The Displaying of Yard Signs as a Form of Political Participation

Although we focus on motives for participation, our general approach also draws upon the social influence literature, particularly Huckfeldt and Sprague (1992) work on parties and mobilization, and McClurg (2006a) examination of the relationship between neighborhood disagreement, interpersonal networks, and political participation.

The 2008 Franklin County Presidential Election Panel Study consisted of a three-wave panel survey of registered voters, and focused on tracking political network composition and across the 2008 presidential election cycle. 4,548 registered voters were randomly sampled from the Franklin County voter file and sent a questionnaire by mail. Across two waves (1 initial, and 1 reminder mailing), 817 respondents returned usable surveys, yielding an AAPOR response rate of 22.5 % for the March sample (930 sampled voters were invalid). Follow-up surveys were then sent to these respondents—496 provided usable responses to the post-convention questionnaire for a response rate of 62 %, and 493 the post-election survey (yielding an almost identical response rate of about 62 %). Not all initial respondents responded to all waves (i.e., some skipped wave 2, and re-entered the panel at wave 3; some did the opposite)—427 of the original respondents completed all three questionnaires, for an overall panel retention rate of 53.5 %. In this paper, we utilize only the general election wave. The 2000 American National Election Study—part of the broader ANES time series—consisted of 1,807 pre-election interviews, and 1,555 post-election interviews; we utilize it as it includes standard questions on political characteristics and demographics, but also items on participation (including a question on yard sign displaying), and an interpersonal political network battery. Complete information on the study and design can be found at: http://​www.​electionstudies.​org.

Such “expressive” motivations are likely preceded by emotions such as enthusiasm, as these have been found to predict political action (e.g., Marcus and MacKuen 1993).

Our work is also situated in a long-line of research looking at the distributions of behaviors and/or opinions in contexts, whether defined in terms of neighborhoods or other units (e.g., Huckfeldt 1986; Leighley 1990; Putnam 1966). In distinguishing—and observing the relationships—between levels of context, in particular we draw upon the work of Huckfeldt and Sprague (1992, 1995) and McClurg (2006a), though our work also speaks to efforts by Djupe and Gilbert (2006) and Huckfeldt et al. (2004), among others.

Other than traditional yard signs, we observed home-made signs, stickers and posters attached to doors and windows. We also witnessed a pro-Obama statue, a pro-Obama lawn chair, a demagogical sign posted on a tree (accompanied by a video camera and an threat toward anyone who vandalized the sign), and even a sign mocking the practice of putting up yard signs. We did not include bumper stickers because of the measurement error associated with the presence/absence of such symbols at the time of observation.

Among the metropolitan areas of the 30 largest U.S. cities, Columbus is at or near the mean in most important demographic categories including Black percentage, median income, poverty rate, home ownership, educational attainment, percent married, and median age. The only category in which the Columbus area differs significantly is its lack of a sizable Hispanic population. According to an index of the previous eight categories, Columbus’ average deviation from the mean ranks second.

There are at least three other reasons why surveying non-participants would have been problematic in the context of this study. First, asking people why they chose not to display a yard sign would have required different survey questions in some cases, and the total exclusion of questions in other cases, leading to concerns over comparability. Second, it is not clear that asking people about motives for inaction is straightforward, especially in the case of displaying a yard sign. With a socially prescribed act such as voting, people who choose not to vote may have considered reasons for not doing so, but in the case of displaying a yard sign, the answer may be, in essence, that the act never occurred to them. Third, concerns about the accuracy of self-reported motivations may be multiplied when it comes to non-participation: Nesbitt and Wilson (1977, p. 252) note that when it comes to such introspective reports, “non-behavior will generally be less available than behavior.”

Sampling of precincts was based on the 2006 Ohio Colleges exit poll, which used stratified systematic probability-proportional-to-size (PPS) sampling to generate a representative sample of voters. Polling places were divided into five strata based on percentage of 2004 presidential vote for John Kerry. Systematic PPS sampling was used to identify polling places within strata, and after matching these polling locations with the precincts from which they draw, we sampled precincts within polling locations.

Upper Arlington is a close-in suburb of Columbus with a population of approximately 34,000, and was chosen as a subject for a case study not due to its representativeness (it is upper income and lacks any racial diversity), but due to its self-apparently high level of democratic participation. Upper Arlington has long had an active local Republican party, but is politically competitive, and Democratic participation has been bolstered in the last four years by the creation of Upper Arlington Progressive Action, a grassroots organization that has been active in the last two presidential campaigns. At more than half of the Upper Arlington homes that posted a sign supporting Barack Obama, at least one of these signs was a sign distributed by UAPA. In addition to its turnout rate of well over 80 %, Upper Arlington was inundated with political yard signs (with over 2,700 out of 11,500 housing units having at least one).

For the analyses that follow, we exclude the eight respondents who voted for Ralph Nader or Bob Barr in the general election, and the one voter who claimed to have voted for Ron Paul (presumably by write-in since he was not on the Ohio general election ballot).

Of the 15 voters who gave their preferred candidate a rating of ‘5’ or lower, all were Republican voters, and 14 voted for a candidate other than McCain in the primary. All but one of these individuals gave Sarah Palin scores of ‘8’ or higher, suggesting their support for McCain was primarily based on enthusiasm toward Palin. In fact, a number of these individuals posted yard signs with messages such as “Sarah!”, which quite unusually did not contain the presidential nominee’s name at all.

Because of the small number of respondents choosing the “not at all important” option, all models were re-estimated with the third and fourth response options combined. Results were nearly identical in all cases, and all conclusions are robust to this change.

Specifically, the elections chosen were races for President (Obama–McCain 2008, Bush–Kerry 2004 and Bush–Gore 2000), governor (Taft–Hagan 2002), Senator (Brown–DeWine 2006, Voinovich–Fingerhut 2004, and DeWine–Celeste 2000) and Attorney General (Dann–Montgomery 2006). Other races were excluded from the index because they produced very little variance across precincts.

While there is some debate over the measurement of interpersonal disagreement, and by extension, the meaning of homogeneity in social circles (e.g., Klofstad et al. 2013), measures based on shared partisanship have been widely used (e.g., Mutz 2002, 2006).

The measure is based on the authors’ judgments regarding the amount of visibility a sign might have on a given property. Streets are coded between ‘1’ (for a ‘dead end’) and ‘6’ (for a ‘major artery’), and the score is multiplied by a weight of 1.5 if the property is located at an intersection with a busier street. Inter-coder agreement was more than 94 %, with discrepancies resolved by the authors through discussion. Full details regarding this measure are available upon request, and in a companion paper (Makse et al. 2010).

This variable was excluded from the Upper Arlington models since only 4 out of 705 respondents were African American.

One might argue that it is the relative propensity to select these prompts that is important—after all, individuals were unconstrained in agreeing with as many of these prompts as they wished, and it seems that Obama supporters and strong partisans were more likely to agree with the prompts generally. For this reason, we also conducted pairwise analyses of the propensity to agree with the prompts: (1) “show pride” versus “let neighbors know”; (2) “show pride” versus “catch the eye of traffic”; and, (3) “let neighbors know” versus “catch the eye of traffic.” We found that individuals whose neighbors posted signs were significantly more likely to cite both of the communicative motives, relative to the “showing pride” motive. Sign placement, however, did not affect the relative likelihood of choosing the two communicative motives. We also found that individuals with homogeneous networks were relatively more likely to choose ‘show pride’ compared to either of the other prompts, but there was no difference between the propensity to choose between the two communicative prompts. Finally, individuals who live on busy roads and individuals in homogeneous neighborhoods are relatively more likely to choose ‘catch the eye of traffic’ compared to either of the other two prompts, but there was no difference in the propensity to choose between the other two.

Indeed, among our observations, we saw a dozen individuals with 15 or more signs on their property.

Throughout this section, unless otherwise indicated, changes in predicted probabilities for continuous variables are based on changes from one standard deviation below the mean to one standard deviation above the mean. The changes in predicted probabilities for the ordered scale questions (folded party identification, folded ideology, political knowledge, education, religious importance, religious service attendance, “taking politics personally,” and “received encouragement”) are based on changes from the “minimum” response to the “maximum” response.

We also find two sets of strangely contradictory results. Turnout in general elections is negatively associated with distributing signs, while turnout in primary elections is positively associated with it. Meanwhile, political knowledge is positively associated with distributing signs, while education is negatively associated with the act. Although we find no particularly notable levels of multicollinearity in the models, we suspect that these unusual results stem from difficulty in parsing these similar effects apart.