Jump to ContentJump to Main Navigation
The Wartime PresidentExecutive Influence and the Nationalizing Politics of Threat$

William G. Howell, Saul P. Jackman, and Jon C. Rogowski

Print publication date: 2013

Print ISBN-13: 9780226048253

Published to Chicago Scholarship Online: January 2014

DOI: 10.7208/chicago/9780226048420.001.0001

Show Summary Details
Page of

PRINTED FROM CHICAGO SCHOLARSHIP ONLINE (www.chicago.universitypressscholarship.com). (c) Copyright University of Chicago Press, 2017. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in CHSO for personal use (for details see http://www.chicago.universitypressscholarship.com/page/privacy-policy). Subscriber: null; date: 30 March 2017

(p.292) Appendix B Alternative Bridging Criteria, Chapter 5

(p.292) Appendix B Alternative Bridging Criteria, Chapter 5

Source:
The Wartime President
Publisher:
University of Chicago Press

In lieu of interest groups, we rely upon individual members of Congress. From the outset, we acknowledge that this approach is fraught with dangers. For starters, we must assume that at least some members of Congress do not change their voting behavior when the nation enters or exits war. Should every member of Congress shift by a common unit, this bridging criterion will necessarily recover estimates indicating that peace and wartime votes are identical. Should we select the wrong member of Congress, it is possible to generate estimates of ideal point changes that are completely wrong. If in the 107th Congress, for instance, we select as bridge actors the Democratic and Republican Senators who in fact shifted the most in the conservative direction, our analysis for all other members will falsely reveal modest shifts in the liberal direction.

To select members of Congress whose voting behavior is likely to be least affected by war, we apply two different decision rules. The first draws on our analysis of the 107th Congress, wherein we found that the war had a larger impact on Democrats representing jurisdictions where Bush performed well in 2000 and on Republicans representing jurisdictions where Bush performed poorly. To wit, in World War II and the Korean War we select (data permitting) the member of the president's party from the state or district in which the president performed the best in the most recent presidential election, and the member of the opposition party from the state or district in which the president fared the worst.1

As a second decision rule, we focus on those members who exhibit high levels of ideological stability—members, we suggest, who are demonstrably less likely to adjust their voting behavior in response to war, presidential appeals, or anything else. To identify these individuals, we estimated (p.293) member ideal points for the entire set of roll calls that span peace and wartime periods. The Bayesian credible intervals associated with each member's ideal points provide information about the ideological stability of his or her voting record. Members with larger intervals, ceteris paribus, can be said to vote less ideologically consistently, and members with smaller intervals vote more consistently.

We cannot simply select as our bridge actors the Democrat and Republican with the smallest credible interval. These intervals, after all, also reflect each member's location within the larger distribution of ideal point estimates. Centrist legislators tend to have very precisely estimated ideal points, because there are many cutting lines that separate them from other centrists. It is more difficult to know exactly how extreme a very conservative or liberal member is, however, because there are fewer cutpoints between them and other like-minded legislators. We therefore regress the estimated credible intervals on a polynomial expression of their associated ideal points, and then we recover the residuals. Members with the largest negative residuals, we suggest, vote most consistently given their location in the ideological distribution, whereas those with the largest positive residual vote least consistently given their location.2 Members of the former group qualify as plausible bridges. Because the differences in the size of the residuals are quite small, and because we do not want to make strong claims about the accuracy of the regression's specification, we select as bridges the three Democratic and three Republican members with the largest negative residuals. We then calculate the aggregate shift in chamber ideal points using each of their nine combinations, and we report the results from the pair that yields the median mean shift.

Recall that in our analysis of the 107th Congress, party leadership constituted another significant predictor of members' voting shifts. Particularly among Republicans, party leaders moved significantly further in the conservative direction after the outbreak of war than did rank-and-file members. Hence, for both decision rules we consider only members who are not party leaders. As a practical consideration, we further limit our selection of Democratic bridge actors to non-southerners so as to ensure adequate coverage of the liberal and conservative regions of the (assumed) unidimensional continuum. (p.294)

Table A.B1 Assessing Member Bridges

Congress and Chamber

Expected Direction

Interest Group Results

Presidential Vote Share Criterion

Regression Residuals Criterion

Reverse Presidential Vote Share

Reverse Regression Residuals

107th House

Conservative (+)

+0.54

+0.06

+0.15

−0.16

−0.01

107th Senate

Conservative (+)

+0.69

+0.10

+0.00

−0.01

−0.08

Note: Cell entries reflect the mean shifts in chamber ideal point estimates. Entries in bold indicate statistical significance at p 〈 .05 or lower, using one-tailed t-tests. Entries in the first three columns of results test the hypothesis that shifts in the expected direction (second column) are observed, and the entries in the two rightmost columns test the hypothesis that shifts opposite those expected are observed.

(p.295) As an initial check on both decision rules, we replicated the analysis for the Afghanistan war, in which interest group positions are available to anchor our analyses. In all four cases, these two decision rules yield results that are consistent with those generated from interest group bridges. These results are displayed in Table A.B1.

As a further check, we again replicated the analyses for the Afghanistan war, but this time we selected those individuals who, under the two decision rules, we would expect to be the worst possible bridges. Thus, under the first decision rule, we selected the individual from the president's party who represented the district or state where the president performed least well in the previous election, and the individual from the opposition party where the president performed best. Under the second decision rule, we selected those individuals from both parties with the largest positive residuals. If our decision rules do a reasonable job of distinguishing members whose voting patterns are more and less resilient to war, then these “reverse bridges” should generate very different results from the more reliable estimates based upon interest group bridges. And so they do. In all four cases, asTable A.B1 shows, we found shifts in the opposite direction from those observed in the interest group bridges. Further, two of these cases show statistically significant movement in the direction opposite from what we find using the interest group bridges. Together, these two cross validations suggest that we have a plausible basis for examining the impact of earlier wars across a variety of legislation.

Notes:

(1) . House elections data were generously provided by Gary Jacobson. Data on district-level presidential returns are not available until the 1952 election, so we do not use this bridging criterion for House members in World War II or the beginning of the Korean War.

(2) . This approach is similar in spirit to Lauderdale 2010, who develops a Bayesian heteroskedastic ideal point estimator to directly model the legislator-specific variances that describe the extent to which legislators' voting behavior is not conditioned on the primary liberal-conservative dimension.