Attribute Priming Effects and Presidential Candidate Evaluation: The Conditionality of Political Sophistication

Abstract

Traditionally, two competing claims have arisen that attempt to explain the role of political sophistication in media effectiveness. I reassess the positive versus negative impacts of political sophistication on media priming effects by considering a curvilinear approach. I combine public opinion data (National Election Studies) on candidate selection criteria in 1992 and 2000 presidential elections with content analyses of campaign news coverage to see which segment of voters at different sophistication levels is most susceptible to media agendas. Quadratic regression analyses reveal that an inverted U-shaped relationship exists between voters' susceptibility to campaign news and their level of political sophistication. Such a curvilinear relationship means that the moderately sophisticated are more likely to accept news agendas than the least or most sophisticated. The findings illuminate the long-standing debate about the inconsistent linear relationships between the two variables, providing a more cogent explanation underlying media priming effects.

Notes

¹The Candidate Master Code used in analyzing both media and public attribute agendas about the presidential candidates is from NES survey and originally includes 281 specific attribute items, all of which were subsumed into 11 attribute categories: experience/ability, leadership qualities, personal qualities, party connections, government management, government activity/philosophy, domestic policies, foreign policies, group connections, events unique to one campaign, and other candidate characteristics. For the detailed attribute items, please refer to the codebook files in the NES data archive, which is available at http://www.electionstudies.org/studypages/download/datacenter_all.htm.

²Cohen's kappa for the coder reliability test was based on how many attribute codes were agreed between the coders, and the following formula was used to calculate the intercoder reliability coefficient: (PA_O – PA_E)/(1 – PA_E) [PA_O = percentage of agreement observed; PA_E = percentage of agreement expected] (for details, refer to Bakeman, 2000; Dewey, 1983).

³To construct a maximally valid measure of political knowledge, Delli Carpini and Keeter (1993) analyzed the NES surveys and developed five-item knowledge indices, representing individuals' factual and ideological knowledge on politics. Those indices include party control of the house, veto override percent, party ideological location, judicial review, and identifying the vice president.

⁴The current model includes polynomials such as product and interaction terms, which are likely to cause insufficient variability in the values of explanatory variables. To reduce such multicollinearity in the polynomial regression, the procedure of standardization of the raw data—recoding raw data into the deviation from the mean value—has been done (Gujarati, 1995, pp. 343–344). After the transformation of data, the VIF (variance-inflation factor) values—one of the useful indicators of multicollinearity—were sufficiently acceptable for the suggested model, generally not exceeding the level of 10 (for a specific explanation about VIF, refer to Mansfield & Helms, 1982).

Note. Numbers are the percentage of the time each candidate attribute appeared in news stories, and was mentioned by voters as their criteria for candidate selection.

Note. Table entries are ordinary least squares coefficients, and standard errors are shown in parentheses.

+p < .10. *p < .05. **p < .01 (two-tailed).

Additional information

Notes on contributors

Sungtae Ha

Sungtae Ha (Ph.D., University of Texas at Austin, 2004) is a Lecturer in the Department of Journalism and Communication at Kyung Hee University. His research interests include public opinion, persuasion, political communication and cyber democracy.