Affective Polarization in the Face of Crisis: The Impact of the COVID-19 Pandemic and De-Escalation Policies in Spain

Abstract

Escalating levels of sociopolitical confrontation worldwide have become a key concern in the social sciences. At the heart of this concern is affective polarization (AP), a concept that expresses the emotional distance between people’s positive feelings toward the party or parties they identify with and their negative feelings toward perceived opponents (Iyengar & Westwood, 2015; Iyengar, Sood, & Lelkes, 2012; Reiljan, 2020; Wagner, 2021).¹ AP is believed to foster negative consequences, such as institutional gridlock (Iyengar & Krupenkin, 2018; Levendusky, 2018), the erosion of political accountability, and diminished support for democratic norms (see Iyengar, Lelkes, Levendusky, Malhotra, & Westwood, 2019; Kingzette et al., 2021). AP is a usual suspect when it comes to explaining processes of growing incivility, confrontation, lack of accountability, and in general, democratic backsliding in some democracies (Orhan, 2022; Somer, McCoy, & Luke, 2021).

While the existing literature extensively examines factors contributing to AP, both perceptual-psychological (Moore-Berg, Hameiri, & Bruneau, 2020) and structural (Gidron, Adams, & Horne, 2020), less attention has been paid to short-term dynamics, despite their potential, unpredictable effects on political stability and social unrest. Most studies focus on medium to long-term drivers of AP, with only a handful exploring the impact of political events such as elections (Hernández, Anduiza, & Rico, 2021a), political campaigns (Hansen & Kosiara-Pedersen, 2017; Iyengar et al., 2012), elite cues (Bäck, Carroll, Renström, & Ryan, 2023), or coalition formations (Horne, Adams, & Gidron, 2023). The extent and mechanisms through which transient non-electoral political conjunctures influence AP and depolarization remain largely unexplored, hindering public decision-makers from developing effective strategies to mitigate adverse effects on political stability during crises.

To address this gap, we examine the COVID-19 pandemic as a critical external shock with potential to alter AP levels, both positively and negatively, at individual and aggregate levels. The pandemic provides an ideal case to study how citizens react to new party positions on unprecedented issues, which are alien to the content of party manifestos and usual partisan conflict lines. It also serves as an example of other unpredictable crises that health experts warn could occur again. The extensive data collected during this period allows for a comprehensive analysis of short-term changes in AP.

Specifically, this paper relies on three waves of a panel survey conducted in Spain in 2019, 2020, and 2021 to explore the short-term dynamics of AP. First, our work extends previous AP literature by examining whether citizens’ perceptions of the political management of the unprecedented COVID-19 crisis influenced their levels of AP. Second, drawing on the 2020 panel wave, which coincided with a significant de-escalation measure implemented by the Spanish government, it contributes to the depolarization literature by assessing the impact of perceived consensual policies on aggregate levels of AP beyond the initial stages of the pandemic.

Spain serves as an optimal case study for several reasons, starting by its elevated levels of AP, as documented by Gidron et al. (2020), which create a “hard test” environment for any intervention aimed at influencing polarization, thereby strengthening the robustness of our findings. In addition, Spain exhibits similar levels of AP to those found in Central Eastern and Southern European countries (Reiljan, 2020), positioning it as a typical case within a regional context and enhancing the generalizability of our results. Finally, the regional and temporal variability in Spain’s de-escalation measures provides a diverse case, offering insights into the effects of policy changes on AP across different settings.

While Spain had particularly restrictive lockdown measures, reaching 85.5 in the Oxford COVID-19 government response stringency index in April 2020, these measures were similar to those of other severely impacted countries such as Italy, France, or Ireland.² Thus, Spain can be viewed as representative of nations that faced intense challenges and implemented stringent lockdowns during the pandemic. The Spanish government’s de-escalation plan was gradual, asymmetric, and coordinated in four phases, thus allowing us to track the effect of the implementation of this policy on a sample of Spaniards, some treated and some untreated with a significant reduction in lockdown stringency. Specifically, we use a Difference-in-differences (DID) analysis to identify the effect of the long-awaited and consensus-driven de-escalation measures implemented by the Spanish government on aggregate levels of AP. Furthermore, we empirically test that the depolarizing effect resulted from individuals perceiving a narrowing of positions between the different parties after being subjected to de-escalation measures.

This study expands the literature on AP by examining how short-term critical junctures can both heighten and reduce AP, highlighting the complexities behind this phenomenon. Our contribution is threefold. First, we analyze short-term dynamics in the context of the pandemic, providing new evidence to address the limited and mixed results in existing research. Second, we explore the simultaneous influence of individual and aggregate factors, acknowledging that while certain evaluative processes about the government's management and the economy can boost AP for certain individuals, some large-scale policies, such as the lockdown's de-escalation measures, can reduce it. Third, we identify two causal mechanisms: citizens’ perceptions of the political management of the pandemic for increases in AP and perceived elite consensus for decreases, while controlling for alternative mechanisms.

The article is structured as follows: first, we connect the literature on the impact of crises on public opinion with insights from the literature on AP, conceptualizing the pandemic as a crisis marked by a set of focusing events. After that, we explain our case and methodology, then examine our results, and last, we discuss how our findings contribute to a better understanding of the contextual effects that affect AP dynamics and enhance the previous literature examining the political consequences of the pandemic.

Our study concludes that while citizens who perceived the pandemic as a threat to the country’s economy, as well as those with extreme evaluations of the government’s management, experienced surges in AP, certain episodes,—specifically, de-escalation—successfully lowered aggregate AP levels by reducing perceived elite confrontation. Our DID analysis demonstrates that Spanish citizens under the de-escalation protocol experienced a significant reduction in AP compared to those who remained under lockdown, and this effect is notably mediated by perceived elite polarization. Thus, our study explores how individual evaluations and broader contextual factors influence short-term AP dynamics, sometimes working in different directions simultaneously, while revealing causal mechanisms and illustrating the complexity of the forces underlying AP.

Theoretical Background

Affective Polarization Over Time

Previous AP literature has predominantly focused on understanding variations in this phenomenon among individuals and across different nations. Numerous studies have explored the individual-level determinants of AP, such as partisan identification, ideology, and social identity (Druckman & Levendusky, 2019; Iyengar & Krupenkin, 2018; Iyengar & Westwood, 2015). Cross-national analyses have also sought to identify patterns and disparities in AP toward parties and leaders across diverse political landscapes (Iyengar et al., 2019; Reiljan, Garzia, Da Silva, & Trechsel, 2023; Torcal & Comellas, 2022). While research on the temporal dynamics of AP is more limited, it has detected declining or stable levels in Germany, Canada and New Zealand, (Gidron et al., 2020), along an upward trend in AP in the United States (Boxell, Gentzkow, & Shapiro, 2022; Garzia, Ferreira da Silva, & Maye, 2023).

A strand of research explores how party competition along with subjective perceptions of such competition, influence AP levels, suggesting responsiveness to medium-term factors like political controversies (Yarchi, Baden, & Kligler-Vilenchik, 2021). Political campaigns also play a relevant role (Iyengar et al., 2012), increasing AP through negative advertisements in high-choice media environments (Lau, Andersen, Ditonto, Kleinberg, & Redlawsk, 2017). In European contexts, elections intensify voter preference for in-parties and disdain for out-parties (Hansen & Kosiara-Pedersen, 2017). However, elections themselves do not necessarily lead to sustained increases in AP. Proximity to elections tends to heighten perceived ideological differences between parties, which often diminish post-election (Hernández, Anduiza, & Rico, 2021a), as a result of the often necessary inter-party cooperation, which tends to reduce AP (Gidron, Adams, & Horne, 2023).

Instances of inter-party cooperation are key factors in lowering citizens’ AP levels (Bassan-Nygate & Weiss, 2022; Horne et al., 2023; Santoso, 2024; Wagner & Praprotnik, 2023). In turn, critical external shocks, such as economic crises (English, Grasso, Buraczynska, Karampampas, & Temple, 2016) or terrorist attacks (Chowanietz, 2011; Indridason, 2008), can influence elite cooperation by prompting coalitions across cleavages. Citizens may rally around the incumbent party or president in response to these events (Baum, 2003), or focus more on the crisis issue, shifting allegiances (Page & Shapiro, 1992) and becoming more polarized (Briscese, Grignani, & Stapleton, 2022). Despite the potential link between critical events, dynamics of party cooperation, political consensus, and AP, the causal chain connecting these phenomena remains unexplored. To clarify these connections, we first need to characterize the peculiar context of the 2020 pandemic.

On Crises and Public Opinion: the Pandemic as a Focusing Event

The COVID-19 outbreak can be conceived as a “catastrophic event,” similar to terrorist attacks, nuclear accidents, or natural disasters, which are all proven to have effects on public opinion (Boomgaarden & de Vreese, 2007). A concept akin to “catastrophic event” and “crisis,” borrowed from public policy studies, is “focusing event” (Kingdon, 2010); this refers to sudden and harmful events, known to policymakers and the public simultaneously, and necessitating urgent government action. The COVID-19 pandemic can be understood as a focusing event not only because of its unpredictable and harmful nature but also because it required urgent government action, forcing governments—and challenger parties and candidates—to take up new stances on various matters.

Previous studies found a “rally around the flag” effect caused by the COVID-19 outbreak (but see Amat, Arenas, Falcó-Gimeno, & Muñoz, 2020; Sosa-Villagarcia & Lozada, 2021 for different conclusions). That is, citizens gathered around their political institutions (governments, presidents) at the onset of the pandemic, as they had done so in previous crises or in the aftermath of a terrorist attack (Brody & Shapiro, 1989; Hetherington & Nelson, 2003), resulting in widespread political support (Baekgaard, Christensen, Madsen, & Mikkelsen, 2020; Schraff, 2020). However, this literature focuses more on political support indicators than on AP, or else they consider AP as an independent variable (Druckman, Klar, Krupnikov, Levendusky, & Ryan, 2020; Grossman, Kim, Rexer, & Thirumurthy, 2020; Stoetzer et al., 2023). Among the works dealing with AP as a research object during this period, Boxell, Conway, Druckman, and Gentzkow (2021) have shown that the onset of the crisis caused levels of AP to significantly decrease in the United States. However, evidence from Germany suggests that AP increased during the pandemic, particularly among AfD voters (Jungkunz, 2021). The uncertainty during the early stages of the pandemic also fostered AP between different ideological camps (Schmid, Treib, & Eckardt, 2023), and COVID-specific anger contributed to rising AP (Nguyen, Mayer, & Veit, 2022). These findings highlight the need to examine the heterogeneous and complex effects of the pandemic on AP in different contexts and among various individuals over time.

During the COVID-19 pandemic, governments faced crucial decisions, some requiring trade-offs between health and wealth. Prioritizing lives often meant sacrificing economic growth through measures such as lockdowns and sector shutdowns (Oana, Pellegata, & Wang, 2021). Government and opposition stances on pandemic policies highlighted key issues such as the economy and public health, prompting citizens to evaluate political actors based on these issues. Differences among parties regarding these aspects might have contributed to AP, considering the well-known role of policy disagreements in fueling AP (Lelkes, 2021; Orr & Huber, 2020; Webster & Abramowitz, 2017). Indeed, Wagner and Eberl (2024) found that vaccination debates intensified identity-based polarization. Thus, the COVID-19 pandemic might have had the potential to increase AP among citizens who processed the pandemic as a focusing event, that is, that assessed parties in terms of their proposals for overcoming the crisis. Hence, we the expect that:

H1) The pandemic increased AP among citizens who engaged in evaluations of the political management of the COVID-19 as a focusing event.

We acknowledge and test a potential alternative mechanism through which the pandemic might have impacted individuals’ AP levels: their personal experience of the illness. According to the Terror Management Theory (TMT, see Greenberg, Pyszczynski & Solomon, 1986; Solomon, Greenberg, & Pyszczynski, 1991), the pandemic might have heightened mortality salience and risk perception. This could have intensified individuals’ existing attitudes and worldviews, as a means of alleviating death-related anxiety (Burke, Martens, & Faucher, 2010), potentially aligning individuals against those who oppose their worldviews (Arrowood et al., 2017).³ This alternative explanation should be tested alongside the evaluative mechanisms derived from the focusing event rationale.

Pandemic Consensual Policies and Their Influence on AP Dynamics

The pandemic can also be viewed as a series of events continuing beyond its outbreak; these include the WHO’s declaration of the outbreak as a Public Health Emergency of International Concern (January 30, 2020), the initiation of “states of emergency” in different countries, the enforcement (and easing) of various lockdown measures, and the implementation of vaccination campaigns.

A crucial event with significant attitude-changing potential was the relaxation of strict lockdown conditions. While the AP literature has not yet measured the effect of these specific policies on AP, it has examined the role of inter-party cooperation, particularly coalition behavior, on polarization dynamics. Comparative (Gidron et al., 2023) and experimental (Bassan-Nygate & Weiss, 2022; Huddy & Yair, 2021; Wagner & Praprotnik, 2023) evidence highlights the importance of inter-party cooperation and coalitions as significant sources of depolarization. For instance, supporters of parties that are becoming part of a coalition tend to view other coalition partners in a more favorable light, even after the coalition ends (Horne et al., 2023). Previous literature suggests that perceptions of ideological proximity between elites could explain the depolarizing effects of elite cooperation (Hernández, Anduiza, & Rico, 2021a; Gidron et al., 2020; Wagner & Praprotnik, 2023), even though the evidence is still limited (Huddy and Yair, 2021). By the same token, policies agreed upon by incumbents and opposition parties should contribute to a decrease in AP, through the reduction in citizens’ perceived ideological differences between the elites.⁴

De-escalation policies illustrate this potential for cooperation. Boin and Lodge (2021) describe how, by the end of the pandemic’s “first wave,” western political leaders felt pressured to reopen up the economy quickly, driven by polls, social media and the business sector, among others. Adhering to the relaxation of measures gave the impression that the government was aligned with general demands and, in some countries, with the opposition, particularly where challenger parties championed relaxation measures. Evidence of this consensus effect is seen in higher citizen support for de-escalation policies backed by bipartisan coalitions (Flores et al., 2022). Thus, COVID-19 de-escalation can be viewed as an episode that has the potential to reduce perceived ideological differences between parties and ultimately produce a depolarizing effect.

Therefore, we contend that:

H2) De-escalation measures, when supported by parties of different ideological camps, resulted in an overall reduction of AP levels by contributing to a decrease in perceptions of ideological polarization.

We also acknowledge and test the possibility that de-escalation lowered AP levels through different mechanisms. The measures allowed citizens to socialize again after a prolonged period of isolation, likely increasing happiness and positively affecting evaluation of out-groups (Schwarz, 2000; but see Yu et al., 2021 for a null effect of happiness on AP). In addition, renewed social contact in the context of a shared and challenging experience could have fostered empathy and mutual understanding, improving social trust and reducing AP.

Research Design

The Spanish Context in Spring 2020—The Asymmetrical De-Escalation Plan

To empirically test our hypotheses, we focus on the Spanish case, characterized by high levels of AP, making it a rigorous test for hypotheses predicting further increases (H1).The regional and temporal variability in Spain’s de-escalation measures provides a dynamic environment to examine the effects of policy changes on AP, within a context of notable elite polarization (H2).

Following the 2019 elections, Spain saw the formation of its first coalition government at the national level since the democratic period’s inception, further polarizing the political landscape into two blocs. The coalition government, led by the center-left Socialist Party (PSOE) and the left-wing Podemos (We Can), faced opposition primarily from right-wing parties such as Ciudadanos, People’s Party (PP), and Vox (Simón, 2020). The government, additionally, had relatively low levels of popularity during the COVID-19 outbreak (Rubia et al., 2020).

Spain was among the hardest-hit European countries during the pandemic, surpassing one million COVID-19 cases early on. Like other severely impacted nations, Spain’s extended and strict lockdown measures were reflective of similar approaches adopted by Western countries, including Italy, France, the United Kingdom, and Argentina. In Spain, the government responded with a state of alarm declared on March 14, 2020, along with a strict lockdown that lasted until May 11. This period severely restricted movement, allowing exceptions only for essential activities. Political discourse during this time was marked by vigorous opposition, with parties engaging in a blame game over pandemic management (Rubia et al., 2020). Parliament extended the state of emergency six times, initially enjoying broad support but gradually losing it as debates unfolded (see Supplementary Table A4 in the Supplementary Appendix). Right-wing parties criticized the lockdown, with VOX referring to it as a “mass house arrest” and the PP advocating for less restrictive policies to mitigate economic damage (Olivas Osuna & Rama, 2021).

Finally, the government initiated an asymmetric de-escalation process starting May 11, which could be interpreted as a response to the growing demands of various parties and citizens.⁵ On that day, 28 provinces (those with a better epidemiological situation) and some municipalities entered phase 1, resuming non-essential economic activities and permitting social gatherings and outdoor activities within municipal limits. Simultaneously, the other half of the country maintained the existing restrictions, remaining in the preparatory phase (phase 0) for at least two additional weeks.⁶ At this juncture, 51% of Spaniards progressed to phase 1, while the remaining 49% continued in phase 0. We capitalize on this asymmetry in the de-escalation plan to evaluate the effect of this policy (the initiation of phase 1) on AP. The de-escalation policy was welcomed by large sectors of the citizenry, as the strict lockdown measures were eased. Moreover, this policy indicated a closer alignment between the government coalition and the rest of the parties in the parliament compared to previous weeks, signaling a high degree of agreement on the need to start a transition toward a “new normality.”

Data and Measures

To tap our main phenomena of interest, we use a Spanish panel survey, conducted yearly since 2010 by the Democracy, Elections and Citizenship research group (Hernández et al., 2021b). The sample was selected from the online panel curated by Netquest, which actively recruits potential participants by using commercial online services and websites, and then sends potential respondents one-time personal invitations, thus reducing the risk of self-selection and duplication. Quotas were used to ensure a balanced representation of participants in terms of gender, education, size of home municipality, and region. Questions about individuals’ personal experience with and assessment of the pandemic are included in waves 12 (May 2020) and 13 (May 2021), allowing us to observe both the short and long-term effects of pandemic-related shocks.

Our dependent variable (AP) is computed based on the respondents’ reported probability of voting for a party (PTV) (0-No probability at all to 10-In all likelihood), which can capture individuals’ affinity toward the different Spanish political parties. This measure has recently been used by different scholars to successfully capture AP (Balinhas, 2024; Orriols & León, 2020; Pérez-Rajó, 2024). We also present the correlations between like-dislike scales and PTVs for the same parties in a supplementary survey on political polarization in Spain (see Supplementary Table A3 in the Supplementary Appendix). The correlations for the main Spanish parties range from 0.80 to 0.87, indicating a very high association between the two measures and confirming that PTVs are a suitable measure for assessing individuals’ feelings toward parties. The operationalization of AP follows the logic of Wagner’s (2021) distance measure of AP, as seen in (1):

Here like_max,i is the PTV assigned by an individual to her preferred party, and n is the number of parties apart from the most-liked party. This way of assessing AP gives importance to an individual’s identification with only one party. The higher an individual scores, the more polarized she is. We do not weight this measure by party importance, mainly because all five parties included in the like-dislike scales are necessary actors to form left-wing or right-wing coalitions, so despite the degree of their parliamentary representation, all of them are key actors in a highly fragmented and volatile party system (see Simón, 2020).

As for our main independent variables, we rely on several items that gauge to what extent an individual is concerned by the pandemic as an issue, and if this positively affects her AP levels (Hypothesis 1). To begin with, we ask a question about how the government has been managing the country in general (five categories from “very bad” to “very good”). We have considered a curvilinear effect of this variable, as citizens holding either very positive or very negative opinions about the government may develop higher AP levels. Next, we asked about the most important problem facing the country and selected those who answered “health” (value 1 vs. the rest, 0). We also asked to what extent the pandemic is a “threat to the country’s economy and the country’s values,” to which the respondents could answer by selecting one of five options, from “not at all” to “a lot.” We will test the effect of these variables on our measure of AP (H1).

We will control these effects with the alternative explanation related to mortality salience. The survey included a question about the extent to which respondents felt that the pandemic threatened their health and their close ones’ health (from “not at all” to “a lot”). Furthermore, we calculated a more objective measure of mortality salience, based on the infection rate (new cases per population) per province between January and April—covering the four months preceding the fieldwork—for both 2020 and 2021.⁷ The incidence rate is uniform for individuals residing in the same province at the same point in time, and there are a total of 52 provinces. Our models also consider various controls such as gender, age, and self-placement on the left-right scale. Lags of the dependent and main independent variables are also considered, to make sure that we are not observing long-term dynamics that are dependent on previous party identification.

To examine the mechanisms responsible for depolarization (H2), we included two additional variables: perceived left-right polarization and issue-based polarization. Perceived left-right polarization captures the spread of positions that citizens attribute to different parties on the left-right scale, as described by Wagner (2021). Issue-based polarization follows the same logic as perceived left-right polarization, assessing the spread of positions that parties are perceived to have on three controversial political issues: Spanish nationalism, feminism, and the freedom versus order spectrum.⁸ Our measure of issue-based polarization is the average perceived polarization across these issues. These variables are measured in wave 12 (2020) on a 0 to 1 scale, then tested as mediators of the effect of the de-escalation policy on AP. In addition, left-right polarization measured in wave 11 (2019) is included as a control in our initial estimations.

Two additional variables are relevant for our causal identification strategy. According to our theoretical framework, easing lockdown measures could impact life satisfaction and social trust rather than perceived elite polarization. Thus, we consider life satisfaction and social trust as alternative dependent variables and mediators in our H2 empirical tests. Social trust is a dichotomous variable, with a value of 1 for those who tend to trust people and 0 for those who believe one can never be too cautious. Life satisfaction is measured on a 0 to 1 scale, with higher values indicating greater satisfaction. Additional details about the wording and coding of the variables used in our analyses can be found in the Supplementary Appendix.

Method

Given the nature of our dependent variable (a scale ranging between 0 and 1, after normalization), we employ Ordinary Least Squares regressions to estimate the effects of the pandemic.⁹ Specifically, we use static score models (Finkel, 1995; Pop-Eleches, Robertson, & Rosenfeld, 2022). In these models, the most recently measured dependent variable is regressed on a vector comprising a lagged dependent variable (measured at wave 11, June 2019) along with a vector of control variables also measured at the beginning of the time span (Finkel, 1995).¹⁰ The temporal dynamics of the relationships found in our main models will be examined for other waves, specifically measuring AP and the independent variables at the height of the 2020 pandemic lockdown (wave 12) and a year later (2021, wave 13). Comparing the effects of the “focusing event” indicators tapped in 2020 and 2021 on our measures of AP will also shed light on the duration and evolution of the effects.¹¹

For clarification, Equations (2) to (4) explicitly outline the estimation of our controlled models. In all cases, a vector of controls (including a lagged version of the dependent variable) was measured at the earliest point available (2019). In the first estimation (Equation 2), all the relevant variables (dependent and main independents, except controls) were measured during the lockdown, i.e., in 2020. This will illustrate how perceptions of the events unfolding affected individuals’ polarization at the height of the pandemic crisis. The next Equation (3) measures the dependent variable at the most recent available moment in time (2021) to discern if perceptions of the political context and an individual’s personal situation during the lockdown still affected AP one year later. The last Equation (4) measures all the relevant variables (except the controls) at the most recent moment available (2021) to discern if the indicators referring to the pandemic as a focusing event or as an event that was able to boost mortality salience one year after the outbreak—when conditions had objectively improved—are still able to affect individuals’ AP.

As for the general effects of the de-escalation phase on the population, we employ a DID strategy. This quasi-experimental design tracks, over time, a group that is treated with a policy or an event, along with an untreated (control) group, to estimate a causal effect. The goal is to analyze differences in outcomes across the treatment and control groups, which occur between pre-treatment and post-treatment periods. The approach removes biases in post-intervention period comparisons between the treatment and control group that might have existed before the intervention. In our case, individuals in territories treated with de-escalation (onset of phase 1) on May 11 should experience a reduction in their levels of AP, when compared with individuals that remained in phase 0.

Our DID design estimates the average treatment effect for the treated, which is calculated as shown in (5):

Following the classical notation, we first compute the difference in AP levels for the experimental group (EG) before the de-escalation was implemented (t0) and after this policy was applied (t1), representing the first difference. Subsequently, we computed the differences in AP levels between the two time-periods for the control group (CG), consisting of individuals living in regions where de-escalation was not implemented, representing the second difference. Finally, we subtracted the second difference from the first difference to obtain the average treatment effect for those treated (Athey & Imbens, 2006). If this last figure is significant, it indicates that the mean differences in the growth rates of the outcome between the treatment and the control group are due to the effect of the de-escalation policy, accounting for all pre-existing differences between the groups.¹²

We estimated the results by introducing an interaction term between time (pre-/post-intervention) and treatment group dummy variables. In addition, a second model includes further controls, such as sex, age, education, local COVID-19 incidence rates, and dummy variables for each day of the fieldwork before the treatment, except the first day, which serves as a reference, to account for potential pre-treatment biases. To rule out alternative explanations, we replicated our DID model using different dependent variables (i.e., life satisfaction and social trust, see Table 3) where we do not expect significant results. In addition, we conducted mediation analyses to determine whether the observed effects were mediated by a reduction in perceived elite polarization levels, as hypothesized in H2 (see Table 4).

Results

Table 1 displays the results of a series of OLS estimations that follow the static scores approach, both without controls (models 1–3) and with controls (models 4–6). The dependent variable in the first model is our AP indicator measured in May 2020 (wave 12). Two variables tapping the focusing event effect are significant: the assessment of the government’s performance and the perception that the pandemic is a threat for the economy. Both variables retain their significance with the addition of controls, although the explanatory power diminishes, and the significance level for threat perception also decreases (see fourth model). The assessment of government performance is represented by a quadratic term. The coefficients indicate that initially, as individuals’ evaluations of the government improve, AP decreases. However, this decline in AP slows as evaluations continue to rise, eventually leading to an increase in AP. This indicates a U-shaped relationship between government evaluations and AP levels. Thus, perceptions of the pandemic as a threat to the country’s economy, alongside both strong negative and positive views of government management, are positively associated with AP levels in 2020.

Shifting to models that measure the dependent variable in 2021 and key independent variables in 2020 (second and fifth models), we again observe a significant (p < .1 in the controlled model) effect of perceived threats to the country’s economy and a persisting effect of both very positive and very negative evaluations of the government. Considering health as the primary issue in the country a year prior shows a positive and significant effect (p < .05), but this effect disappears once we account for controls.

Lastly, the third and last models measure all relevant variables (except for controls) in 2021, and are meant to capture if the rationales and associations between the key variables have changed over the course of a year. We observe that AP is still affected by strong opinions on the government’s management, even after controlling for individuals’ previous levels of AP, ideology, or perceptions of the government. Considering the pandemic as a threat to the economy is positively and significantly associated with AP in both the uncontrolled (p < .001, third model) and controlled models (p < .05, last column). The rationale according to which the pandemic can be conceptualized as a focusing event that forces parties to reveal new policy positions and fosters issue-based evaluation (and further AP) seems to hold. Explanations related to mortality salience do not exert a significant impact, with two exceptions: the provincial incidence rates measured in 2020 are positively and significantly related to AP measured in 2021 (p < .05, see fifth model) indicating that those living in worse-affected areas developed higher levels of AP. Individuals’ perceptions about the pandemic as a threat to their health or their close ones’ also boost AP in the last estimation, when both the dependent and the independent variables are measured one year after the outbreak of the pandemic. It is worth considering that this effect could be a post-hoc rationalization influenced by partisanship.

Overall, these results provide strong support for our first hypothesis. Strong evaluations—both particularly negative and positive—of the government’s management appear to be a primary mechanism influencing AP during the pandemic, while perceptions of the pandemic as a threat to the economy also play a relevant, albeit somewhat weaker, role. Notably, our estimations show a higher model fit when both the dependent and independent variables were measured at the peak of the pandemic (May 2020, columns 1 and 4), suggesting that the focusing event explanations are most effective in a crisis context.

Finally, we turn to the last contextual effect related to the pandemic, by analyzing the impact of easing the lockdown through a DID approach. To do this, we divided the respondents of our survey into two groups: those treated with the easing of the lockdown (beginning of phase 1 of the de-escalation) and those untreated. Some regions fully entered the first phase on May 11, while others, such as Castilla y León and Valencia, only partially entered, or, like Madrid, did not enter at all. The regions left behind were re-evaluated again to see if they could enter phase 1, but this was after the end of our fieldwork. Figure 1 depicts the evolution of the levels of AP for both groups before and after the treatment (end of strict lockdown due to passing to phase 1). We observe a parallel evolution of both groups before the critical date (May 11) and, afterwards, a steady decline of the treated AP levels, which is not matched by the untreated group’s levels. We also observe that most of the post-treatment differences are concentrated around the weekend (15–17 May), suggesting that citizens may only become fully aware of the implications of de-escalation measures during their leisure time, that is, weekends.

Figure 1.

Evolution of AP levels for those treated with phase 1 (easing of strict lockdown) and those untreated.

Open in new tabDownload slide

Table 2 presents the results of the DID analysis (uncontrolled, left column; controlled, right column). The first uncontrolled model reveals that, before the treatment, the AP levels of those about to enter phase 1 of the de-escalation were significantly higher than those who were not (b = 0.04, p < .1). After the treatment, the AP levels were 8 percentage points lower for those treated. The second model in Table 2 includes a series of relevant controls, such as the 2020 local incidence of the disease between January and April. Importantly, we have included dummies per day for the pre-treatment period, excluding the first day of fieldwork for reference and to prevent perfect overlap with the “post-intervention” variable. This was done to ensure that both the treated and the untreated groups in our sample followed parallel trends before the intervention. The inclusion of control variables has the effect of reducing the treatment’s estimated impact by one percentage point. Consequently, we conclude that the intervention had a substantial seven percentage points effect in reducing levels of AP, signifying a noteworthy average treatment effect for those who underwent the treatment.

To provide initial evidence on the mechanism behind the observed depolarization, Table 3 also presents the (uncontrolled) impact of the treatment on alternative dependent variables, such as life satisfaction and social trust. The null effects for these variables rule out the possibility that they may connect de-escalation with AP, as mediators would typically show significant relationships with the treatments. In contrast, the results suggest that perceived elite polarization may be a plausible causal mechanism for the relationship between the treatment and the outcome.

Table 4 presents the mediation effects for four potential mediators, ruling out the possibility that the treatment’s effect is conveyed by psychological processes such as increases in life satisfaction or social trust.¹³ However, we see that perceived differences in parties’ positions regarding several issues or the left-right scale play, indeed, a significant effect: those treated with phase 1 experienced a decrease in these perceptions, followed by a decrease in their levels of AP. The share of the total effect of the treatment is mediated by 53% through perceived issue-based polarization and 42% through perceived left-right polarization.

In sum, AP levels are sensitive to government measures. The implementation of de-escalation affected the perception of elites’ ideological polarization, ultimately leading to a decline in AP levels of those affected by this popular policy.

Conclusions

The sociopolitical landscape shaped by the COVID-19 pandemic presents a unique opportunity to delve into the impact of exceptional events and specific political measures on one of the most pressing facets of contemporary public opinion: AP. Past literature has primarily focused on the individual and institutional foundations of AP. While some studies have explored the medium to long-term drivers of AP, our study addresses a significant gap in the understanding of short-term non-electoral dynamics, particularly in the context of external shocks and crises.

Building on existing literature and conceptualizing the pandemic as a focusing event, we explore how individual-level variables influenced the temporal dynamics of AP during March 2020. The pandemic compelled some citizens to engage in issue-evaluation processes, heightening their AP levels even a year after the COVID-19 outbreak. In addition, we investigate the depolarizing effects of a specific pandemic management policy, particularly the easing of the May 2020 lockdown. Our study, utilizing three waves of a Spanish panel survey and static score models, provides nuanced insights into the temporal dynamics of AP. Specifically, we found that the pandemic, as a focusing event, widened the affective gap between in-parties and out-parties among citizens with strong views (either very positive or very negative) regarding the government’s management of the crisis. Conversely, citizens who perceived the pandemic as a threat to the country’s economy also experienced an increase in their levels of AP. Both results suggest that citizens who engaged in issue-evaluation processes—either through extreme assessments of the government’s crisis measures or by viewing the pandemic as an economic threat—became more affectively polarized.

While our theoretical framework and data do not enable us to identify individuals more likely to engage in evaluative processes, our static score models account for previous individual levels of AP. This effectively eliminates within-individual time-invariant factors (e.g., political sophistication or cognitive styles) that could otherwise confound our results. Consequently, we can attribute the observed effects to individuals’ perceptions regarding the political management of the pandemic. Our dynamic models also reveal that the effects of these extreme assessments on AP levels persisted for at least one year.

In a novel contribution to the affective depolarization literature, we examined the impact of the de-escalation policy implemented by the Spanish government. Using a DID design, we observed a significant seven-point reduction in AP associated with the easing of lockdown measures. Furthermore, to test the main mechanism behind depolarization, mediation analyses reveal that a reduction in perceived left-right and issue-based polarization mediates a significant portion of the depolarizing effect of the de-escalation policy. This finding offers valuable insights for depolarization literature (Huddy & Yair, 2021; Wagner & Praprotnik, 2023) by demonstrating the importance of perceptions of elite ideological disagreements as a mechanism underlying depolarization, thereby contributing to the ongoing debate on the relationship between ideological polarization and AP. In addition, our findings emphasize the role of elite-level dynamics of competition and cooperation in either fueling or defusing affective depolarization (Gidron et al., 2020, 2023).

Our results are based on a quasi-experimental technique combined with a large-N panel survey, offering high levels of external validity in a real-world situation, providing clues on when the causal mechanisms behind this relationship operate. However, we do not address all potential causal pathways. For instance, despite convergence on core de-escalation policies, the Spanish media—known for its generalized tendency to amplify inter-party divisions (Pérez-Escoda, Boulos, Establés, & García-Carretero, 2023; Teruel-Rodríguez, 2023)—likely contributed to the delayed effects of depolarization. These effects only became evident over the weekend following the initiation of the de-escalation measures, suggesting that citizens may have required additional time to internalize the reduced elite-level disagreements amidst ongoing media portrayals of division. Future research should explore how and when citizens’ perceptions of elite polarization shift independently from ongoing media conflict framing. In a similar vein, while our findings are grounded in Spain’s stringent pandemic response, which poses certain limits to their immediate generalizability, they may still be relevant for other countries that implement similarly severe measures in future crises. The political dynamics observed here may reemerge in different contexts, and it will be essential to investigate whether depolarizing effects similarly manifest in other nations or crises.

In conclusion, our study navigates the complex terrain of AP during the 2020 pandemic, shedding light on short-term dynamics and the distinct influences of individual and contextual factors. In complementing previous research on public opinion during COVID-19, especially on AP (Wagner & Eberl, 2024), we highlight the varying contextual effects of the health crisis and its political management on AP. By examining the temporal evolution of AP, our research offers a nuanced exploration of contextual effects during the pandemic and contributes to the broader literature on public opinion and the political implications of crises, particularly the COVID-19 crisis.

Funding

This work was supported by the Spanish Minister for Science and innovation (grant number PID2020-113030RB-I00/ AEI/10.13039/501100011033).

References

Biographical Notes

Carol Galais is an associate professor of political science at the Universitat Autònoma de Barcelona.

Daniel Balinhas holds a PhD in Political Psychology from the Universitat Autònoma de Barcelona.

Footnotes