Presidential Capital and the Supreme Court Confirmation Process
Because our dependent variable is a discrete measure we cannot use traditional linear regression to model this phenomenon. As Long points out, “The use of LRM models for count outcomes can result in inefficient, inconsistent, and biased estimates” (1997, 217). These are not just statistical problems, however. Rather, King (1988, 846) argues that these shortfalls often result in substantive problems where coefficients are the wrong size and the wrong sign. While many reason- able alternatives to OLS regression have been suggested for models with discrete independent variables, King’s (1989) generalized event count model (GEC) allows researchers to model discrete events when unknown dispersion exists in the dependent variable.17 Additionally, King (1988, 859) points out that his tech- nique allows researchers to analyze very small samples (even with an N as small as 10). Thus, his GEC model is appropriate for our purposes.
The model includes three independent variables to test our hypotheses. First, we calculate the absolute value of the distance between the nominee’s ideology and the ideological score of the Senate filibuster pivot.18 For the nominee’s ideology we utilize Segal/Cover scores that are converted to a 0–1 scale (Epstein and Mershon 1996, 269).19 For the filibuster pivot we use DW-NOMINATE scores.20 We expect that as this distance increases the president will have a more
17 While a Poisson model is also an appropriate modeling choice for count outcomes, the data we employ do not lend themselves to this technique. Indeed, the variance of the dependent measure is 13.86, which is greater than its mean of 8.07. This means that the Poisson model would produce con- sistent but inefficient estimates as well as downwardly biased standard errors (Long 1997, 230). We could also use a Negative Binomial Regression model (Long 1997; Greene 2003) which accounts for overdispersion by allowing “the conditional variance of y to exceed its conditional mean” (Long 1997, 230). However, because we did not have a priori expectations for model dispersion, King’s general- ized event count is the most flexible and appropriate model for these data.
18 Using the filibuster pivot comports with Krehbiel’s (1998) theory of “pivotal politics,” which holds that supermajoritarian institutions—such as the Senate filibuster and the presidential veto— have an important influence on the legislative process because policy makers often must obtain large supermajority coalitions to secure passage of legislation. Moraski and Shipan explain why the filibuster is pivotal explicitly during the nomination process: “The reason for this...is that the Senate can filibuster on a nomination, a tactic that the president will need to take into account. While the use of the filibuster for a Supreme Court nomination is a rare event, its infrequency does not denote its unimportance” (1999, 1093).
19 The reader should note that our measures of ideal points (DW-NOMINATE scores and Segal/Cover scores) might not be based on an identical policy space. That is, while both sets of scores have face validity—DW-NOMINATE represents Senator Ted Kennedy as being very liberal and Segal/Cover scores show Justice Antonin Scalia as a staunch conservative—we cannot be sure that a specific Segal/Cover score maps directly onto the DW-NOMINATE space. However, measures that correct this problem do not have measures of ideology for rejected nominees and are wrought with so much imprecision as to make empirical inference impossible (Bailey 2002; Bailey and Chang 2001). To test whether these cross institutional measures change our results, we estimated the model with them in place of the Segal/Cover and DW-NOMINATE scores. The substantive results, presented in Appendix 2 (available at www.journalofpolitics.org) do not change.
20 Other work in this area (Moraski and Shipan 1999) has used adjusted ADA scores to model the dynamics of this game. However, we think DW-NOMINATE scores are more appropriate for this type of analysis for a number of reasons. DW-NOMINATE scores are constructed using an overwhelming majority of the roll-call record (only unanimous and near-unanimous votes are excluded—see Poole