Study 3 was done by counting Pro-Bush and Pro-Kerry bumper stickers in parking lots during March, 2005. For the closest possible replication, the replication study should be planned for March, 2013. If that's not close enough to the original, studies 1 and 2 were standard lab studies with college students and could be replicated more precisely.
The authors hypothesized that people were more likely to express "deviant" opinions if the opinion was like the perceived majority opinion, only stronger ("descriptive deviants"), rather than being opposed to the majority opinion ("prescriptive deviants"). In study 3 they considered people displaying pro-Kerry bumper stickers after the 2004 US Presidential election to be descriptive deviants in a county that voted Democratic, and prescriptive deviants in a county that voted Republican; and vice versa for people displaying pro-Bush stickers. p. 879:
"Consistent with our hypothesis, the ratio of Democratic-to-Republican stickers was significantly higher than the ratio of Democratic-to-Republican registered voters in Santa Clara County, chi-squared = 7.10, p < .01, indicating that the descriptive deviant (Democratic) voice was indeed overrepresented in the population. However, the Democratic-to-Republican sticker ratio was significantly lower than the Democratic-to-Republican registered voter ratio in San Diego County, chi-squared = 3.85, p < .05 (see Table 4). That is, there were disproportionately more descriptive deviant (Republican) stickers than prescriptive deviant (Democratic) stickers."
It was the clearest test of their hypothesis. Study 3 also included a survey, but it was intended to test an alternate explanation (and the response rate was low -- 57% for n = 30). This survey appears to have consisted of four questions: 1 and 2) “How would you describe your feelings toward George W. Bush [John Kerry]?” 3 and 4)“How would you describe the average San Diego [Santa Clara] County resident’s feelings toward George W. Bush [John Kerry]?”
The location in which the survey was distributed was not explicitly collected from respondents, but the questions did have the locations embedded in them (e.g., all data collected from Santa Clara county included "Santa Clara" in questions 3 and 4). The order in which Bush/Kerry were presented was randomized.