Democratic Accountability and Selective Attribution

Democratic accountability is inherently connected to the concept of retrospective voting. Evaluating how well a candidate has done in the past is a heuristic that people can use to help make a cognitively difficult electoral choice (Fiorina Reference Fiorina1981; Woon Reference Woon2012). But the act of evaluating politicians’ past performance can be no less difficult. One key factor is the complexity of government bureaucracy and especially a division of powers across national, state, and municipal leaders, making it hard to know who is responsible when things go wrong and who deserves kudos for policy wins (Thompson Reference Thompson1980). Moreover, sometimes politicians intentionally misrepresent their own role in popular policies; for example, after House Republicans unanimously opposed President Biden’s $1.9 trillion stimulus bill in March of 2021 (which passed and was signed into law), several Republicans leaders nevertheless proceeded to tout the popular bill to their constituents via social media, deliberately attempting to take credit for the legislation that they had, in fact, voted against (Lutz Reference Lutz2021).

Accurately identifying who is responsible for policies is important because it can ultimately guide how citizens vote. For example, Arceneaux and Stein (Reference Arceneaux and Stein2006) find that voters who held their mayor responsible for the aftermath of Tropical Storm Allison became less likely to vote for them in subsequent elections. Indeed, when we consider the meaning of democracy itself—“a system of governance in which rulers are held accountable for their actions in the public realm by citizens” (Schmitter and Karl Reference Schmitter and Karl.1991)—accountability is the sine qua non.

Complicating the difficult task of correctly attributing credit and blame, voters are known for giving their own party members the benefit of the doubt and unduly placing blame on the out-party. When evaluating a policy or politicians is cognitively burdensome, partisan cues are readily available, and the stakes are not especially high, then individuals are more likely to rely on their partisanship to guide their decisions (Lodge and Taber Reference Lodge, Taber., Lupia, McCubbins and Popkin2000). In these cases, it is nearly impossible to evaluate politics evenhandedly. Existing work finds that partisan-motivated reasoning can determine how partisans interpret death statistics in the Iraq War (Gaines et al. Reference Gaines, Kuklinski, Quirk, Peyton and Verkuilen.2007), identify policy positions of the parties (Johnston, Steenbergen, and Lavine Reference Johnston, Steenbergen and Lavine2012), attribute guilt in the case of sexual misconduct (Klar and McCoy Reference Klar and McCoy2021), evaluate the economy (Lewis-Beck, Nadeau, and Elias Reference Lewis-Beck, Nadeau and Elias.2008), and assess the risk and reality of global warming (Joslyn and Haider-Markel Reference Joslyn and Haider-Markel.2014).

Partisans can be motivated to evaluate objective facts more even-handedly, for example with incentives like financial rewards (Bullock et al. Reference Bullock, Gerber, Hill and Huber.2013; Hill Reference Hill2017; Prior, Sood, and Khanna Reference Prior, Sood and Khanna.2015), accuracy scores (Prior, Sood, and Khanna Reference Prior, Sood and Khanna.2015), or even simply an additional request to justify one’s reasoning (Bolsen, Druckman, and Cook Reference Bolsen, Druckman and Cook.2014). But even when partisans do agree on the state of the world—for example, whether the economy is doing well or poorly—they nevertheless can disagree on who is responsible. This phenomenon is referred to as “selective attribution” (Malhotra and Kuo Reference Malhotra and Kuo.2008; Rudolph and Grant Reference Rudolph and Grant.2002). In fact, being forced to accept unflattering facts about one’s political ingroup can actually provide the cognitive dissonance necessary for selective attribution (Bisgaard Reference Bisgaard2019).

Tilley and Hobolt (Reference Tilley and Hobolt.2011) draw an important distinction between policy performance evaluations from responsibility attributions. When confronted with the cognitive dissonance of a poorly performing policy promoted by one’s political ingroup, they suggest that one can resolve that dissonance by either altering one’s evaluation of the facts on the ground (i.e., selective evaluation of the performance of the policy), or one can accept the poor performance of the policy and instead attempt to attribute that poor performance to one’s political outgroup (i.e., selective attribution). Ultimately, they find weak evidence for the former and stronger evidence for the latter. Even non-partisan testimony, as shown in experimental work by Lyons and Jaeger (Reference Lyons and Jaeger.2014), does not mute partisan-motivated reasoning but is rather weaponized by both sides to defend their own party. Tilley and Hobolt (Reference Tilley and Hobolt.2011) argue that different factors (e.g., media exposure, political sophistication) can moderate the degree of responsibility assigned but, ultimately, it is partisanship that tends to determine the direction of attribution (i.e., who gets the blame).

Just as the conditions are specific when it comes to reducing the effect of partisanship on policy performance evaluation, so too are the conditions specific when it comes to reducing the effect of partisanship on selective attribution. Partisans have been shown to attribute less blame to the out-party when partisans are aware that the policy failure was beyond the purview of the out-party office holder (Malhotra and Kuo Reference Malhotra and Kuo.2008). Inversely, partisanship matters more when a disaster is thought to be within the purview of an out-party official (Healy, Kuo, and Malhotra Reference Healy, Kuo and Malhotra.2014). With an experimental study focusing on the government’s inadequate response to Hurricane Katrina, Malhotra and Kuo (Reference Malhotra and Kuo.2008) finds that both Republicans and Democrats are more likely to blame officials of the opposing political party than they are to blame their own, though the effects of a party cue are mitigated among respondents who also received information about the specific offices each official held. However, when partisans are not sure who ought to be responsible, they tend to blame whichever governmental official does not share their party affiliation (Brown Reference Brown2010). We expect that, in the case of COVID-19 policies, we should find support for this earlier work: respondents will be most likely to blame the policies they oppose on officials who do not share their partisanship.

The tendency to follow partisan cues in evaluating decisions is not uniform across the population. Considerable variation can be expected depending on factors including individuals’ preference for their own party over the other. Johnston, Steenbergen, and Lavine (Reference Johnston, Steenbergen and Lavine2012) distinguish “ambivalent voters” from “univalent voters.” The former hold a mix of both positive and negative affect toward both parties and subsequently engage in more even-handed evaluation of a given policy matter. The latter (those who are univalent) hold overwhelmingly positive feelings toward their own party and overwhelmingly negative feelings toward the opposing party. These individuals are most likely to engage in biased processing, even supporting policies they might otherwise oppose simply because their preferred party endorses it.

Affective polarization—the tendency to favor one’s own party relative to the other (often measured with feeling thermometer scales)—produces a similar phenomenon. Individuals who are low in affective polarization are, according to Arceneaux and Vander Wielen (Reference Arceneaux and Vander Wielen2017), most likely to reflect on issues and events. They are—in line with Johnston, Lavine, and Federico (Reference Johnston, Lavine and Federico.2017)’s work—the least biased in how they process political information. Those high in affective polarization, who are least reflective, engaged in more biased processing.

Empirical work demonstrates how affective polarization can indeed predict partisan biases and, conversely, how those lower on the affective polarization scale appear to evaluate politics with greater objectivity. For example, higher affective polarization predicts behaviors that closely follow elite cues even when it comes to public health (Druckman et al. Reference Druckman, Klar, Krupnikov, Levendusky and Ryan.2021); those who are affectively polarized are more likely to bias their judgments regarding sexual misconduct in politics (Klar and McCoy Reference Klar and McCoy2021); affective polarization similarly predicts support for (or opposition to) democratic norms (Kingzette et al. Reference Kingzette, Druckman, Klar, Krupnikov, Levendusky and Ryan2021). Over the past several decades, it appears that Americans overall have grown more affectively polarized (Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). Yet there still remains substantial variation in the degree to which Americans prefer their own party over the opposing one (Krupnikov and Ryan Reference Krupnikov and Ryan.Forthcoming). We thus expect that selective attribution should similarly depend on an individual’s own stated in-party bias.

We next turn to the unique observational setting that allows us to test these hypotheses.

Data and Methods

In response to the outbreak of COVID-19, a variety of preventative measures emerged across some, but not all, 50 states. By late March of 2020, the Center for Disease Control recommended that Americans wear face-masks and maintain distance from one another in order to prevent the spread of COVID. President Trump opted against any type of executive order that might mandate such a policy and instead left it to states to decide whether or not to mandate these precautions. Governors varied in their approaches to these regulations: some mandated masks and restrictions on businesses for their entire state, others left it to local mayors to implement mandates in their cities or towns, and still others signed laws prohibiting such regulations altogether. At the time that we carried out this study (January 2021), 37 states had state-wide mask mandates and the remaining 13 states had left it to local mayors to implement mandates and regulations. Within these 13 states, policies varied widely across towns and cities. This variation across states and municipalities allows us to measure the degree to which respondents attribute policies to their mayors, their governors, or the president himself.

To test our hypotheses regarding the role of partisanship, policy, and affective polarization in selective attribution, we surveyed 833 adults (aged 18 years or older) across the United States. Our sample was recruited by Dynata, an online public opinion research company that recruits large and diverse samples of adults to participate in opt-in online surveys. Our intention is not to infer the distribution of a variable across the broader population, but rather to test the association between multiple variables within a sample in which the covariates of interest are sufficiently distributed (Druckman and Kam Reference Druckman, Kam., Druckman, Green, Kuklinski and Lupia2011). This convenience sample provides excellent variation on these key variables: partisanship, geographic location, and affective polarization. Hill et al. (Reference Hill, Lo, Vavreck and Zaller2007) argue that small bias found in opt-in Internet samples “is not so great as to vitiate the gains of inexpensive, large, and targeted samples that Internet technology make possible” (3) and all survey work, regardless of mode, should acknowledge any bias that might exist in their sample. Most notably, our sample is more white and more educated than the general population. We have no theoretical bases to expect that these differences should distort our findings. Ultimately, using different data sources is what allows us, collectively as researchers, to explore the external validity of a finding—that is, the extent to which the “causal relationship holds over variations in persons, settings, treatments [and timing], and outcomes” (Shadish, Cook, and Campbell Reference Shadish, Cook and Campbell.2002, 83). Our sample is slightly older, more educated, and less racially diverse. An overview of the sample, as well as a comparison with the US census, is provided in our Appendix.

The respondents we reached resided in 724 different towns or cities across the country and in every state, except for Vermont and Wyoming. The geographic distribution of our sample roughly approximates the actual distribution of where the broad American population resides, with more respondents sampled in higher-population states. We display the distribution of where our respondents reside in Figure 1.

Figure 1. Distribution of state residence among respondents.

We asked each respondent for the name of their state and city or town, and for the partisanship of their mayor and their governor. Although we offered a “not sure” option, 94% of respondents were able to tell us if their governor is a Democrat or a Republican. By contrast, only 73% could say whether their mayor is a Democrat or a Republican (a point to which we will return as we contemplate the implications of our results).Footnote ¹

We then asked each respondent whether there was a mask mandate in the city or town where they lived. We provided the following response options: (1) Definitely; (2) Probably; (3) I’m not sure; (4) Probably not; (5) Definitely not. Only 5% of our respondents were “not sure” while the remaining 95% could tell us whether there was or was not a mask mandate. Next, we asked respondents whether they believe that there should be a mask mandate, with the response options: (1) Definitely; (2) Probably; (3) I’m not sure; (4) Probably not; (5) Definitely not. Finally, we asked them a series of questions about who is responsible for the fact that there is, or is not, a mask mandate in their city or town. For those who told us they have a mandate, we asked: “To what extent is your mayor responsible for the fact that there is a mask mandate in the city or town where you live?” We then asked this same question about their governor and about then-sitting President Trump. For those who told us that they do not have a mask mandate, we asked: “To what extent is your mayor responsible for the fact that there is not a mask mandate in the city or town where you live?” We then asked this same question about their governor and about President Trump.

We then repeated this battery of questions, but instead of mask mandates, we asked about restrictions on businesses. We learned whether business restrictions existed in each respondent’s city or town, whether they themselves support business restrictions, and the degree to which they hold their mayor, governor, and president responsible for the fact that there are, or are not, business restrictions where they live. Finally, we asked a series of demographic questions, including partisan identification, as well as thermometer ratings for both parties (in order to measure their affective polarization score). Each of these questions, in addition to the demographic questions we asked, is listed in our Appendix.

Collectively, these questions provide us with all of the information necessary to test our hypotheses. For testing Hypothesis 1, we used information from the respondent about whether there is a mask mandate and business restrictions in their town and whether the respondent opposes their local policies. We subtracted their support for the policy from their certainty that the policy exists, creating a 5-point variable ranging from 0 to 4, where 0 indicates the most support for their local policy and a 4 indicates the most opposition to their local policy. For example, if a respondent is certain the policy exists in their town and they strongly oppose such a policy, they would score a 4. If a respondent is certain that the policy does not exist in their town and they strongly support such a policy, they would also score a 4. A score of 0 would mean that they are certain the policy exists and they fully support it, or that they are certain that the policy does not exist and they fully oppose such a policy. Midpoints on the scale (i.e., scores of 1–3) reflect variation in the strength of their opposition to the local policy. For both the mask mandates and business restrictions, a slight majority fully supported their local policy (60% for mask mandates and 58% for business restrictions). The remaining 40% (for the mask mandate) and 42% (for business restrictions) expressed some degree of opposition.

We next constructed variables to indicate whether a respondent shares the same party identification as their mayor, governor, and president (who, for all respondents, was Trump). We classify leaners as partisans and create a binary variable in each case: 0 if the candidate is from the out-party and 1 if they share the same partisanship.

To test Hypothesis 2, we constructed the standard measure of affective polarization by subtracting respondents’ out-party ratings from their in-party ratings. The thermometer rating ranges from 1 to 10, and so a score of 9 would mean that a respondent expressed the greatest warmth for their own party (10) and the lowest for the out-party (1). The distribution of affective polarization scores is shown in Figure 2. The negative values indicate that just over 1% of respondents reported to favor the out-party over the in-party. We keep these outliers in the dataset and include them in our analyses, but all of the results we present in this paper are identical when we exclude these outliers and instead scale the measure from 0 to 1.

Figure 2. Distribution of in-party bias (also known as affective polarization) among respondents.

We test both hypotheses with one single model, which contains the variables and interactions we require for all of our tests. In this manuscript, we plot all of our results. In our Appendix, we include all of the output from the regression tables. We next turn to our analyses and results. We examine two distinct policies in this study: mask mandates and business closures. We present results for both of these policies as we move sequentially through our hypotheses.

What Makes Mayors Different?

We find that, as partisans increasingly oppose a policy, their tendency to blame out-partisan mayors grows. Yet, contrary to our second hypothesis, this is not exacerbated by affective polarization. In other words, personal feelings of dislike toward the out-party exacerbate selective attribution only against governors and the president, but not against the mayor. In this brief section, we consider why this might be. Druckman et al. (Reference Druckman, Klar, Krupnikov, Levendusky and Ryan.2021) show that affective polarization increases partisan-motivated reasoning only when the target is sufficiently politicized. It seems, with the benefit of hindsight, that mayors are not sufficiently politicized. This is evident in our own data, as well as from existing research. Nearly 27% of our respondents could not identify the partisanship of their mayor. This conforms with a dataset of mayoral elections compiled by de Benedictis-Kessner and Warshaw (Reference de Benedictis-Kessner and Warshaw.2016); only 66% of the mayoral elections they examine occurred between two candidates who can be identified as a Democrat or a Republican. Recently, Das et al. (Reference Das, Sinclair, Webster and Yan2022) examined the political rhetoric used on Twitter by different levels of government. They find that political rhetoric from governors tends to align with national politics and parties, whereas mayors are largely focused on local issues and remain relatively distant from national politics. Das and coauthors conclude that, “unlike elected officials at other levels of government, the average American mayor currently appears to be unaffected by the nationalization of American politics” (28). Given the lesson from Druckman et al. (Reference Druckman, Klar, Krupnikov, Levendusky and Ryan.2021) that politicization is a perquisite for affective polarization to impact policy attitudes, we can speculate that mayors are seemingly insufficiently politicized.

Scholars have debated the extent to which mayors’ partisanship impacts even their own policy-making. Gerber and Hopkins (Reference Gerber and Hopkins.2011) found that mayoral partisanship affected the budget they allocate to public safety, but had no impact on tax policy, social policy, or other areas with overlapping authority. Work by de Benedictis-Kessner and Warshaw (Reference de Benedictis-Kessner and Warshaw.2016), though, demonstrates a greater impact of a mayor’s partisanship: Democrats, they show, increase overall city expenditures in a variety of areas mostly through increasing city debt rather than by increasing taxes. This might be a trend that is shifting, though; Hopkins argues that mayors (and other local officials) are becoming more closely tied to national politicians, and mayoral votes are increasingly cast “with an eye toward national politics” (19). If this is indeed the case, then one might expect that affective polarization becomes relevant for mayoral politics in the future. This is certainly a question for future researchers to revisit.