Abstract
Congressional elections are crucial to the American political system and candidates spend millions of dollars seeking votes with television spots. Unfortunately, the literature in this area is a hodgepodge of studies (with different methods and samples) rather than a comprehensive analysis of congressional television advertising. This study utilized the Functional Theory of Political Campaign Discourse to content analyze 744 television spots for House and Senate candidates from 1980–2004. Candidate discourse in these spots employed acclaiming (positive) strategies much more frequently than attacking (negative) or defending (refutational) strategies. Unlike discourse in presidential campaigns, congressional TV spots tend to place equal emphasis on policy and character (although since 1992 the emphasis has been on policy). Democrats tend to attack more and to discuss policy more than their counterparts. Incumbents acclaimed more and attacked less than challengers, whereas open-seat candidates have a style that lies between these two extremes. Open-seat candidates discuss past deeds less frequently than incumbents or challengers, both of whom tend to rely on the incumbent's record to attack (challengers) or to acclaim (incumbents).
Notes
Note. Rows may not total 100% due to rounding. Boldface shows larger percentage.
Boldface shows larger percentage.
∗Acclaims and attacks for each form of policy and character are reported in first row and the next row contains the total and percentage of policy or character; figures may not total 100% due to rounding.
The tests reported here are very powerful because of the n. Cohen's (Citation1988) power tables only reach ns of 1000; the total number of themes in this study is 4038. The power of a χ2 with df = 1 and an n of 1000 (our n, the number of themes in the chi-square, is much larger) to detect small, medium, and large effects is .89, .99, and .99, respectively. The power of a χ2 with df = 2 and an n of 1000 to detect small, medium, and large effect sizes is .82, .99, and .99. Thus, it is very unlikely that we will commit a Type 2 error when we report a nonsignificant result with these data.
We report η as a measure of strength of association for 2 × 2 χ2s and Cramer's V for 2 × 3 and 3 × 3 χ2s.