“They drive different cars, watch different sports, and drink different beer […] [P]olitics is no longer just about politics. It is about life.” (Hetherington & Weiler, 2018: 120).

Politics is increasingly permeating the everyday lives of Americans. Pundits and scholars alike have posited that there exists a cultural chasm in American politics pitting Latte-drinking, Volvo-driving Liberals against NASCAR-watching, country music-loving Conservatives (Frank, 2004; Hetherington & Weiler, 2018; Nunberg, 2006). Descriptive survey and aggregate-level data confirm that Republicans and Democrats have diverging non-political preferences across a range of issues, including the cars they drive, the coffee they drink, or the sports they watch (for an overview see Hetherington & Weiler, 2018: 89–121). Moreover, far from being simple caricatures, recent studies have documented partisan differences in non-political behavior. For instance, liberals and conservatives differ in the names they give their children (Oliver et al., 2016), the places they choose to call home (Bishop & Cushing, 2008), and the amount of money they give to charity (Margolis & Sances, 2017).

Given the importance of understanding partisan behavior, some scholars examine the antecedents of these “partisan” differences (e.g., Hetherington & Weiler, 2018; Iyengar et al., 2012). In this paper we are interested in two different facets of this phenomenon: (1) does the public associate cultural objects and activities with partisanship, and (2) are these associations consequential for the evaluation of politicians? We first examine the extent to which Americans associate non-political objects, activities, and places with partisanship. Then, building on the literature on heuristics in political decision-making, we test whether these associations influence how citizens evaluate the partisanship of candidates (above and beyond policy information), by manipulating the background of photographs of politicians giving speeches.

Our findings suggest that individuals can and do assign partisan associations to objects and activities – that is, they perceive some objects and activities to be associated with Democrats and others with Republicans. But our findings also make clear that the influence of these subtle cues is limited when presented alongside more explicit policy information.

The Politicization of Culture

Hetherington and Weiler’s (2018) recent book Prius or Pickup? highlights stark cultural divisions between Republicans and Democrats. Their insights reflect broader concerns, both in academic and popular debates, about increasing divides between the culture of conservatives and the culture of liberals (e.g., Fischer & Mattson, 2009). It appears that the divisions between the two major ideological poles of the United States run deeper than just policy and political differences; they are seeping into everyday life such that the animosity felt between partisans translates into animosity towards people with different non-political preferences. The stereotypes of Americans, according to Hetherington and Weiler, are roughly as follows: Democrats drive Priuses, drink at Starbucks and listen to hip hop; Republicans drive trucks, drink at Dunkin Donuts, and like country music. And these, “[n]onpolitical differences seem to be central to why Republicans and Democrats see their counterparts as so alien” (2018: 93).

Recent work by Settle (2018) echoes this point by demonstrating how polarization is fueled through social media, oftentimes through posts that are not overtly political but in which non-political cultural content nonetheless proves politically informative. Settle finds that Americans can infer partisanship from social media posts that include only very implicitly political content (e.g., a post praising the fast-food chain Chick-Fil-A, which is known for its conservative values). These cultural cues, Settle suggests, amplified in the new media environment, may serve to increase political polarization, and thus be a very real concern for sustaining the deliberation and cooperation required in representative democracy.

There are many other facets of this “cultural divide”. Recent work highlights partisan differences in moral concerns (Graham et al., 2012), consumption and boycotting of commercial brands (Simon, 2011), and grammar preferences (Cichocka et al., 2016), for instance. We focus here on factors more central to the accounts provided by Hetherington and Weiler (2018) and Settle (2018) – we examine the associations individuals make between partisanship and a wide range of objects and activities, as markers of lifestyle or culture, broadly construed.

There is some evidence for this relationship already, such as the association between partisanship and drink choice (DellaPosta et al., 2015; Wilson, 2014), or the embracing of pantsuits as a Democratic political symbol during the 2016 US presidential campaign as an homage to Hillary Clinton (Grinberg, 2016). There is, we believe, mounting anecdotal and empirical evidence that Americans readily (and in some cases also accurately) associate Republicans and Democrats with different sets of ostensibly apolitical objects and activities. Our first goal is to further explore a hypothesis already suggested in the work outlined above:

H1: Individuals readily attach partisan leanings to a wide range of objects and activities.

One novel contribution of the work that follows lies in our test of H1 using both open-ended and closed-ended responses on an especially broad range of objects and activities. We regard these tests in large part as a necessary precursor to the second objective of our work, namely, an exploration of the consequences of such associations in the context of campaign communication and electoral competition. More precisely, we are interested in ways in which “partisan” objects and activities matter for candidate evaluation. There is, of course, a rich and long-standing literature devoted to understanding how citizens form their political preferences (e.g., Jost et al., 2009). That literature makes clear that citizens regularly turn to a range of cues or shortcuts to help guide evaluations about politicians and policies (Lupia & McCubbins, 1998; Mondak, 1993). These cues include party identification and organizational endorsements (Arceneaux & Kolondy, 2009; Dancey & Sheagley, 2013; Lupia, 1994). Indeed, a politician’s partisanship is arguably the most important heuristic citizens rely on, as it provides a simple, usually reliable signal to voters about the types of policies that they will seek to enact when in office.

There is also a vast literature on explicit cues such as partisan labels or interest group endorsements, as well as work on more subtle cues. Research on the activation of racial stereotypes tells us that a salient cue does not have to be explicit to function (Hurwitz & Peffley, 2005; Mendelberg, 2001; Valentino et al., 2002). Past scholarship also identifies a range of occupational and demographic heuristics that citizens use to make inferences about a candidate’s ideology and partisanship (e.g., McDermott, 1998, 2005, 1997). These dynamics are a central motivation for the current paper.

So too are the nascent literatures on non-verbal and visual political communication. There are literatures on the impact of news photos on citizens’ interpretations of and attitudes related to current affairs (e.g., Coleman, 2010; Domke et al., 2002; Geise & Baden, 2015; Gibson & Zillmann, 2000; Messaris & Abraham, 2001; Soroka et al., 2016); on the power of symbolic imagery such as flags to activate symbolic attitudes and thereby affect evaluations and vote choice (e.g., Dumitrescu & Popa, 2016; Kalmoe & Gross, 2016); and on the ways in which images – primarily candidate ballot photographs – provide informational shortcuts about political candidates (e.g., Carpinella & Johnson, 2016; Dumitrescu, 2016).Footnote 1

Most central for our work is the fact that the power of symbolism and stereotypes in these literatures depends on there being well-established, widely understood associations between partisanship and a range of phenomena that are at best only indirectly political. We see this as a critical overlap between the literatures on (a) the politicization of culture and (b) the role of heuristics in candidate evaluations. Our purpose here is to extend the scope of what can be considered heuristics. While previous research has focused on candidate appearance,Footnote 2 we suggest that other visible elements, such as a candidate’s association with particular sports, cars, or food can – and perhaps increasingly does – affect our assessments of her partisanship and policies.

It is worth noting that our expectations are in line not just with the work described above, but also with research on the role of images in the development of mental schemas (Wood et al., 2018). Images, and the corresponding cues embedded in them, serve as activation points for existing schemas that citizens hold for partisanship. Simply seeing a picture may be enough to stimulate voters to evaluate the image content through their pre-existing frameworks. This work thus seeks to understand how a range of cues may alter evaluations of candidates through these mental processes.

We also suggest that the changing media environment makes these kinds of cues increasingly prevalent, and thus augments the need for empirical testing of their impact. Politicians have always made efforts to signal their ideas and preferences by visual means, from the use of a flag to the wearing of cowboy boots (Schatz & Lavine, 2007; Skitka, 2005). But this behavior seems increasingly relevant and prevalent in a media environment in which citizens have increasing access to vast bodies of visual content, notably through social media sites and applications.

Do citizens make inferences about candidates and policies based on the objects and activities featured in this visual content? Settle (2018) provides initial evidence pointing in this direction, showing that survey respondents can infer the partisanship of their Facebook connections from purportedly non-political social media posts about hybrid cars or gun racks, for example. We seek to build on these striking findings with tests that more directly speak to the impact that non-political objects can have for assessments of politicians in the context of campaign communication. To our knowledge, there have been limited investigations into this context to date – although Swigger (2012) offers one notable example of prior work on the impact of visual information in political ads, focused on group-related cues (i.e., blue-collar workers, African Americans) rather than the objects and activities that are our focus below. Assuming, first, that citizens can readily assign partisanship to objects and activities (H1), we test the following hypothesis focused on the appearance of these cues in hypothetical candidate stump speech photos:

H2: Assessments of candidates’ partisanship are affected by the object/activity/context in which the candidate is presented.

Note that while H2 focuses on evaluations of partisanship, we also test assessments of political ideology as well as evaluations of policy. We expect both to be influenced in the same way. For the sake of parsimony, we focus on partisanship in our test of H2 in the body of the paper; but include results using political ideology in the Online Appendix.

Testing H1: The Partisanship of Objects and Activities

Sample and Methods

We first explore whether individuals indeed can and do assign partisan associations to objects and activities (H1). Our first test relies on an online survey fielded in February 2020 on the Amazon Mechanical Turk (MTurk) crowdsourcing platform.Footnote 3 We collected data from 501 respondents, with a demographic breakdown that was 43.14% male and 72.76% white, with a median age of 25–34 years, a partisan split that was 44.73% Democratic and 33.3% Republican, and where 63.81% of respondents reported having a 4-year college degree or higher.Footnote 4 Despite the commonly cited drawbacks of using MTurk samples (Casler et al., 2013), we believe the platform provides an appropriate sample for our purposes because we simply consider this as a necessary step to establish that people have these associations. That said, in subsequent studies we rely on larger and more representative samples.

Our first analysis focuses on open-ended responses. At the beginning of the survey, respondents are asked: “Now we would like to ask you about some objects or activities you associate with people who are politically Democratic or Republican – such as clothing, food, cars, hobbies, etc. Please use the spaces below to list off all of the things or activities that you associate with members of these groups.” Responses are recorded in two separate text boxes, one for each party. They provide an entirely un-aided and un-primed examination of whether respondents associate any objects or activities with partisanship.

Our second analysis relies on closed-ended questions. Following the open-ended prompt, we ask respondents for their partisan ratings for a list of 26 objects and activities. This list includes what seem to us to be more obviously partisan items (e.g., church attendance, guns, eating organic food), alongside items that we believe are less readily identifiable as partisan (e.g., hamburgers, wine, dresses). The list was developed by the authors, based on past work (reviewed above), alongside the ongoing public discourse about politics and culture. Note that we do not present this list as an exhaustive or comprehensive representation of all the objects and activities that are associated with politics in the US. Rather, we suggest that our list is representative of some of the objects and activities that are thought of as being associated with partisanship. Participants are asked to rate each on a 0–100 drag-and-drop scale, with Democratic on the left and Republican on the right.

Results

Are there objects or activities that people classify as Republican or Democratic? Figure 1 provides a striking illustration of the ease with which survey respondents are able to name – off the top of their head – objects and activities they believe have partisan associations. The figure is a ‘comparison cloud,’ plotting the words that are both most frequent and most distinguishing between the Democratic and Republican responses.Footnote 5 Democratic words are shown in blue; Republican words are shown in red. The size of each word indicates both the frequency and the strength of the association. Results are clearly in line with what we might expect given past research. Respondents associate Republicans with objects such as guns, trucks, and SUVs and with activities such as hunting, fishing, and golfing; Democrats are associated with electric and hybrid cars, organic food, and yoga.

Fig. 1
figure 1

Words Distinguishing Party. The comparison cloud plots words that are both most frequent and distinguishing between Democratic and Republican responses (see footnote 5 for technical details)

Full size image

We take these results as a powerful signal that people can readily assign partisanship to objects and activities, and that their assignments reveal associations in line with what the existing literature suggests. And while some responses such as “military”, “equality”, or “protesting” constitute words that are directly linked to politics, other words such as “trucks” or “yoga” are less overtly political. There is to our knowledge no salient policy discourse surrounding “golfing” or “basketball,” for instance. The words appearing in Fig. 1 thus reflect a combination of policy-oriented but also cultural differences between partisan groups.

Figure 2 shows ratings of objects and activities from our closed-ended questions. Dots represent mean ratings, and whiskers indicate the interquartile range. These results provide further evidence that objects and activities can be readily linked to partisanship. Mean values align with what we have seen in the open-ended responses in Fig. 1: church attendance and guns are rated as highly Republican whereas items such as organic food and tattoos are rated as highly Democratic. Note that we find clear partisan attachments for objects that may be less obvious as well, such as microbrew beers (Democratic) and suits (Republican).

Fig. 2
figure 2

Party Ratings of Objects and Activities. Mean ratings (dots) and interquartile range (whiskers)

Full size image

We view Figs. 1 and 2 as rather compelling evidence that people can easily ascribe partisan associations to objects and activities. Importantly, we find these associations both when we prompt respondents with a list of objects, and in unprompted open-ended responses.Footnote 6 We next turn to the implications of these associations for political evaluations.

Testing H2: The Partisanship of Background Images

Pretesting Images

Having established that individuals associate partisanship with a wide range of objects and activities, we turn to the possibility that those objects can act as partisan cues. Specifically, we consider whether background images affect survey respondents’ assessments of the partisanship of a (fictitious) politician, in the absence of any other information about the candidate.

We begin by pre-testing a series of candidate images, using three small surveys, each of which uses a similar design: respondents are presented with an image of a candidate giving a speech.Footnote 7 We manipulate the background of the image such that the candidate is standing in different scenes that while purportedly non-political still provide political cues. Background scenes were chosen based on an early fielding of the open- and closed-ended responses (see footnote 6). All backgrounds are drawn from images publicly available online. We use images of a female and a male candidate, both White and middle-aged. The people are not actually political candidates, and we do not expect participants to recognize them. We include all photos used in the Appendix, though note that we do not reveal the female candidate’s face because that image was cropped from a publicly available photo. The male candidate is Jan Van den Bulck, a professor, and he has given us permission to use his image here.We show three examples of our photo manipulation using the male candidate in Fig. 3.

Fig. 3
figure 3

Examples of Stimuli used in Survey 2. Control condition, hybrid (Prius) condition, and guns condition

Full size image

Note that we expect the female candidate to be more likely to be seen as a Democrat than the male candidate (McDermott, 1997), and it is for this reason that we vary the gender of the candidate. That said, since we evaluate the two candidates in different surveys, we are reluctant to directly compare the ratings across candidates. We also became concerned that using a blue curtain behind the female candidate would cue Democratic partisanship, so we changed the color of the curtain for the male candidate. It is consequently the within-candidate comparisons that are of primary interest here. We expect that candidates will be evaluated as more Republican in the truck, guns, and NASCAR conditions, whereas they will be viewed as more Democratic in the organic food, tattoo, and Prius conditions.

Using the background – the “scene” of the press conference – as the means by which our object/activity-based cue is delivered has several advantages. One is pragmatic: moving an individual from one location to another is a straightforward photo manipulation, and ensures that the candidate, and her position, facial expression, etc., remains constant from one condition to the next. Moreover, candidates regularly use locations of press conferences to signal policy priorities and positions. The fact that politicians often use locations such as childcare centers, factories, farms, restaurants, and gun ranges as backdrops for press conferences and photo “ops”, suggests that our manipulation has strong external validity.

After respondents view the image, they (a) fill out a free-form box to describe what they saw in the picture (as a manipulation check), and then (b) provide an assessment of the candidate’s party affiliation. The question is as follows: “What political party do you think this person belongs to?” where response options are Democrat or Republican.Footnote 8

Results suggest that background images do indeed affect assessments of partisanship. Figure 4 shows the percent of respondents who believe that the hypothetical candidate is a Democrat (and the associated 95% confidence interval) for each experimental treatment. Asterisks indicate the conditions for which the mean is significantly different from the control, based on two-tailed t-tests. When presented with the control picture of the female candidate, 68.9% of respondents rated her as a Democrat; the male candidate in the control condition was rated as a Democrat by 40.7% of respondents.Footnote 9 The organic condition using the female candidate is rated as more Democratic (85.4%), but the tattoo condition is rated as being more Republican (60.4%). Both conditions we would expect to be more Democratic for the male candidate are rated as such compared to the control condition (43.3% for organic food and 48.1% for the Prius). Republican cueing conditions are consistently rated as such for both candidates (for the female candidate, 42.86% for guns and 31.71% for NASCAR; for the male candidate, 13.72% for guns and 18.52% for trucks). Not all of these differences are statistically significant, as is indicated in Fig. 4; but all but one of them (the tattoos condition in the female experiment) are in the expected direction.

Fig. 4
figure 4

Candidate Ratings by Experimental Treatment. Dots are percent of respondents who believe candidate is a Democrat. Whiskers are 95% confidence intervals. Asterisks indicate that condition is significantly different from respective control condition based on two-tailed t-tests

Full size image

It is worth noting that the control condition is not statistically distinct from the Democratic-cueing images for either candidate. That said, participants are responsive to a subset of (Republican) partisan cues across the two candidates. In our view, this is sufficient initial evidence that background images affect respondents’ assessment of candidate partisanship.

Sample and Methods

People rarely assess the partisanship of a politician in the absence of any information other than an image. Republicans espouse Republican political positions; Democrats espouse Democratic ones. Our pre-tests thus offer a useful initial test of the possibility that background images matter for assessments of partisanship; but a more externally valid test would examine the possibility that background images matter even in the presence of other relevant signals. This is the objective of the analyses described here.

Our test of H2 relies on data from two different surveys. Survey 1 is a broad national sample of 1,283 U.S. adults collected through Survey Sampling International (SSI) in the fall of 2016. The sample is 50.6% female, 76.1% white, and 14.7% black. 59.4% of respondents are fully employed, and 53.5% have at least a 2-year college degree. In terms of political interest, 57.1% of respondents fall into the “very” or “extremely interested” categories. Survey 2 is a national sample of 1000 U.S. adults collected through YouGov, fielded in the fall of 2018. The sample is 53.2% female, 66.6% white, and 12.2% black. 36.4% of respondents are fully employed, and 53.5% have at least a 2-year college degree.Footnote 10

In Survey 1 we presented respondents with the pictures featuring the female candidate. Each picture was randomly paired with one of two policy conditions, one clearly matching a Democratic position and the other a clearly Republican oneFootnote 11:

[Republican Policy condition] Representative Jenna McMillan was photographed speaking to a crowd at a campaign stop recently while discussing her proposal to stimulate economic recovery. Representative McMillan is proposing a series of business tax cuts. She called for a 5% decrease in corporate taxes over the next 3 years.


[Democratic Policy condition] Representative Jenna McMillan was photographed speaking to a crowd at a campaign stop recently while discussing her proposal to stimulate economic recovery. Representative McMillan is proposing a 15% increase in spending on job retraining and unemployment benefits over the next 3 years.

It is worth noting that these statements do not identify the candidate as being a member of any political party but do identify her as an incumbent. Combining the image and policy treatments, there are ten conditions in a 2 (policy) X 5 (image) factorial design. The main dependent variable of interest is the partisan evaluation the candidate which is a binary 0 – 1 variable with 1 equaling Democratic.Footnote 12

In addition, we asked two manipulation check questions after the evaluations, asking respondents to recall both the picture and the policy they just read about. Recall was rather low: Correct recall of the policy position was 63.9%, while correct recall of the photo contents was just 51.4%. As a result, we constrain our results to respondents who could correctly recall the contents of the photo.Footnote 13

The basic experimental design was unchanged in Survey 2, with the exception of the candidate’s name, in this case, Tom McMillan. We also used background images that we suspected, based on the open-ended responses, would produce somewhat stronger effects: we added Toyota Prius and Pickup Truck conditions, alongside the previous Organic condition, and a Guns photo that more strongly cues guns using a gun store rather than a shooting range.Footnote 14 As in Survey 1, we restrict our analyses to respondents who correctly identified the content of the photos. Part of the incentive to move to a YouGov sample for Survey 2 was to find more attentive respondents, however: in this case, 84.3% of respondents correctly identified the policy position and 50.6% correctly identify the photo contents.Footnote 15

Findings

Figure 5 plots mean evaluations (scaled from 0 to 100) for Representative McMillan across each condition in Survey 1, where higher scores indicate perceptions that the candidate is a Democrat. Left squares indicate average ratings across the Republican policy conditions; right squares indicate average ratings across the Democratic policy conditions; gray whiskers show bootstrapped 95% confidence intervals.Footnote 16 Note that this figure is based on simple means and standard errors of means, but it is possible to model the differences using logistic regression (since the dependent variable is binary). Those logistic regression models are included in Table H1 in the Online Appendix.

Fig. 5
figure 5

Perceived Candidate Partisanship by Background Image and Policy Position in Survey 1 (SSI). Squares represent mean partisan evaluations across photo conditions for respondents who saw the Republican policy cue (left squares) or Democratic policy cue (right squares). Whiskers represent bootstrapped 95% confidence intervals (see footnote 16 for details)

Full size image

The policy treatment has an impact on candidate evaluations: the Republican policy cue consistently leads to a candidate evaluation that is roughly 30 to 40 percentage points more Republican. This difference is similar across all photo treatments.

In comparison with the policy cue, the impact of background images is rather limited. Looking within either Republican or Democratic policy treatments reveals few marked differences. Regressions in Online Appendix Table H1 indicate that candidate evaluations shift significantly only for the Guns and Tattoo images (relative to the control condition), and only within the Democratic policy condition. In both cases, the candidate is more likely to be perceived as Republican. No other pairing of image and policy creates significant variation from the control picture; and while the Guns condition aligns with expectations, the Tattoo one does not. One possible conjecture for this latter result is that respondents believe our candidate, a white woman, is critiquing tattoos and tattoo parlors in her speech, thus seeing her more as a Republican.

Evidence that background images affect perceptions of candidate partisanship is in this instance, rather limited. Most background images have no marked impact on candidate evaluation. Failing to reject a null hypothesis is a reasonable outcome, but accepting it likely requires further evidence. It is for this reason that we pursued our Survey 2 study, this time a using (a) a male rather than a female candidate, (b) a revised control condition that uses a purple rather than a blue curtain, (c) a (partly) different set of background images, and (d) a more attentive national sample. The last point is of real significance, given that there were in this survey only 32 people who correctly identified the Guns condition. While we consider this an issue of attentiveness, it may also point toward an inability to correctly identify the shooting range picture and thus design changes for Study 2 seek to address both of these possibilities.

Figure 6 plots mean partisan evaluations for the candidate. As in the preceding figure, Fig. 6 shows means and bootstrapped 95% confidence intervals for each treatment, and logistic models of the same quantities are included in the Online Appendix. The policy treatments once again shift the partisan ratings markedly, in this case by a little closer to 50 percentage points depending on the photo treatment. As expected, evaluations of the male candidate are generally more conservative than the female candidate across all conditions.

Fig. 6
figure 6

Perceived Candidate Partisanship by Background Image and Policy Position in Survey 2 (YouGov). Squares represent mean partisan evaluations across photo conditions for respondents who saw the Republican policy cue (left squares) or Democratic policy cue (right squares). Whiskers represent bootstrapped 95% confidence intervals (see footnote 16 for details)

Full size image

The impact of background images illustrated in Fig. 6 is relatively similar to what we saw in Fig. 5. Only the Guns treatment produces a statistically significant shift in evaluations, conditional on the Democratic policy condition (see Appendix Table H2). That said, the Organic Food condition approaches traditional levels of statistical significance, conditional on the Republican policy condition. Similarly, the Truck condition pushes Democratic ratings downwards, although the impact narrowly misses traditional levels of statistical significance.

Discussion

We find, in line with recent work (esp. Hetherington & Weiler, 2018; Settle, 2018), that Americans can and do associate a broad range of objects and activities with partisanship. As past work demonstrates, this can matter for perceptions of others. Here, we consider whether it matters for perceptions of politicians. We view this as an increasingly important issue. It has been true for some time that politicians have tried to cue partisanship through the use of objects and activities in television advertising and speech locations; but opportunities for this kind of ceuing are increasing as politicians and other public figures turn to image (and video) sharing platforms such as Instagram and SnapChat. It is increasingly clear that the cues transmitted through those platforms can be of real importance (Munoz & Towner, 2017). The ways in which candidates use objects and activities as a mechanism for partisan cueing nevertheless remains understudied. We suspect that the kinds of partisan cues we consider above are of increasing significance.

That said, our work finds evidence of relatively limited effects. We find very strong support for H1 – people can readily identify objects and activities that are stereotypically Democratic or Republican. Pre-tests of our experimental stimuli also suggest that background images featuring some of these stereotypically Democratic or Republican cues can shift respondents’ perceptions of politicians. Americans do have readily accessible stereotypes about the non-political preferences of average Democrats or Republicans; and they can apply these ideas in the evaluation of political candidates. But when the images are used alongside information about policy content, the impact of the image appears to be marginal at best.

There are a number of limitations that need to be acknowledged. First, our experiments rely on single one-shot exposures. It is possible that the full effects of these cues require repeated and sustained exposure. Our sample sizes are also not large enough to explore heterogeneity in either full-text responses or candidate ratings. We regard both as important avenues for future work. We also see more detailed analysis of full-text responses as a potentially fruitful possibility. This would require questions asking for more lengthy descriptions of party members (rather than just a list of words) across a large sample. This may provide more detailed information about the language used to describe partisans; there may also be important differences in the words used to describe in- and out-partisans. The impact of treatments may also be moderated by political interest or knowledge. However, such analyses require greater sample sizes. In short, our results suggest real potential for further work that explores heterogeneity in the experimental treatments for which we find limited effects above.

Based on the present studies, the background images that appear to make the clearest difference feature guns. Visual cues containing guns – operationalized both through a shooting range image and a gun store image – shift perceptions of a candidate’s partisanship to the right. One explanation may be that guns have become a prominent way of cueing conservatism. It may also be that given the high salience of firearms policy in US politics, gun related imagery cannot truly be considered non-political – it may rather be a very clearly partisan, policy-relevant, cue. The liberal equivalent to guns may be something more explicitly political such as Planned Parenthood. If this is the case, then findings here would seem to indicate that a limited number of deeply partisan and policy-relevant objects and activities can be powerful cues for partisanship, but the large number of objects and activities that are more weakly associated with partisanship, or perhaps not obviously linked in terms of policy, have a limited or even non-existent impact on candidate evaluations, at least when considered in a context in which other policy-oriented information about the candidate is available.

This would align with how we understand mental schemas and the role that political cues play in their activation. Images or symbols that are more powerfully associated with concepts are more likely to trigger the activation of mental frameworks. Thus, the relatively high salience of firearms in the American political landscape may be stimulating mental frameworks around partisanship, producing the changes in candidate evaluations.

We do not regard this as evidence that objects and activities do not cue partisanship in the context of candidate advertising and events so much a sign that we need to better understand the circumstances under which different kinds of cues do and do not matter. There is after all existing work pointing to the potential relevance of photos in the context of candidate evaluations (e.g., Ahler et al., 2017; Banducci et al., 2008; Laustsen & Petersen, 2016). In the absence of other cues, citizens will look for information to help guide their assessment of politicians. Perhaps it is only when other cues do not exist that these more subtle cues matter. Perhaps Instagram pictures, presented with limited commentary, are one context in which the appearance of hybrids, trucks, or organic grocers matters not just to the partisan placements we ascribe to friends, but to politicians as well. Indeed, perhaps it is in that context that audiences are looking for those cues. Our work does not examine this possibility directly, but it highlights the need for work that does.

Our findings also raise questions about the nature of heuristics, and the potential they have to either improve or reduce the quality of citizen decision-making (see Lau & Redlawsk, 2001). Certainly, our view of what is and is not a political heuristic needs to be reconsidered. Put differently: scholars should take seriously the changing nature of the political information environment as we consider how individuals form preferences and evaluations.

For now, our results should provide some comfort to those concerned about the ease with which citizens are swayed by irrelevant, or at least non-policy-focused, information. The fact that voters are readily able to attach partisanship to objects and activities, but yet barely take this information into account when more pertinent political information is available may be good news for representative democracy. Candidates may nevertheless in certain circumstances be able to shift evaluations using simple background cues, and the opportunities for this may be increasing given changes in communication technologies.