WikiSummary, the Social Science Summary Database

Theilmann and Wilhite: Campaign tactics and the decision to attack

From WikiSummary, the Free Social Science Summary Database


Theilmann and Wilhite. 1998. Campaign tactics and the decision to attack. JOP 60:1050-62.


An empirical piece hoping to test two competing theories (though its design does not allow it to do so; see criticism below).

The Literature: Two competing theories

  1. Skaperdas and Grofman published a formal model in which a candidate's decision between positive and negative ads was determined almost entirely by his position in the polls.
    • Negative ads make the opponent's supporters question their support for the opponent.
    • Positive ads make undecided supporters like you more.
    • Thus: If you're behind, run both positive and negative ads. If you're ahead, run only positive (no need to hurt the other guy, so use your money on positive ads).
  2. Harrington and Hess use a spatial model to show that a candidate's ideology and personal characteristics both matter. They argue that correlations between polls and negative ads are spurious; in reality, both reflect a candidate's personal charateristics.

The test


An experimental survey. They contacted 100 or so campaign consultants (with actual experience), presented them with a series of situations (manipulating their candidate's position in the polls), and asked what mix of positive and negative ads they would use.


  1. If Skaperdas and Grofman are correct, the authors expect to find a correlation between the hypothetical candidate's poll position and use of negative ads.
  2. If Harrington and Hess are correct, the authors expect to find no correlation between the candidate's poll position and use of negative ads. Why? Since the consultants are provided no information about candidate characteristics in the survey's hypothetical situations, the candidates will randomly guess about these characteristics and choose a strategy accordingly. Thus, relative to poll data, use of negative ads should be random.


  • Of course they find a correlation between poll position and recommendations to use negative ads.
  • Also, Republican consultants seem to favor negative ads more than Democrats do. (But this may be an historical artifact--in 1996, the Democrats were in trouble and the Republicans were rising, making negative ads a more fitting strategy for Republicans at the time.)

Comments and Criticism

The authors are incorrect to claim that, if H&H are right, there will be no correlation between poll data and recommended spending. Consider this: if candidate characteristics are important, then the consultants will try to extract the information they need from the survey. Since they are provided with little information, they might try to use polling data as an information shortcut to estimate the candidate's personal characteristics. Thus, given the structure of the design, S&G are observationally equivalent with H&H: both models predict a correlation between poll data and negative ads.