The Candidate Effect: Does the Local Candidate Matter?

Abstract

This paper considers how the quality of the local candidate affects vote choice. Specifically, we address three questions: Does the quality of the local candidate influence vote choice? What impact do individual-level differences have on the relationship between vote choice and local candidates? Finally, what is the potential magnitude of candidate effects in terms of change in vote support? To answer these questions, we analyze data gathered from an online voting experiment. Our findings suggest that a local candidate can influence vote choice significantly, but that such effects are tempered by political awareness and partisanship.

Notes

1 All students at Wilfrid Laurier University, Waterloo, Ontario, were invited to participate.

2 Our initial sample consisted of 552 participants. Twenty-seven of these individuals were removed from the final analysis due to missing information on key variables. The response rate for this study was just over 15%.

3 We included four Canadian federal political parties that were represented in Parliament at the time of the study (2012): the Conservative Party of Canada (government), the Liberal Party of Canada, the New Democratic Party of Canada (NDP – official opposition), and the Green Party of Canada. A fifth political option, the Bloc Quebecois, is excluded from our study because it only fields candidates in the province of Quebec, making this party inaccessible to members of our sample.

4 Note that each participant took part in only one election.

5 An alternate means of estimating the average vote share would have been to include a control treatment where all candidates are presented as equally weak or strong. Unfortunately, these data are not available. As such, we rely upon the average of the three non-treatment groups as our baseline.

6 See http://www.elections.ca/content.aspx?section=res&dir=rec/part/estim/41ge&document=report41&lang=e#p4.

7 The gender difference is in line with the proportion of women and men at the university from which the sample was drawn.

8 Our measure of political sophistication is the standardized measure of political interest with political knowledge. Employing this combined measure guards against the multi-colinearity introduced into our model by including each of these measures independently. Note that the results hold with and without the inclusion of these controls.

9 We wish to thank one of our anonymous reviewers for suggesting this line of inquiry.

10 All marginal effects/discrete changes reported in this manuscript are estimated using Stata12's post-estimation margins commands with all covariates held at their means.

11 Note that these results do not reflect differences across partisans, leaners, and non-partisans in regard to the amount or type of information each individual accessed. Comparison of the number and types of links shows no significant difference across individuals (results available from authors upon request).

12 Although note that the sign of the coefficient for partisans in elections where the stronger candidate is competing for another party is negative, and just short of conventional levels of statistical significance (p=.12).

13 All values are held at their mean.

14 To estimate this effect, we subtracted the average predicted vote probability when a party fields a relatively weak candidate from the average vote probability observed for a strong candidate with all other covariates held at their means.

15 We might also consider these results taking into account potential “ceiling effects” (Górecki, 2013). Given the relatively high partisan support of their candidate to begin with (84%), there is only a 16-percentage point increase possible. In the case of leaners, the difference between the support observed for an average candidate and full support is 40-percentage points. In other words, there is a much greater gap between initial support and full support in the case of weak partisans. To account for these differences, we can consider the relative increase in vote probability for each type of voter. To do so, we divide the change in vote probability by 100% minus the vote probability obtained for the weak candidate. This yields a relative candidate effect of 16% for non-partisans, 13% for leaners-rival, 97% for leaners-preferred, 9% for partisans-rival, and 62% for partisans-preferred. While these relative changes differ from the raw changes reported, especially in the case of leaners-preferred and partisans-preferred, the substantive interpretation remains the same; the strongest candidate effect is clearly observed for leaners-preferred.