Nothing to Hide, Nowhere to Run, or Nothing to Lose: Candidate Position-Taking in Congressional Elections

A notable exception is Campbell (1983a) who measures ambiguity with the standard deviation of the public’s perception of a candidate’s position on the issue instead of directly from actual candidate statements as in our analysis.

While candidates might deviate from their strategy in response to another candidate, it is not clear that this is a good idea. For example, Kathleen Brown responded to Pete Wilson’s focus on crime policy in the 1994 California Gubernatorial campaign. Directly addressing crime instead of attempting to shift the debate towards economic policy appears to have cost her votes and possibly the election (Iyengar and Valentino 2000). In our data, we see very little evidence that candidates respond to the position-taking strategies of their opponents. We address this topic more in the supplemental information.

Admittedly, it may also be the case that low quality candidates’ more frequent mention of the issues is because they are inferior campaigners and fail to recognize the risks associated with such a strategy. Adjudicating between these two explanations is not vital to our analyses.

For an exploration of ambiguity and primary elections, see Meirowitz (2005).

Though the salience of the issue has increased as more states have considered legislation explicitly recognizing or banning gay marriage, the issue should not be considered new, as the first court ruling on gay marriage occurred in Minnesota in 1971 (Haider-Markel 2001).

For 2006, one of the authors or a research assistant visited every candidate's campaign website on the Internet Archive at www.​archive.​org on the nearest available date prior to Election Day. For 2008, the archived webpages were accessed at: http://​www.​archive-it.​org/​public/​partner?​id=​316. Viewing websites at a different time during the campaign should not have changed the conclusions. We randomly sampled 50 statements on gay marriage and the Iraq War from 2006 and saw if the statements had changed between Labor Day and the election. Only eight of the 50 statements had changed. Of those eight, only three would have resulted in different codes (i.e. clear, ambiguous, or silent) for the statements. On average, these archives do not include any multimedia components of the original website. The vast majority of websites did not contain any video content on their issues page and none had their issue positions in video without also referencing the issue in text. We do not believe our exclusion of video messages will affect our analysis. Websites with video content tend to be clustered in highly-competitive, well-funded contests with the purpose of providing the same message as the site’s text in a more visually appealing form (Druckman et al. 2007).

For a more extensive discussion on the value of using campaign websites to study candidate strategies see Druckman et al. (2009, p. 345). An alternative approach to website use would be to code campaign advertisements. There are two potential problems with this. First, the representativeness of our sample would be compromised, as many House candidates cannot afford to produce ads while nearly all are able to maintain a website. Second, advertisements necessarily limit what a candidate can say because of time constraints. A comparison of candidates’ website content and their ad content has shown that they are similar, with websites mentioning a greater variety of issues and placing more emphasis on party “owned” issues (Sulkin et al. 2007).

The three examples are taken from candidate websites: (i) Perfecto Rivera, 2006 Republican Candidate in Wisconsin’s 4th Congressional District; (ii) Richard Hough, 2006 Republican Candidate in Hawaii’s 1st Congressional District; (iii) John Thrasher, 2006 Democratic candidate in Arizona’s 2nd Congressional District.

For gay marriage, 212 statements were found that could have been mentions of gay marriage. The three coders then determined if the statement was clear, ambiguous, or silent on the issue. Of those statements, 75.5 % were unanimously coded, with an average pairwise agreement of 83.3 % and Fleiss’ κ equal to 0.649. For the Iraq War, there were 774 statements. The coders were unanimous 60.7 % of the time, with an average pairwise agreement of 73.6 % and a Fleiss’ κ of 0.587. In six cases for the Iraq War and two cases for gay marriage, the coders each independently assigned a different code. The coders discussed these eight cases and unanimously agreed the statements were ambiguous in seven of the cases. The remaining case was judged to be clear.

We include the results from the most recent election because this should serve as a better indicator of district ideology during the campaign even though the 2008 election results were obviously not known until after the campaign was over. For example, Republican candidates who were in marginally favorable districts in 2004 would know that those districts likely were less ideologically congruent by 2008 given the election results of 2006 and the economic conditions of the fall of 2008.

The median number of respondents in a district is 78 with a range from 23 to 162.

We obtained the gender variable from the website for Rutgers’s Center for Women in Politics. In the supplemental materials, we also include a model controlling for which candidate was favored according to Congressional Quarterly (Caughey and Sekhon 2011). Inclusion of this variable diminishes the effects attributable to quality because it is highly correlated with the quality variables. Incumbents in our sample are favored 96 % of the time and low quality candidates are underdogs 94 % of the time. The results are unchanged except for the main effect of quality in the Iraq War model. The effect of quality is only statistically discernible when an incumbent is favored and a low quality challenger is an underdog—which is the modal campaign scenario by far.

We do not believe that an ordered model would be appropriate for this dependent variable. Ambiguous statements can be as lacking in actual policy content as silence on an issue (e.g. “I support our troops overseas.”). Hence, ambiguity is not simply a middle category between silence and clarity. Further, the parallel slopes assumption of the ordered model is violated in this case.

Full results are available in the Table 2 of appendix at the end of the article.

In an alternative model, we measure salience with a binary variable indicating whether the state had a ballot measure concerning gay marriage on the ballot that year. This model, available in the supplementary material, shows no substantive differences from the model presented in Table 2.