reelected and thus provide the group an audience on future issues. Again, the problem here is that
the clear patterns of PAC allocations contradict this view. PACs give overwhelmingly to
legislators who are least in danger of defeat – incumbent members from secure districts, even
members who have no challengers and already have substantial war chests.
Nevertheless, our statistical models include a control for an issue specific measure of how
friendly legislators are to the health care industry.
Including this type of control variable
typically eliminates bivariate relationships between campaign contributions and votes (e.g,
Wilkerson and Carrell 1999; Wright 1998). One would expect it to have the same effect on any
relationship between contributions and lobbying, if that relationship was simply due to a tendency
of PACs to give to their friends expecting nothing in return. Thus, if we find a relationship
between contributions and lobbying contacts in our statistical model, it would provide reasonably
sound evidence that campaign spending helps interest groups gain access to senators.
To refresh, the dependent variable that we analyze is the count of lobbying contacts
between group i and senator j reported by a representative of group i. Thus, our dataset contains
one observation for each senator-group dyad. We employ a negative binomial regression
because the number of lobbying contacts is an overdispered count variable. Given the structure
of our data, it is possible our observations are not truly independent because we have multiple
observations for each senator and for each interest group. To address this potential problem we
estimated our standard errors two ways: with observations clustered by interest group and
clustered by senator. We report the results with standard errors clustered by interest group, but
our inferences are not affected by this choice.
There were 19 interest groups in our sample that clearly favored the bill and 6 that
opposed the bill. Analyzing each set of organizations separately allows us more easily to test our
legislative subsidy hypotheses and our campaign money hypotheses. For example, our legislative
subsidy hypothesis holds that lobbyists will seek contacts with legislators who are likely to be
12
Clearly the better solution is to estimate a simultaneous set of equations for money and access. We
think we have identified good instruments to identify such a system, and we are currently collecting the
necessary data.
17
Convention