How does Polarization Moderate Economic Voting?

Individual-level findings in the literature on polarization suggest that polarized politics has a chief effect on voting behaviour: as the ideological distance between parties becomes more evident for a larger group of citizens, the number of swing voters decreases. Elite polarization clarifies the ideological position of political parties for ordinary citizens (Hetherington Reference Hetherington2001). As a consequence, voters can increasingly rely on partisan cues to make their decisions (Druckman, Peterson and Slothuus Reference Druckman, Peterson and Slothuus2013; Parker-Stephen Reference Parker-Stephen2013). Rather than searching for information about the performance of incumbents and observing candidates’ proposals, voters tend to assume that the ideological difference between parties suffices to inform vote choice. Among partisans and independents who lean towards one of the major parties, polarization enhances affective attachment to parties and stimulates motivated reasoning. Partisans, therefore, are less likely to waver in their support under polarized environments.

But the impact of polarization on voting behaviour is not restricted to partisan voters. By illuminating the ideological divergence between parties, polarization also influences how independents, less-engaged voters – and even nonvoters – perceive candidates (Smidt Reference Smidt2017). Since the influential work of Converse (Reference Converse1964), scholars have characterized voters who are prone to change their support as less informed, indifferent, or ambivalent (Zaller Reference Zaller2004; Mayer Reference Mayer2008). These voters are less likely to distinguish the different positions of political parties on policy issues. For this reason, they would be more responsive to new information such as the performance of incumbents (Key Reference Key1966). However, as Smidt (Reference Smidt2017) shows, independents are more likely to recognize the importance of parties’ issue positions when polarization clarifies party differences. The recognition of parties’ ideological differences reduces ambivalence among independent voters, who also become less likely to move across party lines in successive elections.

In terms of economic voting, these processes, induced by polarization, lead to the expectation that the number of voters defecting to the opposition to punish incumbents for a poor economy decreases as political polarization increases. Similarly, voters who identify with the opposition become less likely to cast a ballot for the party in government when the economy prospers. In other words, voters tend to stick with their parties in polarized politics even if economic conditions indicate that they should switch parties at the polls. Empirically, this expectation leads to the first hypothesis of this paper:

Besides affecting vote choice, polarization can also influence participation. When politics is polarized, what is at stake in the election tends to be higher because the distance between parties’ policy agendas is larger and sharper. Also, as mentioned before, more voters can notice this ideological difference and thus become more concerned about election results. Additionally, affective polarization makes the victory of a party excessively costly for those voters who dislike it – particularly if voters perceive candidates in the other party as extremists (Hall and Thompson Reference Hall and Thompson2018). Finally, the social pressure to participate in polarized environments tends to be high as polarization enhances the number of politically engaged citizens. Social pressure and network are, in turn, crucial determinants of participation (Gerber, Green and Larimer Reference Gerber2008; Sinclair Reference Sinclair2012; McClendon Reference McClendon2014). For these reasons, the benefit from participating may increase for many citizens who are now more likely to notice and care about election outcomes (Downs Reference Downs1957). In a nutshell, polarized politics may increase political engagement and participation (Abramowitz Reference Abramowitz2010; Hetherington Reference Hetherington2008). It follows from this expectation that:

Under low polarization, voters who are dissatisfied with economic outcomes but do not intend to support the opposing party can still opt for a third option: abstention.Footnote ⁸ However, as the environment becomes more polarized, both the cost of abstaining and the benefit from participating should also increase. Therefore, fewer voters will abstain as a means to penalize their preferred party for negative economic conditions under polarized environments. Hypotheses 1 and 2a reflect my theoretical expectations of a sticking effect of polarization: voters are less likely to blame their party for policy outcomes and thus desert it at the polls when politics is polarized.

Hypothesis 2a implies that turnout depends on the benefits voters link to election outcomes. These benefits are, in turn, a function of the set of candidates available in the race (Downs Reference Downs1957; Riker and Ordeshook Reference Riker and Ordeshook1968). When the benefits voters associate with candidates in the choice set are low, these voters will be less likely to participate. Hence, abstention should be modelled as one of the options available to voters at elections. Although the idea of abstaining as a voting option is not new, few empirical analyses incorporate abstention as an option voters can take whenever they are not satisfied with the choices on the ballot (Horiuchi and Kang Reference Horiuchi and Kang2018; Weschle Reference Weschle2014; Dostie and Dupré Reference Dostie and Dupré2012; Arzheimer and Evans Reference Arzheimer and Evans2010). I am unaware of any work theorizing how abstention in response to economic conditions varies as a rational reaction of voters to the political context, particularly to increasing polarization.

There is, however, some evidence that polarization can also increase abstention (Rogowski Reference Rogowski2014). Although polarized environments stimulate partisan-motivated reasoning, scholars have found that some voters can still update their assessment based on incumbents’ performance (Lewis-Beck and Stegmaier Reference Lewis-Beck and Stegmaier2018; Healy and Malhotra Reference Healy and Malhotra2013), candidates’ characteristics (Stone Reference Stone2017), and new information (Peterson Reference Peterson2017; Boudreau and MacKenzie Reference Boudreau and MacKenzie2018). Particularly, weak partisans and independents would not rely solely on partisan cues to make their voting decisions even when they perceive the ideological distance between parties more objectively. Under polarized environments, they could still react to negative performances of their favourite party in managing the economy. Yet, the affective impact of polarization influences how they perceive the other party (Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019). To the extent they no longer see the other party as an alternative, abstention could be a more attractive option to express dissatisfaction with their favourite party in polarized politics. In this context, some voters would not see themselves represented by either side, although they may still prefer one major party to another. This expectation leads to an alternative hypothesis about the effect of polarization on abstention:

This hypothesis represents the expressive abstention effect of polarization, which could increase the number of voters who choose abstention to signal that they are disappointed with how their preferred party managed the economy.

Hypotheses 2a and 2b describe opposite expectations about how polarization can affect abstention in response to economic conditions. These contradictory predictions stem from the fact that the effect of polarizing politics on abstention depends on the ideological distribution of the electorate. In a hypothetical scenario in which voters’ ideological distribution has a single peak between two polarized parties, moderates and partisan leaners who are disappointed with their preferred parties may choose to abstain. The other party is no longer an attractive alternative as it is too far away from their ideological positions. Panel A in Figure 1 depicts this context in which polarization can provoke expressive abstention. Yet, this scenario would also create the opportunity for entrepreneurial politicians to move towards the median voter (Downs Reference Downs1957). As a result, political polarization should decrease over time, leading to a trend opposite to the one observed in the data.Footnote ⁹ Hence, this hypothetical scenario, which is necessary to confirm the expressive abstention effect of polarization, would be out of equilibrium.

Figure 1.

Scope Conditions of Abstention in Response to Economic Circumstances.

Note: Curves represent the distribution of voters across the ideological continuum that ranges from extreme liberal (-1) to extreme conservative (1). Dashed lines indicate the ideological placement of parties.

However, if the electorate has two peaks at opposite sides of the political spectrum that map to the position of polarized parties, the cost of abstaining will be high for a large set of voters. Even if they are dissatisfied with their preferred party, their ideological distance from the other viable option encourages motivated reasoning and reinforces their partisan choice. Panel B in Figure 1 displays a context in which both the masses and the elites are polarized. Under these circumstances, polarization will decrease the number of voters who abstain as a means to sanction their favourite party.

In summary, Hypotheses 1 and 2a predict that political polarization has the same attenuating effect on both channels through which economic voting can operate. Polarized politics decreases both the number of voters who cast a ballot for the opposition to sanction incumbents for poor economic conditions and the number of those who abstain in order to penalize the party in government for poor economic outcomes. Although the necessary condition for Hypothesis 2b would be out of equilibrium, I test for it as an alternative to Hypothesis 2a. The next section describes the data and method I use to test Hypothesis 1 and adjudicate between Hypotheses 2a and 2b with election results at the county level.

At the aggregate level, though, I cannot investigate which voters behave in line with these hypotheses and how polarization affects them.Footnote ¹⁰ Motivated reasoning suggests that partisans are particularly affected by polarization in their response to economic conditions. For these voters, the cost of having the party they dislike in the presidency increases with polarization. Given the evidence that partisan cues bias overall judgement of responsibility among partisans and that polarization detaches presidential approval from economic assessments and weakens the link between economic conditions and economic opinion (Rudolph Reference Rudolph2006; Donovan et al. Reference Donovan, Kellstedt, Key and Lebo2020; Goidel and Kellstedt Reference Goidel and Kellstedt2022), I argue that as the perception of polarization increases among partisans, they become less likely to negatively evaluate the economy when their preferred party holds the Presidency. As such, I expect that:

Yet, besides affecting economic assessments, there is another mechanism through which polarization can moderate economic voting. Some partisans may still be able to assess the economy in polarized environments, as Ang et al. (Reference Ang, Reeves, Rogowski and Vishwanath2021) show. But it does not follow that those partisans intend to use this assessment to inform their vote choice, assigning importance to economic issues and blame to their preferred party (Kane and Anson Reference Kane and Anson2023). Under high levels of polarization, partisans are less likely to hold their preferred party accountable for poor economic performances. Instead, they tend to resort to motivated reasoning to accommodate party attachment and economic perceptions. Therefore, I predict that:

Figure 2 depicts the two links of the economic voting chain that polarization can moderate. Polarized politics may weaken the connection between economic conditions and economic perception, but it can also shape how assessments of the economy influence vote choice. In the fourth section, I describe my empirical strategy to test Hypotheses 3a and 3b with ANES data and present the results.

Figure 2.

Polarization and the Chain of Economic Voting.

Data and Research Design

An ideal case to test the impact of polarization on economic voting would hold institutional factors that may influence the clarity of responsibility constant while the level of polarization varies. The results of the U.S. presidential elections combine these conditions.

The level of national polarization has changed over time in American politics, but the institutional framework that can moderate economic voting remained constant. Moreover, the two-party system makes the US appropriate for measuring polarization and estimating its effect. Although other democracies have experienced the emergence of radical leaders, multiparty systems and different levels of party attachment create challenges to conceptualizing and studying polarization. In multiparty democracies, this phenomenon may indicate the radicalization of politics rather than the ideological polarization of the whole party system.Footnote ¹¹ In the US, changes in the ideological distance between the two major parties express variation in elite polarization, which is the measure I employ to capture changes in polarizing politics.

Figure 3 shows that the ideological distance between the median Republican and median Democratic representatives in the House has increased since the 96th U.S. Congress in 1981. However, most of the change in ideological polarization occurred after 1990. Between 1957 and 1989, the ideological distance between median representatives in the two major parties increased by 16 per cent, from 0.57 to 0.66. From 1991 to 2019, which is the period analyzed in this paper, the gap has increased by 35 per cent, from 0.66 to 0.89. The increasing distance between parties over this period seems to be driven mostly by the Republican party. Whereas the median Democratic has moved about 18 per cent to the left – from −0.32 to −0.38 – the conservative position of the median Republican increased approximately 48 per cent during the same period – from 0.35 to 0.52.Footnote ¹²

Figure 3.

Polarization in the U.S. Congress over time.

Note: The median Republican (red line) and the median Democratic (blue line) representative in the House. The vertical dashed line shows when my aggregate-level analysis starts. Source: Voter View Project.

To test my hypotheses at the aggregate level, I implement a seemingly unrelated regressions (SUR) strategy in which the electorate is divided into three categories: the share of votes for the incumbent party, for the opposition, and abstention. Following the approach developed by Aitchison (Reference Aitchison1982) to model compositional data and used by Tomz, Tucker and Wittenberg (Reference Tomz, Tucker and Wittenberg2002) to study election results in multiparty systems, the outcome variables of the SUR model are the $J - 1$ log-transformed ratios between the percentage of voters in each $j$ piece of the constituency – voters casting a ballot for the party in government, for the opposition, and abstainers – to a baseline category ( $j\,=\,1$ ).Footnote ¹³ Thus, for a county $i$ in election $t$ , the outcome variables are defined as follows:

(1) $${s_{jit}} = \log \left( {{{{y_{jit}}}}\over{{{y_{1it}}}}} \right)\forall j \ne 1$$

where ${y_{jit}}$ is the share of voters in the category $j$ . I create the log-ratios ${s_{jit}}$ with data on presidential election results, turnout, and voting-age population by county from 1992 to 2020.Footnote ¹⁴ My main models use a measure of effective abstention, which combines votes for third parties and the number of voters who abstained. This measure is appropriate to test if voters can express their dissatisfaction with the incumbent for economic conditions through a channel other than voting for the opposition. A vote for a third party in U.S. presidential elections is equivalent to a wasted vote. Yet, casting a ballot for a third party can still express voters’ dissatisfaction with their preferred party.Footnote ¹⁵

For the explanatory variables, polarization is measured with the difference in the DW-nominate scores of the median House Republican and the median Democrat in the U.S. Congress immediately before each election (Theriault Reference Theriault2008; Lewis et al. Reference Lewis, Poole, Rosenthal, Boche, Rudkin and Sonnet2021).Footnote ¹⁶ Given the growing nationalization of local and state politics (Jacobson 2015 Reference Jacobson2013), this national-level measure is appropriate for capturing changes in polarized politics and estimating their effect on presidential elections. My economic indicator is the average annual unemployment rate in the county reported by the U.S. Bureau of Labor Statistics, which is expected to be negatively associated with votes for the incumbent party in presidential elections.Footnote ¹⁷ Besides increasing the cross-sectional variation of my data and, as such, reducing the potential influence of temporal patterns, the local economy is close to the reality that voters experience and can affect presidential election outcomes (Healy and Lenz Reference Healy and Lenz2017). After all, the national economy is the aggregation of local indicators.Footnote ¹⁸

In addition to the large cross-section variation in my data introduced by the economic indicator and outcome variables, I address methodological concerns regarding the upward trend of national polarization with three other empirical strategies. First, my aggregate-level analyses control for the lagged dependent variables (LDV), ensuring that my compositional outcomes in Equation 1 are stationary. Second, the results are robust to models that, besides the LDV, include a time-trend variable that can capture the influence of other factors that might be trending in national politics.Footnote ¹⁹ Finally, my individual-level models use election fixed effects to exclude unobserved variables that change over time, and their results are consistent with my findings at the aggregate level.Footnote ²⁰ Altogether, these strategies demonstrate that the results in this paper are not spuriously driven by temporal influence.

My final dataset has 3,108 counties across eight consecutive elections. With these data, I estimate a system of $J - 1$ equations using the following interactive model:

(2) $${s_{jit}} = {\beta _{1j}}{\rm{Unem}}{{\rm{p}}_{it}} + {\beta _{2j}}{\rm{Po}}{{\rm{l}}_t} + {\beta _{3j}}{\rm{Unem}}{{\rm{p}}_{it}}^{*}{\rm{Po}}{{\rm{l}}_t} + {\boldsymbol \delta} _{kj}{\boldsymbol x}_{it} + {\boldsymbol\alpha _j}{\bf{Stat}}{{\bf{e}}_i} + $$

$${\beta _{4j}}{\rm{CloseEle}}{{\rm{c}}_{it}} + {\beta _{5j}}{\rm{RepPr}}{{\rm{e}}_t} + {\beta _{6j}}{\rm{IncCan}}{{\rm{d}}_t} + {\phi _j}{s_{jit - 1}} + {\varepsilon _{jit}}$$

where ${\rm{Unem}}{{\rm{p}}_{it}}$ is the unemployment rate, ${\rm{Po}}{{\rm{l}}_t}$ stands for polarization, ${x_{it}}$ is a vector of $K$ demographic variables at the county level, and ${\bf{Stat}}{{\bf{e}}_i}$ is a vector of fixed effects indicating the state in which the county $i$ is located. ${\rm{CloseEle}}{{\rm{c}}_{it}}$ measures the distance in vote share between the two presidential candidates in the state where the county $i$ is located, ${\rm{RepPr}}{{\rm{e}}_t}$ is a dummy variable indicating whether the Republican party held the presidency at $t$ , ${\rm{IncCan}}{{\rm{d}}_t}$ is an indicator of whether an incumbent candidate ran for reelection at $t$ , and ${s_{jit - 1}}$ is the LDV, the temporal lag of the dependent variable.Footnote ²¹ ${\varepsilon _{jit}}$ is an error term that may be correlated across the $J - 1$ equations of the SUR model.

The parameterization of the SUR model is similar to a multinomial logistic regression. For each explanatory variable, the SUR model estimates $J - 1$ coefficients. Each of these coefficients is the estimated impact of a one-unit increase in the explanatory variable on one of the ${s_{jit}}$ outcomes, which is in the log-ratio form. A positive coefficient means that its predictor makes the category in the numerator of the logged ratio in Equation 1 larger at the expense of the baseline category in the denominator. Negative values, in turn, have the opposite interpretation. But the substantive interpretation of model results involves ‘translating quantities of interest back into their original composition structures’ (Philips, Rutherford and Whitten Reference Philips, Rutherford and Whitten2016, 271) – share of voters – as I show in Figure 5.

As demographic controls, ${x_{kit}}$ , all models include the following predictors: share of black voters, share of Hispanic voters, share of Asian voters, share of young voters – less than 30 years old – share of the elderly population – more than 70 years old – the county population, share of people with a college degree, share of workers in the manufacturing sector, a rural-urban code, and the average per capita income in the county.Footnote ²² I use state fixed effects to account for electoral rules that can influence abstention and other state-specific factors.Footnote ²³ Additionally, I control for the difference in vote share between the first two candidates in the state ( ${\rm{CloseEle}}{{\rm{c}}_{it}}$ ) as the closeness of the election may affect turnout. The lagged log-ratio, ${s_{ijt - 1}}$ , controls for persistent features of the county that influence voting behaviour but cannot be observed. By controlling for the trade-off between categories in the last election, it can also address concerns about factors that may be gradually changing over time—such as turnout—and could be associated with the uptrend of elite polarization.Footnote ²⁴

Table 1 summarizes the expected relationship predicted by each hypothesis. In the trade-off between incumbent and opposition – ${s_{jit}} = \log \left( {{{{\rm{Incumben}}{{\rm{t}}_{it}}}}\over{{{\rm{Oppositio}}{{\rm{n}}_{it}}}}} \right)$ – Hypothesis 1 predicts that higher polarization reduces the impact of economic factors on voting across party lines. Thus, I expect that ${\hat \beta _{3j}}$ will be significant in the opposite direction of ${\hat \beta _{1j}}$ in the equation that estimates the flow of voters between the incumbent party and the opposition. Since an increasing unemployment rate is expected to harm the incumbent party in favour of the opposition ( ${\hat \beta _{1j}} \lt 0$ ), Hypothesis 1 predicts that polarization will attenuate this negative effect of unemployment on the incumbent’s vote share – which implies that ${\hat \beta _{3j}}\, \gt \,0$ . As polarization increases, fewer voters will cast a ballot for the opposition to punish the party in government for rising unemployment.

Table 1.

Summary of Theoretical Expectations

Note: The ratio indicates the trade-off between parties or a party and abstention. Bracketed signals are effects that are not clearly predicted by traditional economic voting theory. Hypotheses 3a and 3b cannot be tested at the aggregate level.

In Hypothesis 2a, I predict the attenuating effect of polarization on abstention in response to a poor economy. Motivated reasoning, affective polarization, and social pressure in a polarized electorate would increase the cost of abstention. In this context, a large set of voters will prefer to cast a ballot for their preferred party – even when economic conditions suggest that they should punish it – than abstain. Hence, ${\hat \beta _{3j}}$ should be positive in the equation between the incumbent and abstention – ${s_{jit}} = \log \left( {{{{\rm{Incumben}}{{\rm{t}}_{it}}}}\over{{{\rm{Abstentio}}{{\rm{n}}_{it}}}}} \right)$ – reducing the number of voters who abstain in order to express dissatisfaction with economic outcomes.

According to the expressive abstention argument, Hypothesis 2b represents the opposite effect of polarization on abstention. That is, ${\hat \beta _{3j}} \lt 0$ in the equation for the trade-off between the share of voters who cast a ballot for the incumbent party and the share of voters who abstain— $\log \left( {{{{\rm{Incumben}}{{\rm{t}}_{it}}}}\over{{{\rm{Abstentio}}{{\rm{n}}_{it}}}}} \right)$ . As economic conditions deteriorate – that is, the unemployment rate increases – more voters would abstain from punishing the party in government. In other words, political polarization would exacerbate the effect of the unemployment rate on the trade-off between a partisan vote and abstention in response to economic conditions.

Findings

In this section, I present the results of county-level analyses that interact the unemployment rate with political polarization. I focus on the coefficients for the economic indicator and its interaction with polarization. Full tables of coefficients are available in a Supplementary Material document.Footnote ²⁵ Each panel in Figure 4 displays the marginal effects of unemployment across levels of political polarization in one of the log-ratios estimated with Equation 2. Each icon indicates the predicted value of the marginal effect – in the log-ratio form – at a certain level of polarization with its respective 95 per cent confidence interval.Footnote ²⁶ These marginal effects are calculated as ${\hat \beta _{1j}} + {\hat \beta _{3j}}*Po{l_t}$ . Negative values mean that unemployment benefits the category in the denominator at the expense of the one in the numerator.

Figure 4.

Marginal Effects of Unemployment (Log-ratio Form) across Levels of Polarization.

Note: Light-green triangles and dark-red upside-down triangles represent the marginal effects of unemployment when polarization is moderate (0.6) and high (0.85). These marginal effects are computed as ${\hat \beta _{1j}} + {\hat \beta _{3j}}*Po{l_t}$ .

The marginal effect of the unemployment rate on the trade-off between the incumbent party and the opposition ( ${\hat \beta _{1j}}$ from Equation 2) is negative and statistically significant (Column 1 in Table S3). This indicates that, in a hypothetical scenario of no or extremely low polarization, unemployment would have the negative effect expected by economic voting theory. In other words, as the unemployment rate increases, the share of voters who cast a ballot for the opposition increases as well – negative effects benefit the category in the denominator – at the expense of the party in government. Likewise, unemployment has a similar effect on the trade-off between the share of voters supporting the incumbent party and the share of those who abstain (Column 3 in Table S3). Of course, a scenario with no polarization is an implausible condition not observed in real-world data. As such, Figure 4 shows the marginal effect of unemployment across a moderate and a high level of polarization observed in the data.

Panel A in Figure 4 shows that moving from a moderate level of polarization observed for a long period in American politics – 0.6, the light-green triangle – to a high level similar to what exists nowadays – 0.85, the dark-red, upside-down triangle – drastically decreases the magnitude of the effect of unemployment in the trade-off between the incumbent party and opposition. This result supports Hypothesis 1: polarizing politics reduces the number of voters crossing party lines to sanction incumbents for poor economic conditions. The same pattern is observed in the trade-off between the vote share for the incumbent party and the share of abstainers (Panel B in Figure 4). This finding supports Hypothesis 2a, which states that polarization reduces the number of voters who abstain in response to a deteriorating economy to express dissatisfaction with the party in government. Consequently, these results reject the expressive abstention effect predicted by Hypothesis 2b.

Besides corroborating the expectation that polarization attenuates economic voting, these findings show that the moderating effect of polarized environments is substantial. As polarization increases, the impact of the unemployment rate tends to disappear. The marginal effect of unemployment in Panel A is close to zero when polarization is equal to 0.85 – a typical value observed in the late 2000s. It indicates that polarizing politics almost cancels out economic voting in response to local unemployment. Similarly, polarization decreases the share of voters moving from the incumbent party towards abstention in response to unemployment (Panel B). The effect on the trade-off between abstention and the incumbent party becomes about 50 per cent smaller as polarization moves from 0.6 to 0.85.

These results confirm that voters have two channels to punish incumbents for a poor economy – voting for the opposition and abstention – and demonstrate that polarizing environments affect both in the same way. Polarization decreases the number of voters willing to desert the incumbent party for poor economic performance. Under polarized politics, voters are so attached to their preferred parties that they are less likely to defect to the opposing party or abstain when the economy declines.

The coefficients in Figure 4 are substantively uninformative because they show marginal effects in log-ratio form. We can use them to interpret the direction of the marginal effects and whether they are statistically distinguishable from zero or not. But to capture the size of these marginal effects in terms of changes in the share of voters in each category, I follow the strategy presented by Philips, Rutherford and Whitten (Reference Philips, Rutherford and Whitten2016) to visualize the results of models with compositional data. I simulate a non-parametric bootstrap 1,000 times to estimate the impact of a one-standard-deviation increase in the unemployment rate of a hypothetical county in Pennsylvania under two scenarios, moderate (0.6) and high (0.85) ideological distance between parties at the national level.Footnote ²⁷

Since my models use state fixed effects, I employ the checklist suggested by Mummolo and Peterson (Reference Mummolo and Peterson2018) to improve the interpretation of fixed-effect regressions. They argue that simulations must use within-unit typical shocks since fixed-effect models restrict the analysis to within-unit variation. Hence, I simulate the estimated difference in the share of voters for each category $j$ – the incumbent party, opposition, and abstention – when the unemployment rate increases by a within-state standard deviation (a 2.3 per cent increase) from its mean (6.12 per cent).

Figure 5 presents the results of these simulations. When the ideological distance between parties is moderate (0.6), a shock that increases the unemployment rate by one standard deviation in the county decreases the vote share of the incumbent party by approximately 2 per cent in that county. The size of this punishment for bad economic conditions becomes smaller as polarization increases. When polarization is equal to 0.85, the same shock in unemployment diminishes the vote share of the party in government by just 1 per cent. These results are similar to marginal effects estimated with OLS that regress the share of votes for the incumbent party on polarization and unemployment.Footnote ²⁸ But with SUR models, I can also simulate changes in the other two categories. The gain in vote share for the opposition due to unemployment tends to vanish as polarization goes up. It moves from 1 per cent to practically no change in the share of voters moving towards the opposition. Equally, the increase in the share of voters who choose to abstain after a shock in the unemployment rate decreases from more than 1.5 per cent to approximately 1 per cent as polarization moves from 0.6 to 0.85.

Figure 5.

Predicted Change in the Share of Voters.

Note: Results generated with bootstrap. The y-axis is the change in the predicted share of voters in a category j (incumbent, opposition, and abstention) when there is a one-standard-deviation increase in the unemployment rate from its mean.

These results illustrate the attenuating impact of polarization on economic voting predicted by Hypotheses 1 and 2a. Political polarization negatively affects both channels through which voters can punish incumbents. Under polarized politics, voters stick with their parties instead of penalizing them for poor economic performances. These findings differ from previous works that did not find an effect of local economic indicators on vote choice (Healy and Lenz Reference Healy and Lenz2017) or did not identify changes in economic voting over time (de Benedictis-Kessner and Warshaw Reference de Benedictis-Kessner and Warshaw2020). These studies specify the dependent variable as the Democratic vote or change in the Democratic vote share which are typically stable at the local level due to party loyalties in the electorate. The outcome in Equations 1 and 3, in turn, measures the trade-off between the president’s party and one of the other two categories in the electorate (the opposing party or abstention). This outcome may capture more nationwide oscillations in voting behaviour across successive elections than local attachment to parties.Footnote ²⁹ As such, it may represent more national variations than the dependent variables of previous works. Moreover, by including election fixed effects, other studies eliminated any common trends and temporal changes (for example, polarization), limiting the analyses to differences across localities (counties) while holding constant all shifts over time (Kittel and Winner Reference Kittel and Winner2005, 272; Kropko and Kubinec Reference Kropko and Kubinec2020, 8).Footnote ³⁰ In this paper, the model in Equation 2 is dynamic, comparing changes in vote choice across successive elections without time fixed effects. This strategy can estimate how increasing polarization has influenced voters’ reactions to economic conditions. As an observation study, however, this specification may raise concerns about other national-level changes that could also affect vote choice. To mitigate these concerns, I employ the empirical strategies described in the last subsection: LDVs, time-variant controls, a time-trend robustness check, and the individual-level analyses in the following section.Footnote ³¹

Perceived Polarization and Economic Assessment

According to Hypothesis 3a, polarization reduces partisans’ propensity to negatively evaluate the state of the economy when their party holds the Presidency. I use ANES data from 1992 to 2020 to estimate the impact of the reported ideological distance between parties – which measures how respondents perceive political polarization – on economic assessment. In the survey analyses, I use respondents’ perceptions of polarization to estimate the impact of polarized politics on economic voting. I do so because polarizing environments should affect voting behaviour through how voters sense ideological distancing between political parties. Moreover, increasing observed polarization is expected to increase the perceived ideological gap between parties among voters (Hetherington Reference Hetherington2001; Smidt Reference Smidt2017). In the subsection before the conclusion, I discuss the association between observed and perceived polarization. I measure perceived polarization as the gap between the location of the two parties reported by the respondent on a 7-point scale that ranges from extremely liberal to extremely conservative.Footnote ³² After normalizing this variable, the perceived gap ranges from 0 to 1.Footnote ³³ To measure how respondents evaluate economic conditions, I use the question that asks whether the national economy ‘has gotten better, stayed about the same, or gotten worse’ over the past year before the election.

To investigate how partisans evaluate the economy as the perception of polarization increases, I estimate an ordered logistic model that regresses economic evaluation on an interactive model specified as follows:

(3) $${\rm{Pr}}({y_i} = e) = F({\beta _1}{\rm{IdeoGa}}{{\rm{p}}_i} + {\beta _2}{\rm{Partisa}}{{\rm{n}}_i} + {\beta _3}{\rm{IdeoGa}}{{\rm{p}}_i}^*{\rm{Partisa}}{{\rm{n}}_i} + {{\boldsymbol\delta} _k{\boldsymbol x}_{ki} + {\boldsymbol \alpha} { }\bf Electio{n}_i} + {\varepsilon _i})$$

where the probability that respondent $i$ will select the category $e$ – the economy has gotten better, stayed the same, or gotten worse – is a function of her perception of the ideological gap between the two major parties ( ${\rm{IdeoGa}}{{\rm{p}}_i}$ ), a partisanship variable that indicates if the respondent is a member of the president’s party, Independent, or a member of the opposing party ( ${\rm{Partisa}}{{\rm{n}}_i}$ ), an interaction between these two variables ( ${\rm{IdeoGa}}{{\rm{p}}_i}^*{\rm{Partisa}}{{\rm{n}}_i}$ ), a set $k$ demographic controls ( ${x_{ki}}$ ), and a vector of election fixed effects ( $Electio{n_i}$ ).Footnote ³⁴

Figure 6 shows predicted probabilities that a respondent will report that the economy ‘has gotten worse’ across levels of perceived polarization (shapes). In support of Hypothesis 3a, as perceived polarization increases, members of the party holding the presidency are less likely to negatively evaluate the economy. The probability a supporter of the president’s party will report that the economy ‘has gotten worse’ decreases from 0.35 to 0.15 as their perception of the ideological distance between parties moves from 0.1 to 1. This perceived polarization makes members of the opposing party, in turn, more likely to report a negative evaluation of the economy: the predicted probability of a negative economic assessment moves from 0.53 when the ideological gap is 0.1 to 0.68 when the reported distance between parties is 1. The ideological gap has no systematic effect on Independents. These findings corroborate Hypothesis 3a and previous studies (Donovan et al. Reference Donovan, Kellstedt, Key and Lebo2020; Goidel and Kellstedt Reference Goidel and Kellstedt2022), demonstrating that perceived polarization reduces partisans’ disposition to negatively assess the economic performance of their preferred party.Footnote ³⁵

Figure 6.

Probability of a Negative Evaluation of the Economy.

Note: The horizontal axis indicates if the respondent is attached to the president’s party, the opposition, or is an Independent. Shapes indicate the ideological gap between parties perceived by respondents. Full results can be found in Table S4.

Perceived Polarization and Economic Accountability

As Figure 2 illustrates, polarized politics can affect another link in the economic voting chain. Among members of the president’s party who negatively evaluate the economy (Ang et al. Reference Ang, Reeves, Rogowski and Vishwanath2021), polarization can weaken their willingness to use this economic assessment to inform their vote choice. In other words, partisans become less likely to punish their preferred party for poor economic conditions as polarization increases. This is the expectation of Hypothesis 3b. To test it, I split the ANES data described above by respondent’s partisan attachment ( ${\rm{Partisa}}{{\rm{n}}_i}$ in Equation 3). For each subsample – members of the president’s party, members of the opposing party, and Independents – I use multinomial logistic regressions to estimate the probability that a respondent $i$ selects category $m$ – voting for the incumbent party, the opposition, or abstention – as a function of the following interactive specification:Footnote ³⁶

$${\rm{Pr}}({y_i} = m) = F({\beta _{1j}}{\rm{EconEva}}{{\rm{l}}_i} + {\beta _{2j}}{\rm{IdeoGa}}{{\rm{p}}_i} + {\beta _{3j}}{\rm{EconEva}}{{\rm{l}}_i}^*{\rm{IdeoGa}}{{\rm{p}}_i} + $$

(4) $${{\boldsymbol\delta} _{kj}{\boldsymbol x}_{ki} + {\boldsymbol{\gamma _j}}{\boldsymbol{Stat{e_i}}} + {\boldsymbol \alpha _j}{\bf Election}_i} + {\varepsilon _i})$$

where the probability of a respondent $i$ selecting category $m$ is a function of her evaluation of the economy, ${\rm{EconEva}}{{\rm{l}}_i}$ , her perception of the ideological gap between the two major parties, ${\rm{IdeoGa}}{{\rm{p}}_i}$ , the interaction between these variables ( ${\rm{EconEva}}{{\rm{l}}_i}^*{\rm{IdeoGa}}{{\rm{p}}_i}$ ), the same set of $k$ demographic controls ( ${{\bf{x}}_{ki}}$ ) in Equation 3, and vectors of state and election fixed effects ( $Stat{e_i}$ and $Electio{n_i}$ ). With this model specification, I can investigate how partisans react to economic evaluations under different presidencies and levels of the perceived ideological gap between parties.Footnote ³⁷

Figure 7 presents predicted probabilities of a respondent to vote for the incumbent party (Panel A), for the opposition (Panel B), or to abstain (Panel C), given that she negatively evaluates the economy. Icons represent the level of the perceived ideological distance between parties. In support of Hypothesis 3b, as the perceived ideological distance between parties increases, supporters of the president’s party who negatively evaluate the economy become less likely to punish the incumbent party by either voting for the opposition or not participating. When the ideological gap is 0.1, the probability that a member of the president’s party who reports that the economy is deteriorating will vote for the incumbent party is equal to 0.60. If s/he locates parties at extreme sides of the spectrum – that is, an ideological gap equal to 1 – this probability goes up to 0.80. Supporting Hypotheses 1, 2a, and 3b, Panels B and C show that under higher polarization, supporters of the president’s party are less likely to sanction their party for economic conditions by voting for the opposing party or abstaining.Footnote ³⁸ The probability a supporter of the incumbent party will cast a ballot for the opposition in response to the deteriorating economy decreases from 0.22 to 0.09 as the perceived ideological distance between parties moves from 0.1 to 1 (Panel B in Figure 7).Footnote ³⁹

Figure 7.

Vote Choice Given a Negative Economic Evaluation.

Note: Panels indicate respondents’ vote choice. Shapes indicate the ideological gap between parties perceived by respondents. Full results in Table S5.

Supporters of the opposing party and Independents who negatively evaluate the economy become more likely to vote for the opposition. Independents reporting a negative economic assessment present a small probability of voting for the incumbent party regardless of their perception of the ideological distance between the two major parties. Moreover, consistent with the literature showing that nonpartisans are less interested and involved in politics (Lewis-Beck et al. Reference Lewis-Beck, Norpoth, Jacoby and Weisberg2008; Magleby, Nelson and Westlye Reference Magleby, Nelson and Westlye2011), Independents are always more likely to abstain than partisans. However, their propensity to abstain in response to a negative assessment of the economy decreases as they become more able to distinguish the ideological difference between parties. The probability of voting for the opposition among Independents who negatively evaluate the economy increases with the perceived ideological gap at the expense of abstention. These findings are in line with the results of previous studies which demonstrate that polarization clarifies the ideological positions of political parties for ordinary citizens and, as a result, reduces ambivalence among Independents (Hetherington Reference Hetherington2001; Smidt Reference Smidt2017). Moderate voters who perceive a larger ideological gap between parties are not only less likely to move between parties across elections (Smidt Reference Smidt2017); they are also less likely to abstain in response to a negative economic evaluation.Footnote ⁴⁰

To sum up, individual-level results support Hypothesis 3b and the patterns predicted by the sticking effect of polarization on electoral accountability. As the perceived distance between parties increases, members of the president’s party become more reluctant to punish their party for a negative evaluation of the economy.

Observed Polarization and the Perceived Ideological Gap

These micro-level analyses confirm my early findings with aggregate data. Results at both levels imply that polarization weakens economic voting. Yet, is perceived polarization – the perceived ideological gap between parties – associated with observed levels of political polarization? In line with Lupu’s (Reference Lupu2015) findings showing that citizens in polarized systems perceive their parties to be more polarized, Figure S2 in the Supplementary Material provides evidence to support an affirmative answer to this question. It shows that results at the county level are robust to models that use the average perceived ideological distance from ANES data instead of the distance between representatives in the House. Indeed, both observed elite polarization and the perceived ideological distance between parties are moving up over time (Lelkes Reference Lelkes2016; Smidt Reference Smidt2017). Figure S9 plots trends in observed and perceived polarization among voters by party attachment. Consistent with Lelkes’ (Reference Lelkes2016) results using the mass public’s placement of parties on policy issues, the perceived gap between parties increases in the electorate over time (yellow line). Partisans are driving these results (blue and red lines).

To further observe the association between political polarization and voters’ perceptions about the ideological distance between parties, I regress the perceived ideological gap reported by ANES respondents on an interaction between observed polarization and partisanship, controlling for all variables described in Equation 4. Results from this regression cannot be interpreted as a causal relationship, given that the model does not address issues concerning reverse causality. Yet, they indicate a strong relationship between observed party polarization and how voters perceive it. Figure S10 in Supplementary Material plots the marginal effects of this interaction. Polarization is positively associated with the perceived ideological gap across all categories of party identification but it is stronger among Republican identifiers.Footnote ⁴¹ These results suggest that observed polarization influences perceptions of polarization in the electorate, especially among partisans. Hence, the moderating effect of polarized politics on economic voting observed with the aggregate-level data is likely operating through how voters perceive the ideological difference between parties (Hetherington Reference Hetherington2001; Smidt Reference Smidt2017).Footnote ⁴²