A Test of the Structure of PAC Contracts: An Analysis of House Gun Control Votes in the 1980s.

Sutter, Daniel

Joseph P. McGarrity [*]

Daniel Sutter [+]

We examine roll call votes on gun control in the U.S. House of Representatives during the 1980s to determine whether political action committees (PACs) make spot market purchases, prepay for votes in the prior election cycle, or make long-term investments. Previous tests generally employ PAC contributions from only one cycle, which could impose the wrong structure on contracts between PACs and politicians, causing researchers to misestimate a contribution's impact. We find that money from more than one election cycle influences roll call votes, which suggests that PAC expenditures are not simple spot market or one-period prepayment contracts. Most remarkably, we find that the National Rifle Association buys votes with contributions from three election cycles.

Enactment of the bill was a major victory for the National Rifle Association, which had worked for years to win relaxation of the landmark Gun Control Act of 1968.

Congressional Quarterly Almanac (1986, p. 82)

1. Introduction

Every election the same political action committees (PACs) contribute millions of dollars to candidates for office. Well-informed participants in the political process surely must get value for their money. Yet empirical research fails to provide systematic evidence that campaign contributions influence congressional roll call votes. [1]

A mischaracterization of the exchange between PACs and politicians may produce these contradictory results. Empirical studies of roll call voting typically use PAC contributions from only one election cycle. Earlier studies generally assume prepayment in the prior Congress, while more contemporary studies assume a spot exchange. Recent empirical work, however, suggests another possibility: Interest groups may buy current influence with money spent over multiple election cycles. Stratmann (1995) shows that contributions from both the current and the previous election cycles influence floor votes on agricultural price supports. Snyder (1990, 1992) presents evidence that interest groups make long-term investments in politicians. Previous studies, by arbitrarily imposing a payment structure, may have produced erroneous estimates of the influence of PAC money on legislative votes.

We examine roll call votes on gun control in the U.S. House of Representatives during the 1980s to determine whether PACs make spot market purchases, prepay for votes in the prior election cycle, or make long-term investments. We investigate whether PAC contributions have a durable, long-term investment effect, as Snyder (1992) suggests. We consider two types of evidence concerning the campaign contribution contract. First, we examine the pattern of expenditures by the two dominant interest groups concerned with gun control: the National Rifle Association (NRA) and Hand Gun Control (HGC). Their expenditures provide strong evidence against the spot market exchange hypothesis: Both groups made significant expenditures in sessions of Congress in which no floor vote occurred. Nor were these expenditures purchasing services such as committee work in a spot market exchange. Further, 1986 NRA expenditures are significantly higher for first-term representatives, which is consistent with long-term investments but con tradicts a spot exchange (since freshmen have the least influence to sell).

Second, we include expenditures from both the current and the five previous election cycles as separate independent variables in a model of roll call voting. To date, no empirical paper tests whether PAC money from more than two cycles influences roll call voting, and only Stratmann (1995) and Grenzke (1989) use a model that specifies money from more than one cycle in an equation explaining roll call votes. Our probit estimates indicate that NRA expenditures in the three most recent election cycles and HOC expenditures in the two most recent cycles buy votes. The independent effect of expenditures from different cycles suggests that contributions are neither an unqualified one-session prepayment nor a spot exchange. The influence of NRA expenditures two sessions prior to the roll call votes is evidence of a long-term investment.

The literature on the determinants of roll call voting assumes that PAC contributions alter representatives' votes. Interest groups might instead simply provide campaign assistance to their supporters to affect the composition of Congress (Stratmann 1991; Bronars and Lott 1997). If a difficult to quantify variable (and thus omitted from regressions) signals a representative's true position and interest groups make campaign contributions on the basis of this variable, PAC contributions might appear to influence votes. The pattern of contributions and votes on gun control, however, indicates attempts by the NRA and HGC to influence votes, not merely the composition of Congress. In particular, neither group targeted contributions to supporters in greatest electoral danger, and almost 60 representatives switched their position on gun control between 1986 and 1988.

We proceed as follows. Section 2 discusses the advantages of the three different payment plans and the appropriateness of gun control as an issue for distinguishing between these alternatives. Section 3 specifies our econometric model of roll call voting. Section 4 examines evidence from the pattern of PAC contributions. Section 5 presents our econometric results. Section 6 contains stability tests of our model that provide evidence that the NRA and HGC are not primarily buying services other than floor votes. Section 7 considers the issue of reverse causality. Section 8 concludes.

2. Alternative Payment Plans for Buying Votes

Incumbent legislators seek to win reelection. Legislators can use a roll call vote to directly attract constituent support or to secure contributions that might indirectly win votes by funding campaign activities. In a world of imperfect information, legislators must advertise those acts that benefit their constituents. Voting against constituent interests improves reelection prospects if the expected number of votes lost because of this roll call is less than the positive impact of campaigning. Constituency interests determine the representative's reservation price for a roll call vote, and at least some representatives should be willing to trade their votes for a modest contribution. [2]

What type of payment scheme will interest groups employ in making contributions to politicians? Each of the three arrangements has beneficial properties. The simplest exchange involves PACs and representatives transacting in a spot market, trading roll call votes for contributions in the present election cycle. In such a trade, expenditures in the 1997-1998 cycle influence a 1998 vote. Advantages of a spot market exchange stem from potentially high transaction costs in a political market. The lack of court enforcement subjects political bargains to opportunism. A spot market allows a nearly simultaneous exchange, reducing the ability of one party to renege after the other performs.

Alternatively, a single payment for a current House vote could occur in the immediately previous election cycle. Prepayment allows interest groups to influence both the composition of the Congress that casts the roll call vote and the candidates' announced positions (Stratmann 1991). Political deals involving prepayment must be self-enforcing. Reputation effects may suffice to prevent a politician from accepting prepayment and then reneging on the promised vote. [3]

Prepayment allows representatives to access contributions by selling future votes today. The money can be used to deal with immediate challengers or banked in the campaign war chest to deter potential competition. Selling commitments on possible votes in the next session benefits congressmen by smoothing out their contributions. Of course, a PAC will not pay the full value of a vote since the issue may not come to the floor. The payment will not exceed the product of the vote's value if cast, the probability a vote occurs, and the probability the candidate wins the intervening election. Since a floor vote can occur in any future Congress, a PAC will make a contribution each session. With spot exchanges, only interest groups concerned with the issues brought to a vote in a particular Congress make contributions, which consequently may vary enormously between sessions.

The third option entails purchasing a vote with several payments. Long-term investment buys votes occurring in future sessions on the "installment plan." Investments, like a one-session prepayment, also smooth out contributions. Furthermore, PACS may need to spread their payments over several sessions if campaign finance laws set the limit for the maximum contribution at less than the price of a legislator's vote. [4] Long-term contracts further separate payment from delivery, however, and so increase the potential for opportunistic behavior. However, Snyder argues that "since this self-enforcement must be based on something like trust or reputation, it is not obvious that the long-term contract is much more difficult to sustain than the short-term one" (1992, p. 18). Long-term investments are subject to electoral attrition: A representative you contribute to now may be defeated (or retire) before he can reciprocate. [5]

The structure that best characterizes exchanges in the roll call market is an empirical question. Studies typically assume either prepayment or a spot market purchase and implicitly ignore the installment plan alternative. The absence of econometric evidence for the payment arrangement employed effectively renders the choice arbitrary. Specifying the wrong payment scheme may yield an erroneous estimate of the influence of PAC money.

Determining the form of PAC transactions when Congress votes on an issue each session is practically impossible. Today's contribution may pay only for a current floor vote, prepay for later floor votes, or both. Congress votes infrequently on gun control, making this a promising issue on which to distinguish contract structures. The federal government enacted no major gun control legislation between 1968 and 1993, and no floor votes occurred in the House between 1968 and 1986. [6]

The two main interest groups concerned with gun control are single-issue PACs, and this further facilitates our efforts. We do not have to worry about these PAC contributions influencing other votes. An environmental interest group may be buying votes on air pollution control, Superfund, and endangered species protection with one contribution. The NRA, one of the most influential interest groups in the United States, defends a person's ownership rights while HGC supports gun control. [7] We investigate the House instead of the Senate because payment for votes more than one session in the future is more likely with the two-year House term than with the six-year Senate term.

Two previous studies, Stratmaun (1995) and Grenzke (1989), employ contributions from more than one election cycle in a vote equation. Three differences suggest that our paper provides a more convincing test of the payment structure. First, these papers examine issues frequently voted on: farm price supports or quotas for farm commodities (Stratmann) and interest group ratings summarizing many votes (Grenzke). Second, their relevant interest groups desire influence over a wide range of policies, making isolation of the votes purchased with any one payment difficult. Third, both studies specify current and previous contributions as simultaneously determined with current voting. Expectations about future votes may induce contributions, but an unexpected component of a vote can never impact contributions given in the previous session. We consider a simultaneity problem for current but not past contributions. [8]

3. The Model

We examine six roll call votes in the House of Representatives in 1986 and 1988. [9] Five are from the 1986 House attempt to weaken the 1968 gun control law; we distinguish the votes using the numbers assigned by Congressional Quarterly, 64, 65, 66, 68, and 69. The sixth vote, 321 in 1988, is on an amendment offered by Bill McCollum (R-FL) to strike the Brady Bill provisions from an omnibus drug bill. The amendment passed, killing the Brady Bill for the term. Descriptions of each vote and the yea and nay summaries appear in Appendix A.

The following structural equations explain the market for 1986 roll call votes:

NRA86 = F(VOTE, FRESHMAN, TENURE, VOTE%84, CRIME, JUDICIARY, DEM) (1)

HGC86 = F(VOTE, FRESHMAN, TENURE, VOTE% 84, CRIME, JUDICIARY, DEM) (2)

VOTE = F(NRA86, HGC86, NRA84, HGC84, NRA82, HGC82, NRA80, HGC80, NRA78, DEM, %URBAN, INCOME, %OVER65, VIOLENT, %BLACK, CC1985). (3)

The equation of primary concern, Equation 3, allows estimation of the effect of current and previous contributions on gun control votes. We normalize our vote dummy variable so that a value of one always represents a vote in support of gun control. The specification of our money variables deviates from typical practice. Most scholars consider only a link between direct PAC contributions to a candidate and roll call voting. However, PACs often make sizable independent expenditures for or against a candidate. To control for this, we include all three components of PAC money in our NRA and HGC contribution variables. We define expenditures for each group in a cycle as money given directly to a candidate plus independent expenditures by the group on behalf of a candidate minus independent expenditures by the group against a candidate. The NRA's 1986 expenditures for a candidate ranged from -$604 to $56,500, while HGC's 1986 expenditures ranged from -$9,675 to $2,737. All amounts are in 1982-1984 dollars. [10]

The remaining variables in Equation 3 are two types of controls. The first controls for the demographic composition of a representative's district. These include %URBAN, INCOME (median family income), %OVER65, VIOLENT (the violent crime rate), and %BLACK. The second type measures a representative's preferences. These variables include the representative's party affiliation, DEM, and ideology as measured by his Conservative Coalition score from the previous year, CC1985.

Equation 3 contains three potentially endogenous variables: the dependent variable, VOTE, and two right-hand-side variables measuring the current cycle contributions, NRA86 and HGC86. Equations 1 and 2, based on Grier and Munger (1986, 1991, 1993) and Grier, Munger, and Torrent (1990), explain current NRA and HGC expenditures. Equations 1 and 2 contain five independent variables not in Equation 3 that we use to identify the 1986 expenditure variables in an estimation of Equation 3. These include FRESHMAN, TENURE, VOTE%84 (the representative's vote percentage in the previous general election), JUDICIARY, and CRIME (the relevant committee and subcommittee for gun control). Appendix B contains data sources and summary statistics for all the variables.

Treatment of the potential endogeneity of current money is our most significant econometric challenge. Estimation of a two-step probit is one possible approach to the endogeneity problem. [11] Predicted values obtained from regressing NRA86 and HGC86 on all the exogenous variables in the system replace the actual values in Equation 3; this produces asymptotically consistent estimates, although the standard errors must be corrected (Murphy and Topel 1985). Asymptotic results, however, do not necessarily indicate the best course of action in finite samples. Bollen, Guilkey, and Mroz (1995) present Monte Carlo evidence that in realistically sized samples, correcting for a simultaneity bias can create more inaccurate results than retaining the unadjusted variable in the equation. We test for the endogeneity of NRA86 and HGC86 to determine the appropriate specification of the current expenditure variables. Appendix C details these tests.

We treat NRA86 and HGC86 as exogenous on the basis of two types of evidence. First, we ran an ordinary-least-squares (OLS) regression of NRA86 and HGC86 on all the exogenous variables in the system; these estimates are presented in Table 1. The estimated error terms from these regressions, and the actual 1986 expenditure figures were plugged into Equation 3. The estimated error terms from the first step were jointly insignificant in the majority of the probit estimates of Equation 3, suggesting that the 1986 expenditure variables are best specified as exogenous. [12] Second, the exogenous variables in Equations 1 and 2 but not Equation 3 do not provide enough variation to adequately identify the predicted values of NRA86 and HGC86. [13]

4. The Pattern of PAC Contributions on Gun Control

We start our empirical analysis by examining NRA and HGC spending patterns for clues to the type of contract in use. Table 2 presents NRA and HGC expenditures for every election cycle from 1978 to 1988. These figures provide strong evidence against the spot exchange hypothesis. The NRA made expenditures in each of the six election cycles between 1978 and 1988, and HGC did so in the five cycles between 1980 and 1988, while gun control bills reached the floor only in 1986 and 1988. Expenditures prior to the 1986 cycle did not buy contemporaneous floor votes. Further, each group made significant expenditures in these years: HGC spent more money in 1982 and 1984 than in 1986 and 1988, while the NRA spent more in 1982 and 1984 than it did in 1988. NRA expenditures in 1986 were only 14% greater than in 1984.

The 1978 through 1984 expenditures could constitute a spot market exchange if, as Welch (1982) contends, PAC contributions purchase committee work, for example, the effort of members to either kill a bill in committee or report it out for a floor vote. Table 2 presents the percentage of NRA and HGC total expenditures made to Judiciary Committee members. Again the evidence rejects a spot market exchange. Only between 2.7% and 30.6% of each PAC's total expenditures in cycles before the 1986 bill went to committee members. Further, the percentage of money allocated to committee members did not change very much in sessions when gun control votes were taken. The NRA made, on average, 5.95% of its expenditures on committee members in sessions where a floor vote did not occur and 3.85% of its expenditures on committee members when floor votes did occur. HGC averaged 21.1% and 16.3% of its expenditures on committee members when a vote did not and did occur, respectively. Interest groups do not pay significantly more attention to the committee when there is no vote.

Table 2 lists the percentage of total expenditures that the NRA and HGC gave directly to House candidates, which undermines the campaign contributions limit motive. Spending limits do not affect independent expenditures on behalf of or against a candidate. A substantial amount of money was not subject to restrictions. HGC gave as little as 8.7% of their total expenditures directly to candidates, while the NRA gave as little as 73.5% of their total expenditures directly to candidates. The ability to make large independent expenditures may eliminate the need for PACs to stretch direct contributions over many years. Table 2 provides some support for a one-session prepayment, which can nicely explain the sharp increase in NRA contributions in the 1982 election cycle. The Reagan assassination attempt in March 1981 increased the expected probability of a floor vote in the next session, increasing the value of commitments.

The OLS regressions of 1986 NRA and HGC expenditures on all the exogenous variables in the model presented in Table 1 and used in our endogeneity tests in section 3 shed additional light on our inquiry. Previous work establishes the importance of tenure and the committee system in the organization of Congress: Senior representatives and members of the committees with jurisdiction over the legislation in question have greater influence. [14] First-term legislators have the least influence, with little more to offer a PAC than their roll call Vote. Since a spot market requires PACs to purchase a representative's influence with one payment, the spot market hypothesis predicts that TENURE and the two committee variables, JUDICIARY and CRIME, should increase while FRESHMAN decrease current expenditures. [15] The long-term-investment hypothesis provides contradictory implications. Long-serving members have greater influence but have received more previous contributions. Similarly, PACs could have already invested in the additional services that committee members offer. However, PACs have not yet invested in new members. The long-term hypothesis predicts that FRESHMAN positively influences current expenditures and predicts a zero coefficient on TENURE, JUDICIARY, and CRIME.

The results in columns 2 and 3 of Table 1 strongly contradict the spot market hypothesis but are consistent with long-term investments, at least by the NRA. FRESHMAN is positive and significant for NRA86, and the coefficient estimate is large. TENURE and JUDICIARY are insignificant in both the NRA86 and the HGC86 equation. CRIME, which is insignificant in HGC86, moves in an unexpected direction in NRA86 and has a t-statistic of 1.3 in this equation. The coefficient for FRESHMAN is negative but insignificant for HGC86.

The long-term-investment hypothesis further predicts that past contributions by the same PAC should increase current contributions while past contributions by the other PAC should decrease current contributions. Based on these considerations, the NRA should continue to invest in the same politicians in 1986 as in 1984 or 1982, while HGC should avoid contributions to representatives who have already sold their votes to the NRA. Previous NRA expenditures all positively influence NRA86, and NRA84, NRA82, and NRA80 are each significant at the .05 level or better. The coefficients on HGC84, HGC82, and HGC80 are insignificant in the NRA86 regression. In contrast, only HGC84 is significant at the .10 level in the estimate of HGC86, even though all previous HGC money variables have the expected sign. NRA84 and NRA80 are negative and significant determinants of HGC86.

The pattern of PAC expenditures rejects a spot market exchange of votes and contributions. Several features of long-term investments are evident (as in Snyder 1992), and some support exists for the one-period-prepayment hypothesis. The stronger evidence for long-term investments by the NRA than HGC is unsurprising since HGC is a relative newcomer and the NRA typically outspent HGC by a ratio of 10 to 1. HGC perhaps is less willing to use its limited funds to make long-term investments.

5. Which Contributions Influence Roll Call Votes?

We now analyze House roll call votes on gun control in the 1980s. [16] We first test whether money has a durable influence in the full specification of our model. Since this specification allows money from five election cycles to influence roll call votes, it may be too inclusive; the insignificant variables may produce inefficient estimates. We therefore reestimate the model with insignificant lagged NRA and HGC expenditure variables omitted.

Table 3 presents the estimates of Equation 3 for the five 1986 roll call votes when possible lagged PAC influence extends five election cycles. Expenditures from more than one cycle matter, especially for the NRA; this casts doubt on either a pure spot exchange or a pure prepayment for votes. The coefficients for the 1986 expenditure variables (NRA86 and HGC86) in the five estimates of Equation 3 always have the predicted sign, and nine of the 10 are significant at the .1 level or better in a one-tailed test. [17] The results provide weaker evidence of influence by money spent in the previous Congress. Again all 10 1984 expenditure coefficients (NRA84 and HGC84) have the correct sign, but only three are significant at the .1 level in a one-tailed test. The NRA, but not HGC, appears to make long-term investments: 13 of the 15 coefficients for NRA82, NRA80, and NRA78 have the correct sign, with five significant at the .10 level or better. Three of the significant coefficients are on 1982 expenditures. Eight of HGC's 10 estimates of HGC82 and HGC80 have the correct sign, but only one of these is significant at the .10 level.

The model overall has good explanatory power, correctly predicting between 81% and 92% of the representative's votes with an average prediction success rate of 87.5%. The simple majority position ranged from 54.5% to 68.1% with an average of 61.5%. The control variables, when significant, had a similar effect on the different votes. Democrats consistently favored gun control. INCOME, %URBAN, and %OVER65 were consistently significant (at the .05 level) determinants of support for gun control. The Conservative Coalition score was always negative and significant at the .05 level. A district's violent crime rate and percentage of black residents were generally insignificant determinants of voting on gun control.

We next estimate a parsimonious specification of the model omitting the NRA80, NRA78, HGC82, and HGC80 expenditure variables, which are jointly insignificant at conventional levels. Table 4 presents these results, which are similar to Table 3 except for somewhat higher t-statistics for the remaining money variables. [18] The NRA influences floor voting with current money and funds spent two sessions ago. All 10 of the estimates of NRA86 and NRA82 have the predicted sign, and nine are significant at the .05 level. NRA money from the previous election cycle (NRA84) always has the predicted sign but is significant at the .10 level only once. All the HGC86 and HGC84 coefficients in Table 4 have the predicted sign, and four out of five estimates of each variable are significant at the .10 level or better. [19] Removing the irrelevant variables did not change the explanatory power of the equations. The likelihood ratio index ranged from .42 to .69, and the percentage of votes predicted correctly ranged from 82% to 92%. The control variable results remained similar. [20]

Table 5 presents estimates of our model for the 1988 Brady Bill vote. We again treat current money as exogenous. Table 5 reports the full model and a parsimonious specification excluding insignificant lagged expenditures. [21] In the full model, the 1982-1988 NRA expenditure variables are significant at the .05 level, but the 1984 variable has the wrong sign. Current HGC expenditures are significant at the .05 level, while the one period prepayment spending is very insignificant (a t-statistic of .03). The insignificant HGC estimates and the positive NRA84 coefficient may result from a problem much of the current literature faces. Money given before the 1988 cycle may have purchased influence over both the 1986 votes and the 1988 vote. Without a way to determine precisely which votes the earlier money influenced, puzzling results may arise, as in this case. We therefore place more weight on analysis of the 1986 votes than the 1988 vote.

6. Stability Tests

The model in section 5 implicitly assumes an equal marginal effect of pro-gun money for all legislators. A legislator possessing institutional assets like committee membership can offer more than just roll call votes, and PACs will pay for these additional services. If PACs buy more than just votes, the influence of the marginal dollar on floor voting will vary inversely with the value of other services a congressman can offer. For example, a dollar paid by the NRA to a member of the Judiciary Committee may buy committee work and a small amount of roll call influence. A representative not on the Judiciary Committee has to sell more roll call influence to earn an NRA dollar. A dollar may buy less voting influence for a committee member than it does from an otherwise identical congressman. [22]

We test the stability of the effect of PAC contributions by adding three sets of interaction terms to the estimates of the 1986 votes reported in Table 4. The interaction terms allow the influence of PAC money to vary with two measures of the value of non--roll call services a congressman can offer: membership on the Judiciary Committee and seniority. Three dummy variables create the interaction terms. JUDICIARY indicates membership on the Judiciary Committee. JUNIOR equals one if a member has less than four terms of tenure and zero otherwise. SENIOR equals one if a member has at least 10 terms of experience and zero otherwise. We create interaction terms by multiplying each dummy variable by the NRA and HGC expenditure variables in Table 4. The parsimonious model is then successively reestimated using the JUDICIARY, JUNIOR, and SENIOR interaction terms.

Table 6 presents the joint levels of significance for the interaction terms. The sale of other services by representatives appears not to alter the relationship between PAC expenditures and roll call voting, as only one of the 15 possible sets of interaction terms achieved significance at the .1 level. In sum, a dollar of expenditure buys the same amount of influence on gun control roll calls whether a representative serves on the Judiciary Committee and regardless of his seniority. We conclude that at least in 1986 the NRA and HGC bought primarily floor votes, not committee work or other influence.

7. Do PACs Buy Roil Call Votes or Influence Elections?

Our analysis assumes that interest groups make contributions to influence the votes of representatives. As such, our inquiry fits into the extensive literature investigating roll call votes. However, as Bronars and Lott (1997) argue, the correlation between contributions and votes does not necessarily demonstrate causation. Alternatively, interest groups may identify representatives who share their position on the question and make contributions to help their supporters win office. Several aspects of the 1980s gun control votes indicate that interest groups are likely influencing votes.

We first note two points about our analysis so far. No roll call vote on gun control occurred in the House between 1968 and 1986. Thus, for the vast majority of members in 1986, the NRA and HGC had no objective measure of preferences on gun control on which to base their contribution decisions, and our vote Equation 3 contains a measure of representative ideology, the previous year's Conservative Coalition score. Our PAC money variables are still significant even with inclusion of this measure of general preferences. Interest groups can attempt to discern representatives' preferences through informal discussions, which our ideology and constituency variables omit, but talk is cheap. Indeed, both sides made mistakes in their 1984 contributions by giving money to representatives who voted against their position in 1986. The NRA made seven such mistakes, while HGC made five mistakes. [23]

The pattern of contributions suggests that PACs are not primarily interested in using PAC money to influence who gets elected. If they were, a PAC should target contributions to swing districts. Freshman representatives in the 1985-1986 session either defeated an incumbent (who was in electoral danger) or won open-seat elections, which are more competitive than average. Yet freshmen in the 1985-1986 Congress account for only 16% of the NRA and 6% of the HGC contributions during the 1984 election cycle. Similarly freshmen in the 1987-1988 Congress account only for 19% of the NRA's and 2% of HGC's 1986 election cycle contributions to members. The number of representatives receiving PAC money suggests that PACs were not exclusively trying to change the composition of Congress. In 1984, 187 (98) members of the 1985-1986 Congress received NRA (HGC) contributions. Clearly, over half the seats were not at electoral risk in 1984. Finally, PACs concerned primarily with influencing elections should contribute to suppo rters who face close elections. The NRA in 1984 made contributions to 180 of the 295 representatives who voted to weaken gun control in 1986. The average 1984 vote share of members who received 1984 NRA contributions was 69.59%, while the average vote percentage of supporters who did not receive 1984 NRA contributions was 71.83%. Similarly, the average 1984 vote percentages of the 93 representatives who voted with HGC in 1986 and received 1984 HGC money was 70.63%, while the 45 supporters who did not receive 1984 HGC money had an average vote percentage of 71.76%. In neither cases are these differences statistically significant. The mean vote share for members receiving NRA contributions is more than one standard deviation above the majority vote share necessary to win the election, so most recipients were not at electoral risk.

The pro-gun rights position lost considerable support between 1986 and 1988. Vote changes constitute a serious challenge to the alternative hypothesis that preferences determine representatives' votes and PAC contributions merely assist supporters. Turnover in membership does not explain the change in support for gun control. Of the 55 freshmen in the 1987-1988 Congress, 43 voted the way their predecessors did on the final vote (no. 69) in 1986. In fact, the NRA position gained a net of four votes because of turnover. Of the 375 representatives who voted in both 1986 and 1988, 59 (15.7%) switched from the NRA to the HGC position, while no representatives switched to the NRA position. [24]

NRA and HGC contributions may explain the vote switches between 1986 and 1988. First we note that HGC made contributions to 10 members who voted against their position in 1986. If preferences solely determine roll call votes, this money was totally wasted. [25] More important, the time pattern of contributions is quite suggestive. Table 7 presents the average contributions by the NRA and HGC to representatives who voted with their positions in both 1986 and 1988 and to the vote switchers. NRA contributions to these two groups in the 1984 election cycle were approximately equal, $1,456.05 for loyalists versus $1,137.44 for switchers. However, contributions to the vote switchers fell dramatically in the 1986 and 1988 cycles, to approximately one-fifth of their 1984 levels. Loyalists received increased contributions. Note that even if 1986 contributions and the 1986 vote were simultaneously determined, the NRA could not have been punishing these representatives, as they voted with the NRA in 1986. Meanwhile, HG C contributions to the vote switchers increased from $34.75 in 1984 to $86.44 in 1988. The ratio of HGC money to NRA money for vote switchers rose from .031 in 1984 to .351 in 1988. If preferences determine voting, given that these representatives revealed their opposition to gun control with their 1986 votes, the 1988 HGC contributions were irrational. The sharp decline in NRA funding between 1984 and 1988 and the increase in HGC funding in 1988 to vote switchers is consistent with the hypothesis that money influenced the roll call votes.

The simultaneous change in votes and PAC contributions is worthy of examination in its own right although beyond the scope of this paper. McGarrity and Sutter (1999) analyze vote switches on the Brady Bill between 1988 and its passage in 1993. Again switches overwhelmingly favor the gun control position; the decrease in NRA contributions is the only statistically significant determinant of the switches. Did these representatives stop taking NRA money to vote for gun control, or did NRA neglect lead to vote switching? Sabato (1984, p. 129) notes the importance of independent expenditures against candidates made by ideological PACs and cites as an example a 1982 negative television ad by HGC against pro-NRA representatives. Independent expenditures by HGC against supporters of the Volkmer Amendment, however, do not seem to explain the 1988 switches. HGC made expenditures against only one of the 59 vote switchers in any of the 1984, 1986, or 1988 cycles. [26]

8. Conclusions

We offer the first empirical test of whether PAC expenditures exert a long-term influence over roll call voting. Both interest groups buy influence that spans more than one Congress. The NRA's contributions buy influence that extends three Congresses, while HGC purchases influence with current dollars along with expenditures made the previous Congress. The difference in the longevity of influence among expenditures made by the two interest groups may be due to their competitive imbalance. The NRA usually spends 10 times more than HGC in House races. HGC may not have the luxury of making long-term investments to candidates who may not be around in future Congresses.

Our results imply that the spot exchange, the most common specification assumed in the more recent literature, is by itself not totally appropriate. As mentioned previously, money given in other sessions affect the vote. Leaving out significant money variables can bias the estimates. Future tests of the link between money and roll call voting should not arbitrarily assume a payment structure. Models should examine the possible long-term investment properties of PAC contributions.

(*.) Department of Economics and Finance, University of Central Arkansas, Conway, AR 72035, USA; E-mail joem@mail.uca.edu; corresponding author.

(+.) Department of Economics, University of Oklahoma, Norman, OK 73019, USA.

Walter Block, Andy Glenn, Kirsta Glenn, William Greene, Kevin Grier, Bernard Grofman, Laura Junor, Nolan McCarty, Marc Poitras, Robert Tollison, two anonymous referees, and seminar participants at UCA, Oklahoma, and the 1997 Public Choice meetings provided helpful comments. Sutter thanks the Social Philosophy and Policy Center at Bowling Green State for financial support.

Received November 1998; accepted August 1999.

(1.) Silberman and Durden (1976); Kau, Keenan, and Rubin (1982); Wilhite (1988); Withite and Theilmann (1987); Langbein and Lotwis (1990); and Stratmann (1991, 1995) find evidence that PAC contributions buy congressional roll call votes. Contrastingly, Chappell (1981, 1982), Wright (1985), Welch (1982), and Grenzke (1989) reach the opposite conclusion.

(2.) See Mueller (1989, PP. 206-14), Stratmann (1991), and Denzau and Munger (1986).

(3.) Informal campaign contribution contracts require punishment of violations by politicians and interest groups. For evidence that the willingness of legislators to punish violations is low, see McCarty and Rothenberg (1996).

(4.) Although contribution limits force installment purchases, if the preference of politicians and PACs were for lump sum payments, the question then becomes why spending limits were adopted.

(5.) Note that a durable investment differs from repeated spot market or prepayment purchases. A long-term relationship does not imply that earlier contributions directly influence today's vote.

(6.) The 1968 law, passed in the aftermath of the assassinations of Robert F. Kennedy and Dr. Martin Luther King, Jr., banned the mail order and interstate shipment of firearms and stiffened licensing requirements for dealers. The 1993 Brady Bill mandated a five-business-day waiting period on firearm purchases, required notification of police of purchases of multiple handguns, and increased licensing fees for gun dealers.

(7.) No PAC will be exclusively concerned with only one issue, if only because of the incentive to use already organized groups for new causes. The Congressional Quarterly Almanac for 1981-1988 mentions the NRA in stories on only a few other matters: banning "Saturday Night Specials" (1981); funding for the Bureau of Alcohol, Tobacco and Firearms (1982); an Alaska sport hunting ban (1983); armor-piercing "cop killer" bullets (1984, 1985); a plastic gun ban (1988); and a proposal to extend RICO legislation to the sale of fireams to felons (1988). The NRA and HGC are probably as close approximations as exist to single-issue groups.

(8.) To illustrate, consider the following system of equations:

[VOTE.sub.t] = [a.sub.0] + [a.sub.1].[PAC.sub.t-1] + [e.sub.1] (A)

[PAC.sub.t-1] = [b.sub.0] + [b.sub.1].P([Vote.sub.t]) + [e.sub.2]. (B)

[VOTE.sub.t] is the roll call vote in the current election cycle, [PAC.sub.t-1] is the contribution given the previous cycle, and P([VOTE.sub.t]) is the predicted vote. Typically, a simultaneity problem occurs when an increase in [e.sub.1] leads to an increase in [VOTE.sub.t], the dependent variable in Equation A. If [VOTE.sub.t] were the independent variable in Equation B, as Stratmann and Grenzke assume, then this increase in [VOTE.sub.t] would, if [b.sub.1] were positive, increase the dependent variable in Equation B, [PAC.sub.t-1]. [PAC.sub.t-1] is also in Equation A and would be increasing with [e.sub.t]. Since [PAC.sub.t-1] and [e.sub.1] are not independent, a simultaneity correction would be called for. The independent variable in Equation B, however, is the predicted vote, P([VOTE.sub.t]), not the actual vote.

(9.) The previous empirical analysis of gun control votes includes Langbein and Lotwis (1990) and Langbein (1993). Both papers consider the 1986 roll call votes of 143 House members. They include NRA and HGC current contributions and dummy variables indicating whether the member received money from either interest group in the 1984 cycle. The 1990 paper, which attempted to explain the 1986 roll call votes, finds that both NRA variables and current HGC spending are significant at the .1 level or better. 1984 HGC spending, while not significant, did have the predicted sign. The 1993 paper attempted to explain which members switched votes as the roll call votes progressed.

(10.) Use of direct contributions to a candidate, the measure of PAC money typically used in the roll call voting literature, does not affect any of our results.

(11.) See Greene (1997).

(12.) See Greene (1997), Maddala (1992), and Bollen, Guilkey, and Mroz (1995).

(13.) For a discussion of these points, see Bollen, Guilkey, and Mroz (1995).

(14.) For examples, see Shepsle and Weingast (1987), Weingast and Marshall (1988), Weingast and Moran (1983), and Roberts (1990).

(15.) The prepayment hypothesis yields similar predictions even though 1986 contributions buy influence in 1987-1988.

(16.) Offsetting expenditures in equilibrium offer another explanation for a lack of influence of PAC contributions on roll call votes. However, on gun control the NRA and HGC only rarely make expenditures to the same representative. In the 1986 cycle, 215 representatives received money from one of these two PACs. Only 11 members had expenditures from both groups, and in nine of these cases HGC made expenditures against a representative who received support from the NRA. Thus, in only two of 215 cases could NRA and HGC expenditures possibly offset each other.

(17.) In the full specification model, when predicted values are used instead of the actual values, current NRA expenditures are significant at the .1 level twice, and current HGC expenditures move in the unexpected direction and are significant at the .05 level twice. The remaining current expenditure estimates are insignificant at the .1 level. The 1984 expenditure variables always move in the expected direction and are significant at the .1 level or better in eight out of 10 estimates. The results on older expenditures are generally comparable to those reported in Table 3.

(18.) When predicted values are used for the 1986 expenditure variables in the parsimonious equation, the model performs poorly. The predicted value of HGC86$ is significant at the .1 level in two of the five estimates. The predicted value for NRA86$ is never significant. Only one of the 15 estimates of the three remaining money variables is significant at the .1 level. The control variables also do not perform quite as well as the results reported in Table 4. CC1985 and INCOME are significant at the .05 level in all five estimates. %OVER65 is significant at the .05 level three times and the .1 level twice. %URBAN is significant at the .05 level once and the .1 level once. %BLACK is significant at the .1 level once. VCRIME is always insignificant.

(19.) The magnitude of the coefficients on the HGC expenditure variables may trouble some readers. For example, the coefficient on HGC86 in the Vote 68 regression in Table 4 (.0019) suggests that a $1,000 contribution would increase a representative's probability of voting for gun control from 0 to .97. Seemingly legislation can be purchased very cheaply. However, the average contribution by HGC in 1986 was $84, and only five (out of 83) of HGC's contributions in this cycle exceeded $1,000. We cannot simply infer from our results how many votes HGC could have bought with much larger expenditures than they actually made.

(20.) Inclusion of current and lagged PAC contributions in the roll call regressions subjects our results to multicollinearity problems. If an interest group continually invests in the same representatives, the influence of any one cycle's expenditure may be lost in a regression containing expenditures from other cycles. Examination of NRA and HGC expenditure patterns diminishes fears of multicollinearity because the PACs do not give to the same representatives every cycle. The NRA gave money at least once to 120 of the 193 members in our sample who served in all five Congresses from 1978 through 1986. Only 25 of these 120 received money from the NRA in all five cycles (20.8%); 18 more received money four times. HOC contributed to 61 of the 193 members at least once. Only 11 received money in all four cycles in which HOC made expenditures (18.0%), and 24 more received HGC money three times.

(21.) When predicted values are used instead of the actual values in the two 1988 vote model specifications, none of the predicted values is significant at conventional levels. NRA86$ and NRA82$ are significant at the .05 level and move in the expected direction in both estimates. NRA84$ is significant at the .1 level in both equations; however, like the results reported in Table 4, it has an unexpected sign. HGC84$ is significant in both equations at the .1 level or better. NRA78$ is also significant at the .1 level in the full equation. The remaining money variables are insignificant. The control variables estimates are roughly equivalent to those reported in Table 4.

(22.) See Grier and Munger (1991, 1993).

(23.) The NRA made contributions to 187 representatives in the 1984 cycle, while HGC made 98 total contributions. While the error percentages for each group's contributions are low, 148 representatives who voted in 1986 did not receive contributions from either group. Undoubtedly, some of these individuals comprise errors by the PACs who were unable to identify their supporters.

(24.) Preferences could explain these votes if the Brady Bill were closer to the vote switchers' ideal points. However, the Brady Bill provisions were more severe than those that the Volkmer amendment repealed in 1986. With single-peaked preferences, any representative who preferred the Volkmer amendment to the 1986 status quo would prefer Volkmer to Brady. With only two years between the votes, constituent demographics and preferences are unlikely to have changed significantly.

(25.) The NRA, however, did not make any 1988 cycle contributions to representatives who voted against their position in 1986.

(26.) Police officers, specifically the Law Enforcement Steering Committee, became an important lobbying force during this period; Langbein and Lotwis (1990) emphasize police lobbying in their study. They use a measure of lobbying contacts and letters received for a sample of representatives that cannot be applied to our broader sample.

References

Bollen, Kenneth, David Guilkey, and Thomas Mroz. 1995. Binary outcomes and endogenous explanatory variables: Tests and solutions with an application to the demand for contraceptive use in Tunisia. Demography 32(l):111-31.

Bronars, Stephen G., and John R. Lott. 1997. Do campaign donations alter how a politician Votes? Or, do donors support candidates who value the same things that they do? Journal of Law and Economics 40(2):3l7-50.

Chappell, Henry. 1981. Campaign contributions and voting on the Cargo Preference Bill. Public Choice 36:301-12.

Chappell, Henry. 1982. Campaign contributions, and congressional voting: A simultaneous Probit-Tobit model. Review of Economics and Statistics 62:77-83.

Congressional Directory. 99-100th Congress. Washington, DC: Government Printing Office.

Congressional Quarterly Incorporated. Various issues. Congressional Quarterly Almanac. Washington, DC.

Denzau, Arthur, and Michael Munger. 1986. Legislators and interest groups: How unorganized interests get represented. American Political Science Review 80(l):89-106.

Federal Election Commission. 1995. FEC computer listing from a file search in D Index, February 21.

Greene, William. 1997. Econometric analysis. 3rd edition. Upper Saddle River, NJ: Prentice Hall.

Grenzke, Janet. 1989. PACs and the congressional supermarket: The currency is complex. American Journal of Political Science 34:2-24.

Grier, Kevin, and Michael Munger. 1986. The impact of legislator attributes on interest group campaign contributions. Journal of Labor Research 7(4):349-61.

Grier, Kevin, and Michael Munger. 1991. Committee assignments, constituent preferences, and campaign contributions to House incumbents. Economic Inquiry 29:24-43.

Grier, Kevin, and Michael Munger. 1993. Comparing interest group PAC contributions to House and Senate incumbents. Journal of Politics 55(3):615-43.

Grier, Kevin, Michael Munger, and Gary Torrent. 1990. Allocation patterns of PAC monies: The U.S. Senate. Public Choice 67(2): 111-28.

Kau, James, Donald Keenan, and Paul Rubin. 1982. A general equilibrium model of congressional voting. Quarterly Journal of Economics 97:271-93.

Langbein, Laura. 1993. PACs, lobbies, and political conflict: The case of gun control. Public Choice 75:254-71.

Langbein, Laura, and Mark Lotwis. 1990. The political efficacy of lobbying and money: Gun control in the U.S. House, 1986. Legislative Studies Quarterly 15:413-40.

Maddala, G. S. 1992. Introduction to econometrics. 2nd edition. New York: Macmillan.

McCarty, Nolan, and Lawrence S. Rothenberg. 1996. Commitment and the campaign contribution contract. American Journal of Political Science 40(4):872-904.

McGarrity, Joseph P., and Daniel Sutter. 1999. The Brady Bill: Turning legislative defeat into victory. Working paper.

Mueller, Dennis. 1989. Public Choice II. Cambridge: Cambridge University Press.

Murphy, Kevin, and Richard Topel. 1985. Estimation and inference in two step econometric models. Journal of Business and Economic Statistics 3(4):370-9.

Roberts, Brian. 1990. A dead senator tells no lies: Seniority and the distribution of federal benefits. American Journal of Political Science 34(1):31-58.

Sabato, Larry J. 1984. PAC Power: Inside the world of political action committees. New York: Norton.

Shepsle, Kenneth, and Barry Weingast. 1987. The institutional foundations of committee power. American Political Science Review 81(1):85-104.

Silberman, Jonathan, and Garey Durden. 1976. Determining legislative preferences on the minimum wage: An economic approach. Journal of Political Economy 84:317-29.

Snyder, James M., Jr. 1990. Campaign contributions as investments: The U.S. House of Representatives, 1980-1986. Journal of Political Economy 98:1195-227.

Snyder, James M., Jr. 1992. Long-term investing in politicians; Or, give early, give often. Journal of Law and Economics 35:15-43.

Stratmann, Thomas. 1991. What do campaign contributions buy? Deciphering causal effects of money and votes. Southern Economic Journal 57:606-20.

Stratmann, Thomas. 1995. Campaign contributions and congressional voting: Does the timing of contributions matter? Review of Economics and Statistics 77:127-36.

Weingast, Barry, and William Marshall. 1988. The industrial organization of congress; Or, why legislatures, like firms, are not organized as markets. Journal of Political Economy 96(1):132-63.

Weingast, Barry, and Mark J. Moran. 1983. Bureaucratic discretion or congressional control? Regulatory policymaking by the Federal Trade Commission. Journal of Political Economy 91(5):765-800.

Welch, William. 1982. Campaign contributions and legislative voting: Milk money and dairy price supports. Western Political Quarterly 35:478-95.

Wilhite, Allen. 1988. Union political action committee contributions and legislative voting. Journal of Labor Research 9:79-90.

Wilhite, Allen, and John Theilmann. 1987. Labor PAC contributions and labor legislation: A simultaneous Logit approach. Public Choice 53:267-76.

Wright, John R. 1985. PACs, contributions, and roll calls: An organizational perspective. American Political Science Review 79:400-14.

Appendix A: Roll Call Vote Descriptions and Record of the Votes

1986 Votes on Gun Control

Vote 64: Amendment by Hughes (NJ) to continue prohibition of silencers and interstate handgun sales, disallow interstate transportation of handguns, and retain existing firearm transaction record keeping requirements. A yes vote was for gun control. Vote: Yes 177, No 256.

Vote 65: Amendment by Hughes (NJ) to prohibit interstate transportation of handguns. A yes vote was for gun control. Vote: Yes 178, No 255.

Vote 66: Amendment by Hughes (NJ) to continue prohibition of interstate handgun sales. A yes vote was for gun control. Vote: Yes 197, No 236.

Vote 68: A substitute amendment by Volkmer (MO) permitting interstate sales of rifles and shotguns and interstate transportation of firearms, relaxing record-keeping requirements for firearm transactions, and limiting federal agents to one unannounced visit per year to a gun dealership. A no vote was for gun control. Vote: Yes 289, No 144.

Vote 69: A bill to revise the 1968 Gun Control Act along the lines of the Volkmer amendment. A no vote was for gun control. Vote: Yes 295, No 138.

1988 Gun Control Vote

Vote 321: Amendment by McCollum (R-FL) to strike provisions requiring a seven-day waiting period for handgun purchases and requiring the Justice Department submit a proposal to Congress for a system of background checks to determine whether potential customers were eligible to buy a handgun from HR 5210, an omnibus drug bill. A no vote was for gun control. Vote: Yes 231, No 198.

In the vote tabulations, "paired for" counts as a yes vote, and "paired against" and not voting count as no votes. The votes were transformed so that a one indicated support for gun control. Source: Congressional Quarterly Almanac, 1986, 1988.

Appendix B: Data Sources and Summary Statistics

PAC data were obtained from the Federal Election Commission's "Computer Index of Campaign Activity." Tenure, which measures terms of service, was found in the appropriate Congressional Directory. Violent Crime, which measures violent crimes per 100 people, was derived from county-level information in the County City Data Book % URBAN, MEDIAN FAMILY INCOME, and % BLACK were obtained from Congressional Districts in the 1980s. The remaining variables were found in the appropriate Congressional Quarterly Almanac. Tenure measures the terms that a congressman has served. Vote%84 (vote%86) is the percentage of the vote a member received in the 1984 (1986) election.

Appendix C: Is Current Spending Exogenous?

To answer this question, we first regress NRA86 and HGC86 on all the exogenous variables. We save the two estimated error terms and plug them, along with the actual 1986 expenditure variables, into Equation 3. We then use a probit estimate of Equation 3 and perform a Wald test to determine the joint level of significance of the estimated error terms. If the Wald statistic is near zero, we would be unable to reject the null hypothesis of exogeneity of 1986 expenditures, and the use of the actual values, not the predicted values, provides a better specification of Equation 3.

Table 1 provides the reduced-form estimates of NRA86 and HGC86 for the two different specifications of Equation 3 used in this paper. The first two columns present an estimate of the reduced form of the full model, while the final two columns estimate the reduced form of a more parsimonious specification omitting some expenditure variables (NRA80, NRA78, HGC82, and HGC80). The levels of significance for the Wald statistics obtained when the error terms from the full regressions estimated in Table 1 are added to the full specification of Equation 3 are listed in the order in which the votes occurred: .20, .68, .41, .99, and .99. When the error terms from the parsimonious specifications reported in Table 1 are added to the parsimonious specification of Equation 3, the levels of significance for the Wald statistics are .39, .41, .54, .71, and .85. The Wald Statistics are generated in a similar manner with the 1988 vote equation. The joint significance levels in the 1988 equations are .40 (full specification) an d .48 (parsimonious specification).

As just mentioned, the two error terms added to the 1986 vote equation are jointly significant at the .05 level in two of the 10 probit estimates. Both incidences of significant sets of error terms occur in the specification that includes all the expenditure variables. In the parsimonious specification, the error terms are jointly insignificant at conventional levels and suggest that the 1986 expenditure variables should be considered exogenous. The mixed results in the full model imply that additional considerations should be weighed.

The reduced-form estimates of the model when all expenditure variables are considered suggests that the 1986 expenditure variables should also be considered exogenous. The variables in Equations 1 and 2 but not in Equation 3 do not provide enough independent variation to adequately identify the predicted values. In the NRA86 equation, only FRESHMAN is significant at the .05 level; the remaining four variables are not jointly significant at conventional levels. A Wald statistic of 3.9 at 4 degrees of freedom generated from the joint test has a confidence level of .59. The HGC86 equation fares even worse. None of the five variables exclusive to Equations 1 and 2 is independently significant at the .1 level or jointly significant. A Wald statistic of 2.9 at 5 degrees of freedom generated from the joint test has a confidence level of .29.

The explanatory power of the reduced-form estimate of the full specification has respectable [R.sup.2] of .48 and .50. However, the most significant variable in both these estimates is NRA80 with t-statistics of 13.1 and 16.2 in the two estimates. When this variable, along with some other earlier expenditure variables, are removed from the reduced-form equation in the second specification of the model, the [R.sup.2] of the two reduced-form estimates becomes very low, .26 and .17. Again we see that the variables contained in Equations 1 and 2 but not Equation 3 explain little of the variation in NRA86 and HGC86.

In the two estimates where we rejected the null hypothesis of exogeneity, the estimated error term from the NRA reduced-form equation had an insignificant r-statistic, while the estimated error term from the HGC equation was significant at the .05 level. Since the error term from the HGC equation drives the endogeneity test result and the HGC equation is not identified adequately, we do not replace the actual 1986 expenditure values with predicted values in our estimates of Equation 3.