Gender Stereotypes, Information Search, and Voting Behavior in Political Campaigns

See http://​www.​processtracing.​org for the Dynamic Process Tracing Environment (DPTE) software and user guide. Any researcher may request access to the system for research purposes by clicking on the appropriate link on the website. Funding for DPTE and some of the research reported here was provided by support from the National Science Foundation and the University of Iowa.

See Figure 1 in the Online Appendix.

In Study 1, participants could also learn information from campaign television ads which periodically took over the computer screen, interrupting the information search process. These 20 s ads were typical of presidential campaign advertisements, except that for the most part they were positive in nature, highlighting one particular issue for each candidate. There was nothing in the ads that could not be learned by clicking on the scrolling boxes and reading the resulting detailed positions. Video ads were not used in Study 2.

Extensive details on the study are available in Lau and Redlawsk 2006. The number of participants was 194 for Experiment 1, 97 for Experiment 2, and 107 for Experiment 3. Participants, while not a random sample, generally represented the demographics of the area from which they were recruited.

There were four possible candidate “personas” in each party’s primary. Among the Democrats, there was an extremely liberal candidate, a “mainstream” liberal candidate, a somewhat conservative candidate, and a “mixed” candidate who took both conservative and liberal positions, but averaged right down the center. Likewise, Republican primary voters could choose among a mixed-ideology Republican candidate, a relatively liberal Republican candidate, a mainstream (conservative) Republican and an extremely conservative Republican. In each party, the “mixed-ideology” candidate and the “mainstream” liberal or conservative candidate were designated for the gender manipulation; all other candidates were always male. See Figure 2 in the Online Appendix.

Participants chose the party in which they wished to vote in the primary; when we refer to in-party this is the party they voted in, the out-party is the other party. Where appropriate in the analyses below we control for the number of primary in-party candidates.

In four additional cases there were female candidates in both parties. Because we are interested in comparing male candidates to female candidates, we dropped those cases from the analysis.

These include exposure to either two or three campaigns simultaneously, as well as variation in the office at the “top of the ticket” between the Presidency and Governor. Participants always saw a race for the House of Representatives, while half saw a presidential race and half saw a gubernatorial race. Half of the sample also saw a Senate race, while the other half did not. There were a total of 279 participants overall, but since we manipulated gender only in the presidential campaign our effective sample is the 132 participants who saw that campaign. The other unrelated manipulations affected the information environment during the campaigns. Half of participants saw campaigns with a “realistic” Distribution of information, where there was more information available for the presidential candidates than for the House candidates. Others experienced campaigns that had “equal resources,” so there were equal numbers of information boxes no matter the level of the office sought. Finally, the media attribution for some information was varied so that some participants saw certain information from conservative outlets, others saw information from liberal outlets, and another third saw no attribution. All of these treatments were randomly assigned and are controlled for in our analyses.

Unlike Study 1, candidates did not vary by ideology—the Democratic candidate always had “mainstream” liberal Democratic issue stances, while the Republican always had “mainstream” conservative Republican positions.

Participants who identified as independent were asked which party they felt closer to, and were placed in the appropriate group. Those who could not choose a party were dropped from these analyses.

Pictures of the candidates in the race are available in the online appendix.

Because the vast majority of the information voters could view was text, and the length of each item varied, the time it took participants to read items would also influence information search. Thus we control for the number of words in each item and the reading ability of the voter. Reading ability was measured as the time it took each subject to read the instructions and scenario presented before the experiment began, which was automatically calculated by the computer. Full models are available from the authors upon request.

We examined whether there were partisan differences between voters in information search for male and female candidates. The results (not shown) also fit the voting model, with partisans of both parties equally likely to focus on female candidates.

As with the primary election the general election model controls for total information search, the number of words in each item and participants’ reading ability, as well as for the vote preference.

Again, we only present figures for statistically significant results. The results for total search and trait-based search can be found in Figure 7 in the Online Appendix.