Introduction

Democracy relies on citizens holding governments to account at the ballot box for the outcomes of their policies (Fiorina, 1981; Key, 1966; Powell, 2000). Yet research has repeatedly cast doubt on the ability of voters to objectively evaluate government performance on important issues. One frequently raised concern is that many voters have a strong attachment to a particular political party, a ‘partisan identity’, and that this leads to biased assessments of real-world conditions (Anderson, 2007; Green et al., 2002; Sniderman et al., 1991). When outcomes are hard to dispute, government and opposition partisans can agree about the current state of affairs but attribute responsibility for the outcomes to different actors (Bisgaard, 2015; Tilley & Hobolt, 2011). In this paper, we use panel data analysis combined with an experimental approach to examine whether these consequences of partisan bias are present in the context of a highly salient non-economic issue; the UK government’s handling of the Covid-19 pandemic, which was noticeably less party polarized than in the more commonly studied US context. We also use the opportunity to test a new potential consequence of partisan reasoning; that partisans may exhibit biased recall of the government’s past performance.

The Covid-19 pandemic provides an excellent opportunity for the study of partisan motivated reasoning and its effects on evaluations of government competency.Footnote 1 Perceiving real-world conditions in line with partisan leanings is to be expected when issues are of low salience to voters and signals about the government’s performance are mixed. The Covid-19 pandemic, however, was highly salient and performance signals were relatively clear. People cared about the government’s handling of the pandemic for over two years,Footnote 2 and the media provided constant coverage and offered continual information about simple indicators like death and vaccination rates throughout (Nielsen et al., 2020). Moreover, Covid-19 handling in the UK was not strongly polarized by party: the government was unchanged throughout the pandemic and the main parties were relatively consensual in their positions on the measures adopted to address the pandemic and implementation of vaccination programs (Klymak & Vlandas, 2022). This contrasts with the polarization over Covid-19 seen in, for example, the United States (Rodriguez et al., 2020). In this respect, Gadarian et al (2022) provide substantial evidence that Donald Trump’s administration tied the pandemic to the president’s political fate, emphasizing partisanship over public health, while Democrats depicted the crisis as evidence of Trump’s lack of concern with public well-being.Footnote 3 In contrast, the comparatively de-polarized UK pandemic provides a particularly tough test of partisan motivated reasoning and the thesis that partisans view the world through a perceptual lens (Achen & Bartels, 2016; Campbell et al., 1960; Green et al., 2002). In this paper we assess whether that lens is strong enough to distort perceptions of the worst global health crisis of recent decades.

In this context, the UK government is widely acknowledged to have performed both extremely badly on some aspects of pandemic management and extremely well on other aspects, thereby creating two strong but opposing performance signals. This allows us to prime respondents to focus on a positive or a negative aspect of handling without lying or distorting reality, increasing the external validity of our results as well as allowing causal inference about the impact of partisanship. As a pre-pandemic wave of our survey was fielded at the time of the UK’s last general election, in December 2019, it further allows us to incorporate pre-Covid partisanship into our models and thus to provide a temporal basis for inferring possible causal impact.

In addition to testing whether partisans display bias when evaluating performance and attributing responsibility for outcomes, we also examine an additional and novel way in which partisan voters may draw different conclusions about the competence of the governing party. Existing literature shows that motivated reasoning can affect the accuracy of a person’s memory (Greene et al., 2020; Murphy et al., 2019). We build on these findings by hypothesizing that government partisans will selectively recall positive aspects of the government’s past performance, whilst opposition partisans focus on more negative aspects. These hypotheses have not been tested in previous research.

Overall, our results show that voters assess government performance in a partisan-biased manner. Panel data analysis provides evidence for almost all of our proposed mechanisms. Government partisans are more positive about the UK’s coronavirus performance than opposition partisans, slightly less likely to attribute responsibility to the government for the pandemic overall, and more positive when recalling the government’s handling of the pandemic a year before our experiment compared to how they perceived it at the time. In contrast, when recalling past performance, opposition partisans are significantly less likely than government partisans to think about a positive aspect of crisis management.

Our experimental results mostly support our panel data findings. We find evidence that government partisans focus on positive aspects of the crisis when forming evaluations of the UK’s performance overall, whilst opposition partisans are less likely to consider the success of the UK’s vaccination program unless explicitly forced to do so. We also find that government partisans are less likely to attribute responsibility for the UK’s pandemic experience to the government even when reminded of the UK’s high death toll.

Theory and Hypotheses

People are not only motivated to reach accurate conclusions, but also to reach conclusions that accord with their prior opinions (Mercier & Sperber, 2017). This process of (directionally) ‘motivated reasoning’ (Kunda, 1990) results in a range of cognitive biases that are frequently observed when voters reason about politics (Flynn, Nyhan, and Reifler 2017; Leeper & Slothuus, 2014; Lodge & Taber, 2013; Taber & Lodge, 2006).

The role of motivated reasoning in politics is most clearly apparent in the case of partisanship. Rather than acting as a ‘running tally’ of party performances (Fiorina, 1981), partisan identity can create a ‘perceptual screen’ through which voters see different realities (Butler & Stokes, 1969; Campbell et al., 1960). In particular, partisans tend to view conditions more favorably when their party is in power (‘selective evaluation’), and to attribute more responsibility for positive outcomes and less responsibility for negative outcomes to the government when their party is in power (‘selective attribution’). As we shall elaborate below, there are also reasons to expect partisans to recall the government’s past performance differently according to whether or not their party held power. These processes allow voters to reduce the cognitive dissonance that results from supporting a political party which is failing to deliver desirable outcomes.

The tendency to twist new information in service to prior opinions is, however, bounded (Lebo & Cassino, 2007; Redlawsk, Civettini, and Emmerson 2010). Even committed partisans acknowledge economic reality when conditions are extreme or signals about performance are particularly clear. The case of Covid-19 allows us to test the limits of partisan reasoning during an extreme and highly salient health crisis, such as occurred in response to the economic crisis of 2008–09 when the endogenous nature of economic perceptions was mitigated by the strength of the signals resulting from the financial crisis (Chzhen, Evans, and Pickup 2014; Parker-Stephen, 2013). For partisanship to affect evaluations of the government during the Covid-19 crisis, and in a relatively non-polarized political environment such as that in the UK, would be strong evidence of the resilience of partisan biases.

Selective Evaluation

The simplest way for partisans to avoid confronting harsh truths about the failures of their favored party or the successes of other parties is to ignore these truths altogether. For example, when inflation rates are worryingly high, government partisans might focus on the low level of unemployment and therefore judge that the economy is performing well.

Existing literature has established that partisans view conditions more positively when their favored party holds office than when it is in opposition. In the US, VanDusky-Allen et al. (2022) find that during the pandemic onset period, Americans typically rated their state governments’ responses more favorably if their governor was a co-partisan. But even in the UK, panel data analysis of voters in the 1990s showed clearly that socio-tropic perceptions of the economy are themselves affected by prior opinions about the incumbent party and by party choice at the last election (Anderson, Mendes, and Tverdova 2004; Evans & Andersen, 2006), as are egocentric economic evaluations (Johnston et al., 2005). The causal relationship between vote choice and economic perceptions extends beyond the UK (Wlezien et al., 1997), as does the persistent effect of partisanship on perceptions of not just the economy (Wilcox & Wlezien, 1993) but also foreign policy (Bartels, 2002). Accordingly, we expect to find descriptive partisan differences in how respondents view the UK’s experience of the Covid-19 pandemic. We anticipate that government partisans will be more positive about the UK’s coronavirus performance than opposition partisans, because they are motivated to believe that the government has handled the crisis successfully.

We expect that government partisans will have arrived at a positive view of the UK’s performance, in part, by already incorporating positive information like the vaccine rollout success into their perceptions. Opposition partisans, in contrast, are expected to have already incorporated negative information like the high death toll. Accordingly, we expect our negative treatment to have a bigger effect on government partisans, who will avoid thinking about negative aspects of the crisis management unless prompted to do so. Likewise, we expect our positive treatment to have a bigger effect on opposition partisans, who will avoid thinking about positive aspects of the government’s crisis management unless prompted to do so:

H1: Government partisans who are reminded about a negative aspect of the UK’s experience of coronavirus are made more negative about the UK’s overall pandemic performance compared to opposition partisans (and conversely for opposition partisans with regards to the positive treatment).

Selective Attribution

When faced with uncomfortable and undeniable facts about reality, there remain a number of possible options for engaging in motivated reasoning (Gaines et al., 2007). One of these options is selective attribution. Voters who hold a partisan identify may accept objective facts about policy outcomes like the state of the country’s economy or health service, but choose to attribute responsibility for these outcomes differently according to whether their favored party holds power. Multiple studies have found evidence that partisans engage in this form of motivated reasoning, raising concerns about democratic accountability (Gomez and Wilson 2003; Rudolph and Grant 2002). The opportunities to engage in selective attribution are expanded by the fact that attributing responsibility for policy outcomes is often difficult even for a non-partisan voter (Anderson, 2000; Hobolt, Tilley, and Banducci 2013; Powell & Whitten, 1993), particularly given the clear political incentives for parties to intentionally blur the lines of responsibility (Hellwig, 2012; Hobolt & Tilley, 2014).

Previous research has shown that partisans in the United States are more likely to attribute responsibility for good economic outcomes and less likely to attribute blame for bad economic outcomes to officials from their favored party (Brown, 2010; Rudolph, 2003a, 2003b, 2006). They are also more likely to selectively blame officials for the handling of natural disasters (Healy et al., 2014; Malhotra & Kuo, 2008), foreign policy (Nawara, 2015; Sirin & Villalobos, 2011) and health care (McCabe, 2016). Most recently, Graham and Singh (2023) show that selective attribution also applied to understandings of the Covid-19 pandemic, with partisans “disproportionately crediting their party for positive developments and blaming opponents for negative developments”.

Far less research has considered the UK context, but some studies have found that the findings from the U.S. extend quite well to the UK. UK partisans are more likely to blame the government for negative economic outcomes when their party does not hold power (Bisgaard, 2015; Marsh & Tilley, 2009), and similar effects have been found for health care outcomes (Tilley & Hobolt, 2011). Most recently with respect to the latter, Yeandle and Maxia (2023) examined the impact of emphasizing the role of the National Health Service in the UK’s Covid-19 vaccination program on attributions of responsibility to the government. They found that respondents attributed less responsibility to the government, but this was not associated with a change in government approval ratings, indicating limitation in the role of attributions for approval.

Our expectations follow the findings from this existing literature. We anticipate that government partisans will attribute less responsibility for negative aspects of crisis management than opposition partisans, and more responsibility for positive aspects. Accordingly, in our experiment, we anticipate that prompting respondents to think about the UK’s high death toll will result in government partisans affording less responsibility for the pandemic overall to the government compared to opposition partisans. We expect the exact reverse when we prompt respondents with details of the UK’s vaccine success:

H2: Government partisans who are reminded about a negative aspect of the UK’s experience of coronavirus attribute less responsibility to the government for the handling of the pandemic overall compared to opposition partisans (and conversely when reminded about a positive aspect).

Selective Recall

Far less research has investigated the effect of partisanship on memory of political events. Literature from psychology shows that human memories are frequently distorted or even patently false (for an overview, see: Koriat, Goldsmith, and Pansky 2000). There is also a clear neurological link between motivated reasoning and memory (Bavel and Pereira 2018). Given that partisans are willing to selectively evaluate present conditions differently according to whether their favored party holds power, it stands to reason that they are also likely to evaluate past performance differently according to whether their party held power at that time. When attempting to reach a motivated conclusion about the competence of the governing party, individuals may selectively recall or ignore elements of the party’s past performance. For example, a government partisan may forget about some of the government’s past failings and instead conclude that the government was performing well, even if they thought the government was performing badly at the time. Despite the intuition of this possibility, we are not aware of any studies directly testing this potential consequence of partisan bias, which we term ‘selective recall’.Footnote 4

Very few articles come close to testing the type of selective recall examined in this paper. Castelli and Carraro (2011) used an experimental approach to show that ideologically conservative participants were more likely to recall negative facts about immigrants than ideological liberal participants. It seems reasonable to expect similar differences in memory according to partisanship. And Jacobson (2010) used panel data to show that perceptions of the Iraq war in the U.S. changed over time in line with partisan identity, with respondents falsely recalling their earlier opinions. For example, many Democrats forgot that they had once believed Iraq to possess weapons of mass destruction (Jacobson, 2010), which mirrors a similar amnesia with respect to the Gulf War of the early 1990s, where Zaller (1994) found that voters had forgotten the partisan divisions preceding the conflict and believed that there had always been uniform, bipartisan support for expelling Saddam Hussein from Kuwait.

Our paper builds on these findings by considering whether past memories about government performance can be affected by partisanship. The question of whether voters can accurately recall past performance, or whether this is biased by partisan identity, is important for understanding the relationship between time and the limits of democratic accountability.

We expect that partisan bias will result in government partisans recalling the government’s performance a year ago more favorably than opposition partisans. The explanation for this could of course simply be that government partisans were more positive about their handling of the pandemic at the time. In addition, however, we also anticipate that partisan bias actually distorts recollections of the past. Accordingly, we expect that government partisans will recall the government’s past handling more positively than they perceived it at the time, and conversely for opposition partisans.

The motivated reasoning account of this outcome is that people selectively mis-remember their evaluations from the previous year in order to make them consistent with their current beliefs. However, it could also reflect partisan biases in updating. Thus government partisans incorporate new information (e.g. that the vaccine roll out has gone well, so the government were probably fairly effective in preparing the ground for this) whilst opposition partisans incorporate negative information from the present (e.g. long covid rates are high, so perhaps the government was even poorer at handling the crisis than it seemed at the time). So, voters project backwards on the basis of partisan-biased updating procedures.

Whichever of these interpretations is most accurate, we expect this mechanism of selective recall to operate similarly to selective evaluation; government partisans typically avoid thinking about the negative aspects of past performance when evaluating earlier government competence. We therefore anticipate that reminding government partisans of the UK’s high pandemic death toll will have a bigger negative effect than reminding opposition partisans of this fact (opposition partisans are likely to think about the death toll even without a prompt to do so) and conversely for the vaccine success:

H3: Government partisans who are reminded about a negative aspect of the UK’s experience of coronavirus will become more negative about the government’s past handling compared to opposition partisans (and conversely for a positive aspect).

Analytical Strategy

To test our hypotheses, we employ both panel data analysis and a survey experiment. Our data is from the British Election Study Internet Panel (BESIP), and we make use of three waves (19–21) from this survey (Fieldhouse et al., 2021). Wave 19 was fielded between 13th and 23rd December 2019, wave 20 between 3rd and 22nd June 2020, and wave 21 between 7 and 25th May 2021. Our experiment was fielded at the end of the full May 2021 survey wave, in England and Wales, with a sample comprising 6884 respondents who were randomly selected from the full panel (n = 30,821). For tables showing the distribution of the sample across a range of demographic and non-demographic variables, see Appendix 3 of the Supplementary Material.

The timing of these survey waves allows us to exploit exogenous variation in the British experience of the Covid-19 pandemic (see Fig. 1). Covid-19 first reached Europe in early 2020, just after a UK general election that gave then Prime Minister Johnson a large majority in the House of Commons. Yet by May 2020, the UK government was perceived by many as too slow to react to the pandemic, resulting in comparatively high infection and death rates as the virus spread fast during a period of very few restrictions on social activity. By June 2020, when the BESIP wave 20 was in the field, UK citizens had endured multiple weeks of ‘lockdown’, with a police-enforced ‘stay at home’ order that prohibited leaving one’s home except to shop for necessities or engage in one hour of daily exercise. Public approval of the government’s handling of the crisis had dropped from 60% in April to around 40% (Gov.UK, 2023; Yougov, 2023).

Fig. 1
figure 1

Source: Institute for Government Analysis, BBC news, and the BESIP codebook

Timeline of data collection.

Full size image

By May 2021, the time at which we fielded our experiment, the UK situation had dramatically reversed. Though there had been additional (less extreme) lockdowns in the 11 interceding months, May 2021 was a period of relatively few restrictions, and a time at which the UK’s vaccination program was proving extremely successful. Vaccination rates outstripped those seen across Europe, partly due to the government’s successful procurement program, and the UK’s pandemic performance had received praise even from traditionally anti-Conservative media outlets.Footnote 5 Handling ratings had improved accordingly (Yougov, 2023).

Our panel data analysis uses this exogenous variation to examine evaluations of the government’s handling of Covid-19 as a product of pre-Covid-19 evaluations. Our experiment exploits the variation by priming respondents to think about negative and positive aspects of the crisis without lying or distorting reality; we simply call attention to either Britain’s initial struggle with the virus or to the later success that came from the vaccine program. It is also worth noting that early evidence suggests that voters differ substantially in their evaluation of the government’s performance but tend to agree on what the goal should be (Green et al., 2020), which avoids the conflation of performance evaluations with ideological preferences.

The panel data analysis consists of examining the relationship between pre-pandemic partisanship and evaluations of the government’s handling of the pandemic two years later. On the question of recalled past handling, we also make use of the 2021 survey wave to compare respondents’ recollections of pandemic handling ‘this time last year’ to how they actually responded at the time. Over half (n = 3796) of the full sample had participated in the June 2020 wave of the BESIP, which was fielded one year before our experiment, and 3784 had participated in the December 2019 wave of the BESIP, providing a reasonable sample size for panel data analysis.

To corroborate the findings of our panel data analysis we also make use of an experimental approach. We primed respondents to think about a positive or negative aspect of the UK’s Covid-19 performance, either the successful vaccine program or the high pre-vaccine death toll, and observed how this treatment affected responses to questions about the government’s handling of the pandemic.

Experimental Treatment

We randomly assigned respondents to one of two treatment conditions or to a control group.Footnote 6 For experimental treatments we used short vignettes that highlighted either a negative or a positive aspect of the UK’s experience with the Covid-19 pandemic. By comparing the responses of those in the control group (who received no vignette) to those who did receive a vignette, we can infer whether respondents would have been thinking about the points raised in our vignettes even had we not prompted them first.

The exact wording was as follows:

Negative performance reminder:

“Before the rollout of the coronavirus vaccine, the UK had one of the highest coronavirus death tolls per capita in the entire world. Over 125,000 Britons have died after contracting the virus.”

Positive performance reminder:

“Well over half of the UK adult population have already received their first dose of the Covid-19 vaccine, and the UK continues to have one of the best vaccination rates in the entire world.”

Dependent Variables

After receiving treatment in the form of vignettes, respondents were asked a number of questions that allow us to test our hypotheses about selective evaluation, attribution and recall. First, to test for selective evaluation we asked: “How well has the UK performed overall in dealing with the coronavirus crisis?”. Answers were given on a five-point scale from ‘very well’ to ‘very badly’, with a ‘don’t know’ option available. This dependent variable also acts as a useful manipulation check, allowing us to assess whether respondents were affected by our treatments at all.

We then tested selective attribution by asking three questions about the government’s responsibility for crisis management. Respondents were asked: “To what extent are the following the result of decisions taken by the UK government?” and asked to rate responsibility for three outcomes (in a randomized order) on a scale from 0 (“Not at all due to government decisions”) to 10 (“Entirely due to government decisions”), again with an option for “Don’t know”. The three outcomes we asked about were “The UK’s overall experience of the coronavirus crisis”, “The UK’s high coronavirus death toll” and “The fast pace of the UK’s vaccine rollout”.

To test our novel theoretical expectations about retrospective evaluations of distant past performance, we asked respondents: “Thinking back to this time last year, how well do you think the UK Government had handled the coronavirus outbreak in Britain?”.Footnote 7 This was measured on a 5-point response scale ranging from ‘Very well’ to ‘Very badly’ in order to match the question about the UK’s overall performance and matches the question in the BESIP survey for the June 2020 wave in which respondents were asked: “How well do you think the UK Government has handled the coronavirus outbreak in Britain?”.Footnote 8

Independent Variables

Partisan identity is our main independent variable of interest, which we interact with treatment status to see whether government partisans responded differently to our vignettes compared to opposition partisans. We make use of the standard BESIP question for partisan identity: “Generally speaking, do you think of yourself as Labour, Conservative, Liberal Democrat or what?”, coding all Conservative partisans as ‘government partisans’ and all those expressing any other partisan identity as opposition partisans. This was measured prior to our experiment in the main survey. Of those who also participated in the June 2020 wave of the BESIP, we were able to check whether their partisan identity had changed during the pandemic; just 3% of the sample switched from the Conservatives to another party, or vice versa, and the results reported below remove these switchers from the sample. We also remove the 1% of the sample who were Brexit party identifiers because they are not ‘government partisans’ but cannot really be considered ‘opposition partisans’ given the ideological overlap between the Brexit and Conservative parties. The effect of not removing these individuals can be seen in Appendix 5 of the Supplementary Material, as can the effect of restricting the analysis to Conservative vs Labour, while excluding all other opposition party supporters. The main coefficients are not substantively affected in either case.Footnote 9

Method

We analyze our data with ordinary least squares regression models, with each dependent variable modelled in turn as a function of treatment status. For hypotheses concerning differences of treatment effect by partisan identity we also include the interaction of treatment status with these variables. In the case of our evaluation measures (overall performance and retrospective evaluation) we also re-ran the analysis using ordered logit models—the results of this robustness check were not substantively different from using OLS and can be found in Appendix 5 of the Supplementary Material. The use of randomized allocation to treatment theoretically removes the need to include control variables, since demographics do not differ systematically according to treatment status, but demographic variables can help to increase the precision of other coefficient estimates. Accordingly, all models reported below include full demographic controls for gender, ethnicity, age, class and educational attainment. Full details of how these variables are coded can be found in Appendix 1 of the Supplementary Material.

Surveys typically under-represent individuals with low political attention, which may bias our experimental results if these individuals differ in response to treatment compared to individuals with high political attention. Accordingly, we adapt the weighting variable for our analysis to match the levels of self-reported political attention in our experimental sample to those found in the random probability British Election Study (BES). An additional concern is that our vignette praising the success of the UK’s vaccine program might have an opposite effect on those individuals who oppose vaccination on ideological grounds. All reported models therefore omit those individuals who indicated earlier in the survey that they are not generally in favor of vaccinations (3% of the sample).

The effect of including these individuals and of not weighting by political attention can be found in Appendix 5 of the Supplementary Material. The model specification has no systematic effect on coefficient sizes or standard errors and does not affect our substantive conclusions.

Results—Panel Data Analysis

Figure 2 shows the main results of the panel data analysis. The models show clearly that there are partisan differences across almost all dependent variables. In line with our expectations, individuals who were government partisans before the pandemic were more positive about the government’s handling of the pandemic by May 2021 compared to those respondents who were opposition partisans in 2019. Government partisans were less likely than opposition partisans to attribute responsibility for the death toll to the government, but more likely to attribute responsibility for the success of the vaccine rollout. Those without any partisan identity, meanwhile, sat between both of these extremes.

Fig. 2
figure 2

Average predicted position for government and opposition partisans across all six dependent variables. Note: Triangular points indicate that there is a significant difference in the predicted response between government and opposition partisans at the 95% level. The ‘Change in retro handle’ model only includes those individuals who were also surveyed in the pre-experiment wave one year prior to the experiment. The predicted positions displayed in this figure are calculated from a model that also includes controls for gender, ethnicity, age, class and education. The ‘whiskers’ shown reflect 95% confidence intervals. The full results of these models can be found in Appendix 4 of the Supplementary Material

Full size image

Furthermore, we find that government partisans were more positive when recalling the government’s performance than opposition partisans. We also find evidence that this reflects genuine bias among the government partisans, who are actually more positive than they recalled at the time. Surprisingly, and contrary to the expectations, opposition partisans are also significantly more positive about the government’s handling of the pandemic one year ago than they stated at the time, as are non-partisans.

To give some substantive context to how extreme this last result is, we considered the relationship between current and past evaluations of the UK’s coronavirus experience, and respondents’ recollections of the government’s past handling. As Table 1 shows, there is a greater association between evaluations of the UK’s current coronavirus experience and recall of the government’s earlier handling than there is with individuals’ actual evaluations of the government’s handling at the time.

Table 1 Association between evaluations of the UK’s current Covid-19 performance and evaluations of the performance one year ago
Full size table

Furthermore, government partisans rely significantly more on present evaluations when forming their opinions about the past than opposition partisans. This adds further evidence to our suggestion that partisanship affects the way in which individuals recall the past, with government partisans extrapolating back from the sunny present and forgetting about their actual feelings concerning the government’s performance one year ago.

Results—Experiment

Our first dependent variable, perceptions of the UK’s pandemic performance overall, can be used as a manipulation check for our treatments. As Table 2 shows, our treatments had the desired effect. Those individuals who were informed (or reminded) about the UK’s high death toll were significantly more negative about the UK’s experience of the coronavirus pandemic overall than the control group receiving no vignette. Those who were informed (or reminded) about the success of the UK’s vaccination program were significantly more positive about the UK’s experience of the pandemic than the control group.

Table 2 Percentage breakdown of UK overall performance by treatment condition
Full size table

The results of our main analysis, testing whether our successful manipulation affected government and opposition partisans differently, can be found in Fig. 3. The figure shows the marginal effect of our treatment conditions on each of our dependent variables, separated by partisanship. Triangular points indicate that the interaction term was significant in the model, providing evidence at the 95% level that government and opposition partisans reacted differently to our treatment vignettes.

Fig. 3
figure 3

Average marginal treatment effects for government and opposition partisans for all six dependent variables. Note: Triangular points indicate that the interaction term is significant at the 95% level, i.e., that government partisans differ significantly from opposition partisans in response to that treatment condition for that dependent variable. The ‘Change in retro handle’ model only includes those individuals who were also surveyed in the pre-experiment wave one year prior to the experiment. The marginal effects displayed in this figure are calculated from a model that also includes controls for gender, ethnicity, age, class and education. The ‘whiskers’ shown reflect 95% confidence intervals. The full results of these models can be found in Appendix 4 of the Supplementary Material

Full size image

With regards to our hypotheses about selective evaluation, we find partial support for H1. In line with our expectations, opposition partisans were significantly more affected by a reminder about the UK’s vaccine success than government partisans, indicating that they were less likely than government partisans to recall the vaccine success when evaluating the UK’s overall experience of coronavirus unless explicitly prompted to do so. However, the negative treatment was equally effective at dampening evaluations of the UK’s performance for government and opposition partisans, and so does not provide further evidence for H1. In line with the panel data analysis, non-partisans lay between these two extremes, being more affected by the positive treatment than government partisans but less so than opposition partisans.

Turning to responsibility attribution, we find the reverse of our findings on performance evaluations, with a significant partisan difference for the effect of the negative treatment but not for the positive treatment. In line with H2, we find that government partisans attribute significantly less responsibility to the government for the UK’s pandemic performance overall when they are first reminded about the UK’s high coronavirus death toll. In this respect they differ significantly from opposition partisans. However, the positive treatment has no effect on either government or opposition partisans, nor for non-partisans.

For our final dependent variable, retrospective handling evaluations, we find no significant partisan differences in the effect of the negative treatment but a significant difference for the positive treatment. When reminding respondents about the UK’s high death toll, we do not find a significant difference in the resulting evaluations of past handling compared to the control group who received no reminder. This is true for both government and opposition partisans. Somewhat surprisingly, we find that non-partisans became slightly more negative about the UK’s past performance when exposed to either the positive or the negative treatment. We did not put forward any expectations on this front, nor have any explanations post-hoc for why this might be the case. Future research into political recall may shed further light on this apparently counter-intuitive finding.

There is some evidence that, for those respondents who were surveyed a year before the experiment as well as during the experiment, government partisans became slightly more negative about the past performance compared to how they felt at the time. Yet this does not significantly differ from opposition partisans.

We do find a significant interaction effect in the case of the positive treatment, as hypothesized in H3. Opposition partisans reminded about a positive aspect of the UK’s coronavirus handling became significantly more positive about the government’s response to the pandemic a year prior to the experiment. Government partisans, by contrast, appear to have become more negative, though this is not quite significant in the model for the full sample. When focusing on those who took the survey one year before the experiment, we also find a significant interaction effect and, in this case, we find that government partisans reminded about the vaccine success actually became more negative about the government’s past performance compared to how they rated it at the time.

Discussion

The results in this paper offer an important confirmation that government evaluations are heavily endogenous to partisan identity. The findings, derived from panel data analysis corroborated by a survey experiment, suggest that partisans evaluate performance, past performance and responsibility in a manner that reflects positively on their favored party, even in the context of a highly salient health crisis like the Covid-19 pandemic.

This article provides an important update to a body of research that has tended to be limited either in scope or by a lack of external validity. Much scholarship on the topic of partisan bias focusses exclusively on the consequences for economic evaluations, often in the United States. This leaves open questions about whether partisan bias extends beyond the economy to other issues, and about how well the results replicate to less politically polarized countries. Studies that attempt to move beyond the economy or the United States, however, tend to struggle with providing experimental treatments that are realistic. For example, experimenters sometimes present respondents with hypothetical scenarios or non-factual prompts, such as claiming to respondents in the treatment group that ‘experts say that… the British economy is doing considerably worse than most other countries’ (Tilley & Hobolt, 2011). There is good reason to doubt whether voters react to real-world information in the same way that they react to these kinds of hypothetical vignettes. Alternatively, some papers focus on natural disasters like hurricanes or floods, which have a short-term saliency and are often quite localized (Malhotra & Kuo, 2008). Our study builds on existing research by addressing both of these shortcomings—we manipulated factually accurate information about a real-world and enduringly salient non-economic crisis.

It is also worth noting that our experimental set-up provided a very tough test of the mechanisms of partisan motivated reasoning established in the literature. Covid-19 was a highly salient event, covered extensively in the national media, with easy to grasp figures like death rates that could be compared to other countries by ordinary voters.Footnote 10 In such a context, finding evidence of motivated reasoning testifies to the strength of the partisan perceptual screen. It also fits with the findings of Becher et al (2023), who examined self-selection biases in benchmarking of media headlines in France, Germany and the UK, showing that respondents sought out information that was consistent with their prior attitude towards the government, even without the use of partisan labels.

It is possible moreover that our experiment setup may have downwards-biased our estimates of motivated reasoning. We fielded the experiment at the end of the May 2021 wave of the British Election Study Internet Panel, meaning that our sample comprised relatively knowledgeable and politically interested respondents who had just answered a number of questions about politics and Covid-19. Accordingly, our treatment vignettes might have been expected to have no detectable effect on these respondents. The fact that we detected changes in the attitudes of our respondents after treatment, and that these changes fell along predictable patterns of partisanship, further emphasizes the robustness of the partisan perceptual screen.

One interesting nuance of our findings is that the effect of our negative and positive prompts was asymmetric. In the case of selective evaluation, our positive prime affected opposition respondents more than government respondents, whilst the negative prime affected both equally. In contrast, our negative treatment affected opposition partisans significantly more for a question of overall responsibility, but the positive treatment was equally ineffective at changing attributions of responsibility for opposition or government partisans. One possible explanation for this is that respondents reacted to our primes by engaging in one of two forms of motivated reasoning. When reminded of a positive, they chose to update their overall evaluation, but when reminded of a negative they instead changed their attribution of responsibility. In neither case was there a need for partisans to both selectively evaluate performance and to selectively attribute responsibility. This accords with findings from previous literature that partisans need only engage in one form of motivated reasoning to arrive at a biased conclusion, for example by attributing responsibility differently only when forced to confront the fact that conditions are getting worse (Bisgaard, 2015). However, more work is needed to establish whether the specific positive / negative asymmetry we found in our paper generalizes to other cases, or whether it is merely a feature of the context in which we conducted our experiment. For example, the UK’s high death toll may simply be too strong a performance signal for even ardent government partisans to ignore, instead leading them to allocate responsibility differently. It is also worth highlighting that for the panel data analysis we do not find this asymmetry, further suggesting that these caveats may not generalize to other settings.

Our paper is also the first to posit and test that voters may differ with regard to how they recall the distant past performance of governments. Our results suggest that recall differs according to partisan identity. Our findings are not meant to suggest that voters care less about the past, though this may be the case, but that they genuinely recall past performance differently. For example, we find experimental evidence that opposition partisans are unlikely to recall the vaccine success when evaluating early government performance unless explicitly reminded about it. Relatedly, government partisans seem more likely than opposition partisans to essentially extrapolate back from the positive present context when evaluating the government’s performance at a far more negative stage in the pandemic. This carries obvious implications for democratic accountability in a context of increasing political polarization and invites further research.

In conclusion, our findings suggest that voters evaluate the government’s performance with bias even in the context of a deadly worldwide pandemic. Not only are voters biased in their evaluation of conditions and their attributions of responsibility, but they appear to have short and biased memories too. There are many challenges to democratic accountability—one is undoubtedly the bias that voters themselves bring into the ballot box. Democracy relies on voters accurately holding governments to account without bias, and our findings concerning selective evaluation, selective attribution and selective recall therefore have important implications for the limits of democratic accountability.