMINATE Score (Deviation 17,763.87 from Chamber Median) (1,201.91) Vote Share in Previous {655,920.14 Election (10,719.43) NOMINATE Score 1,946.13 (1,048.01) Entrepreneurship/NOMINATE {123.83 Score Interaction (3.139) (Vote Share2in Previous 414,780.64 Election) (6,814.90)
{73.18 {69.77 {70.72 49.48 (.845) (1.128) (.994) (3.146) {2,135.87 14,858.70 1,627.57 75,138.37 (294.41) (233.549) (321.37) (808.43) 8,990.91 {10,350.87 13,994.67 4,033.64 (263.75) (246.372) (370.49) (910.14) 18,177.16 {11,422.80 49,582.32 73,428.10 (1,039.04) (1,059.81) (1,218.58) (1,720.17) {258,208.52 {1,353,719.8 {1,958,774.7 {671,844.56 (8,177.86) (10,169.40) (13,936.19) (32,714.62) {26,866.81 {33,620.57 {37,881.71 {17,692.12 (923.23) (962.65) (1,114.18) (1,558.39) {8.198 {137.48 {213.13 {347.48 (1.442) (1.608) (2.118) (6.342) 161,215.17 831,964.55 1,214,041.6 381,436.43 (5,091.83) (6,149.77) (8,644.20) (20,616.73)
Notes: N = 206. 2152 = 143:273 (p = :681). Standard errors in parenthesis. Estimates and standard errors obtained by using the minimum distance estimator described in Chamberlain 1982, where coecients on the independent variables are allowed to vary over time.
