Inflation in 2022 did not affect congressional voting, but abortion did

Significance

Economic theories of voting dominate public understanding of elections in the United States, largely on the basis of cross-sectional evidence that negative (positive) perceptions of the national economy prompt voting against (for) the incumbent party. In this study, we demonstrate why the role of the economy is easily misinterpreted in research on American elections. We provide evidence that, despite severe inflation and grave concerns about deteriorating economic conditions, changing perceptions of the economy did not affect voting shifting from the 2020 to the 2022 US midterm elections. This study demonstrates that views on abortion and the Supreme Court decision in Dobbs v. Jackson Women's Health Organization directly affected voting in 2022.

Abstract

This study examines voting in the 2022 United States congressional elections, contests that were widely expected to produce a sizable defeat for Democratic candidates for largely economic reasons. Based on a representative national probability sample of voters interviewed in both 2020 and 2022, individuals who changed their vote from one party's congressional candidate to another party’s candidate did not do so in response to the salience of inflation or declining economic conditions. Instead, we find strong evidence that views on abortion were central to shifting votes in the midterm elections. Americans who favored (opposed) legal abortions were more likely to shift from voting for Republican (Democratic) candidates in 2020 to Democratic (Republican) candidates in 2022. Since a larger number of Americans supported than opposed legal abortions, the combination of these shifts ultimately improved the electoral prospects of Democratic candidates. New voters were especially likely to weigh abortion views heavily in their vote-shifting calculus. Likewise, those respondents whose confidence in the US Supreme Court declined from 2020 to 2022 were more likely to shift from voting for Republican to Democratic congressional candidates. We provide direct empirical evidence that changes in support for the Supreme Court, a nonpartisan branch of the federal government, are implicated in partisan voting behavior in another branch of government. We explore the implications of these findings for prevalent assumptions about how economic conditions influence voting, as well as for the relationship between the judiciary and electoral politics.

On the eve of the 2022 midterm elections, scholars, pundits, and politicians widely predicted that Democrats in the United States Congress would experience an unusually sizable loss of seats. This expectation stemmed from mounting inflation and concerns about the economy on the part of the American public. One national poll conducted during the summer of 2022 found that one-third of respondents reported that inflation was the biggest concern facing their family and well over half reported that economic conditions more generally were their most significant worry (1). Many observers concluded that the effect of inflation “on working-class voters [could] be disastrous for Democrats in the midterm elections” since Americans were likely to hold President Joe Biden and the Democratic Party responsible at the ballot box for the rapid and alarming rise in the price of many products, as well as for the perceived deterioration of the US economy (2).

Despite these dire predictions, the Democrats retained control of the Senate and lost control of the House of Representatives by only the narrowest of margins. The nine House seats lost by Democrats in 2022 was far less than the average of 27 seats lost in post–World War II midterm elections by the incumbent president’s party, and far fewer than the 47 seats predicted by traditional models (3).

In this study, we address why predictions that the Democratic Party would suffer a substantial defeat in 2022 were wrong. In so doing, we also question whether the longstanding emphasis on the economy in studies of US voting is warranted (4). Indeed, many existing studies maintain that Americans engage in retrospective economic voting, whereby citizens reflect on whether they or the country as a whole has improved or declined economically, and then support or oppose candidates of the incumbent’s party on that basis.

Journalists frequently assert that Americans are “pocketbook voters,” relying on their financial self-interest in making voting decisions. However, the posited influence of self-interest easily outstrips its actual impact on attitudes and political behavior (5). Scholars of political behavior, on the other hand, emphasize that perceptions of collective national well-being bear more heavily on voting than personal economic circumstances (6).

Regardless, during the run-up to the 2022 midterm elections, many scholars predicted that inflation would have a greater impact on vote choice than abortion, primarily based on survey evidence. Americans in October of that year were far more likely to identify the economy or inflation as the “most important” problem facing the country than abortion, regardless of their party identification (7, 8). Although there is little evidence that subjective issue importance reflects what actually guides people’s vote choice, it often serves as the basis for election predictions (9).

Interpretations of why individuals vote as they do are seldom well-documented. More commonly, observers reason after the fact based on the outcome of an election to establish which issues were most consequential. This practice frequently results in misinterpreting the message conveyed by voters (10, 11). In reality, most issue positions encourage support for one party among some voters, while simultaneously discouraging support for that same party among other voters (12).

Since elections have a variety of moving parts, who won or lost is an inadequate basis for determining what shaped the result. Nonetheless, the economy retains an especially prominent status in election research, in part because all voters prefer a robust and well-functioning national economy. In addition, information about the economy is widely available to Americans on a day-to-day basis in a way that other issue information often is not.

But there are multiple reasons to doubt whether the well-worn maxim, “It’s the Economy, Stupid,” explains voting in the contemporary electoral climate. First, most voting is highly habitual. The overwhelming majority of people repeatedly vote for candidates of the same party, the economy notwithstanding. Increasing affective polarization has contributed to even more strongly held partisan identities for many Americans (13). More intense partisan preferences should make economic voting less likely to occur (14).

Stable partisan identities not only influence voting behavior, they also influence perceptions of the economy and attributions of responsibility for economic improvement or decline. Individuals who support the incumbent president’s party systematically perceive the economy in rosier terms than those who support the party that is out of office; these perceptions reverse direction when the party in the White House changes (15). In other words, this same pattern may stem from reverse causation; that is, from citizens affiliated with the incumbent party judging existing economic conditions to be better than those affiliated with the outparty (15–19). Likewise, attributions of responsibility for the economy are rationalized to deflect blame from their preexisting preferences (20). Members of the outparty may attribute economic downturns to the inparty, but members of the inparty will find another culprit to hold responsible.

Second, regardless of partisanship, voters seldom switch the party for which they cast ballots over time. In presidential elections, vote switchers typically hover around 10%. Vote switching in midterm elections is generally even rarer. Such regularity characterizes the midterms in part because congressional contests tend to be low information events. Relative to presidential elections, there is far less information about congressional candidates that might turn people away from their default predispositions. In addition, midterm voters tend to be more partisan than voters in presidential election years.

A third possible explanation for the surprising outcome in 2022 is that past evidence of economic voting may not be as robust as many observers believe. Most evidence for retrospective economic voting is based either on aggregate-level data, or on cross-sectional data correlating actual or perceived economic conditions with vote intentions. Such relationships are typically statistically significant, with positive or improving economic conditions associated with support for candidates of the incumbent party. Likewise, issue opinions may be correlated with voter preferences, but it is typically unclear whether voters’ issue opinions are being influenced by favorably regarded candidates or, instead, candidates are favored due to championing issue opinions that are already congruent with the opinion of voters.

Evolving understandings of economic voting reflect growing concerns about the strength and consistency of this evidence. On the one hand, “Both officeholders and political observers believe that election returns vary systematically with the economic conditions which prevail in the period prior to elections” (21). Consistent with this belief, a seminal aggregate-level study by Tufte found that a thriving economy significantly reduced the loss of seats in Congress by the incumbent president’s party (22). Subsequent studies also found that the president’s party fared better in midterm elections when the national economy was healthy (23–25).

But other studies have concluded that the national economy has neither a direct nor an indirect effect on midterm elections (3, 26–28). In his review entitled “The End of Economic Voting?” Anderson summarized this evidence, arguing that “economic voting does not function as envisioned by advocates of democratic accountability.” (4) A more recent review of retrospective economic voting concludes somewhat more optimistically that we have yet to achieve an understanding of “the circumstances under which retrospective voting achieves effective democratic accountability and when it fails to do so” (29). Because the public is directly exposed to the economy in ongoing ways, democratic accountability by means of a referendum on the economy is assumed to be the most likely form of issue accountability.

A fourth concern is that even if the health of the economy is, indeed, an important influence on voting, it is unclear which features of the economy are most salient to American voters. The 2022 midterm elections presented voters with an unusual and confusing economic scenario. Joblessness in the US was nearing a historic low. Unemployment had declined steadily since the height of the COVID-19 pandemic, dipping below 4%. At the same time, by November 2022 inflation had surged to its highest rate in over 40 y. Further, while mounting inflation seemed to augur poorly for Democratic candidates in 2022, past microlevel evidence indicates that inflation generally has less influence on vote choice than growth and employment (4, 30, 31).

The 2022 midterm elections provided an ideal setting for evaluating the impact of inflation and unemployment. In most elections over the past 40 y, voters have displayed little awareness of—or concern about—inflation, largely because it has been low and stable. By 2022, however, inflation was so severe that virtually everyone was aware of it. Based on our data, over 90% of Americans recognized that the cost of goods had increased during the past year. Moreover, unlike unemployment, which directly affects only a small percentage of the public even when it is elevated, inflation affected all consumers. Nonetheless, it remains an open question whether Americans’ perceptions of increasing inflation influenced voting.

Evaluating Issue Impact

To evaluate how inflation—and economic conditions more generally—affected the 2022 midterms, we analyze two waves of a probability panel survey that was fielded in 2020 and 2022 by the National Opinion Research Center at the University of Chicago. Respondents were recruited using address-based probability sampling, interviewed in either English or Spanish, and by telephone or by internet, based on their preferences (32). A panel design is essential for these purposes because it allows us to account for the high degree of persistence in voting behavior over time. Of the 3,053 people interviewed in October 2020, 81% were re-interviewed in October 2022. Details on the representativeness of the sample are included in SI Appendix, Table S1, along with exact question wording.

To analyze the role of perceptions of inflation and economic issues more generally, we used two-way fixed effects logit regressions, an extension of difference-in-differences. This approach has two important features. First, comparing voters at the individual level to themselves over time is the best approach for avoiding model specification bias in observational analyses. Temporally stable characteristics of individuals are automatically eliminated from these models as potential spurious influences. Second, by controlling for the average impact of all other unspecified changes between 2020 and 2022 through the inclusion of a dummy variable for panel wave, we eliminate still more sources of omitted variable bias. Relative to other approaches to panel data such as lagged dependent variables, fixed effects eliminates far more potential model misspecification errors (33).

For our dependent variable, we assigned values in both years, distinguishing voting for a Republican (1) from voting for a Democrat or a third-party candidate (0). Panelists who did not vote in 2020, but did vote in 2022, were included in the sample as 0s in 2020 to account for the influence of new voters on the election outcome. However, we also analyze the sensitivity of our results to the introduction of these new voters in the sample. Nonvoters in 2022 were excluded from our analyses since they could not influence the 2022 election outcome. Thus, our analyses of change over time reflect both changes in turnout from 2020 to 2022, which could have favored either Republicans or Democrats, as well as voters who changed party/candidate preferences. In practice, however, only 2% of panelists were new voters in 2022.

Our analysis was designed to determine whether change toward voting for the Republican candidate in 2022, the direction suggested by retrospective economic voting, was predicted by changing economic perceptions. Given that substantial and worsening inflation at the time of the 2022 election was expected to be a salient issue influencing voters, we modeled the impact of current inflation perceptions on vote switching using the interaction between Wave (2020 to 2022) and the perceived severity of Inflation. A statistically significant and positive coefficient of the interaction term would indicate that those individuals who perceived inflation as particularly severe (i.e., they perceived that the cost of consumer goods had increased a lot in the past year) were more likely to change their votes in the direction of Republican candidates than those who viewed inflation as less severe. We later use this same approach with opinions on Abortion and perceptions of Unemployment, two additional political issues that may have influenced vote switching in 2022.

For time-varying independent variables measured for the same individuals in both 2020 and 2022—including perceptions of the national economy, personal finances, favorability toward the Supreme Court, and Biden’s job approval—we estimate the effects of change in each of these factors on change in vote intentions. To further examine the potential electoral impact of views on abortion, we modeled changes in voter preferences as a function of individual changes from 2020 to 2022 in evaluations of the Supreme Court, changes that stemmed primarily from reactions to the Dobbs v. Jackson Women’s Health Organization decision handed down in June 2022.

These two approaches to testing for issue impact on vote change correspond to the two central ways in which issues potentially impact voting. In one case, it is not that people’s opinions on an issue change, but instead that the salience of existing opinions on an issue increase. We hypothesized that abortion views would be more salient after the Supreme Court overturned Roe v. Wade. In what is often referred to as “priming” or heightened issue salience, preexisting opinions become more likely to influence voting relative to their influence in a previous election, even when those opinions do not change.

The second way that issues precipitate change in voter preferences is when perceptions or opinions themselves change, as with perceptions of a declining economy, and this triggers changes in voter preferences. In the case of perceptions of declining economic conditions, citizens should be more likely to shift their votes away from the incumbent president. Unfortunately, analyses of both kinds of issue-based vote switching are plagued by the weaknesses of causal inference using cross-sectional analyses. Such analyses make it impossible to distinguish preexisting issue-candidate alignments from changes in support for candidates that result from changes in issue opinions or changes in issue salience.

Results

The Impact of Inflation.

We first conduct a simple test of the hypothesis that severe inflation caused some voters to switch toward Republican candidates in 2022. Table 1 includes odds ratios associated with each variable that are derived from the fixed effects logit regressions. An odds ratio greater (less) than 1.0 indicates that higher (lower) values of a given variable increase (decrease) the likelihood of switching from voting for a non-Republican in 2022 to a Republican in 2022. As shown in Table 1, Model 1, there is no evidence that perceptions of either Inflation or Unemployment had any bearing on vote switching toward the Republican party from 2020 to 2022. To the extent that these null findings are due to Americans uniformly viewing inflation as more severe in 2022, we should observe a significant Wave odds ratio in Table 1, but this is not the case. Instead, the only significant predictor of vote changing toward Republican candidates was change in voters’ opinions of Biden from 2020 to 2022. Consistent with previous research, increasingly unfavorable views of the incumbent stimulated a greater likelihood of shifting from voting for non-Republican candidates in 2020 to Republican candidates in 2022, as indicated by the statistically significant odds ratios corresponding to the Biden Thermometer that are consistently less than 1.0.

Table 1.

Effects of change in economic perceptions on change in congressional vote choice

	Model 1	Model 2
Wave (2020 to 2022)	0.195	0.209
	[0.016, 2.375]	[0.017, 2.637]
Wave × Inflation (2022)	1.146	1.150
	[0.710, 1.850]	[0.708, 1.867]
Wave × Unemployment (2022)	1.232	1.122
	[0.921, 1.648]	[0.828, 1.521]
Biden Thermometer	0.855**	0.853**
	[0.786, 0.929]	[0.783, 0.929]
National economy		1.141
		[0.942, 1.382]
Personal finances		1.136
		[0.879, 1.468]
Log likelihood	−141.775	−138.065
Number of respondents	218	216
Number of observations	436	432

Note: Entries are odds ratios with 95% CIs in parentheses, based on fixed effects logit analyses of individual-level congressional voting changes from 2020 to 2022. **P < 0.01, *P < 0.05.

Model 2 in Table 1 expands on the previous results by accounting for change over time in perceptions of the National Economy and respondents’ Personal Finances. Our findings suggest that neither changing personal nor national economic perceptions had an effect on the likelihood of changing votes toward Republican candidates. Nonetheless, given the consistent impact of changes in favorability toward Biden, it is plausible that the impact of Inflation in Table 1 could have occurred indirectly by affecting evaluations of Biden. To evaluate this possibility, we ran these same analyses without attitudes toward Biden in the model (SI Appendix, Table S2). Across all models, the impact of economic variables remained null, as did the Wave variable.

Given the constraints of fixed effects logistic regression and the fact that just over 10% of the sample changed vote preferences from 2020 to 2022, it is also possible that these null results are due to an underpowered sample. For this reason, in SI Appendix, Table S3, we re-estimate the models in Table 1 using fixed effects linear regressions that include all voters, including those who did not switch the party of the congressional candidate they supported from 2020 to 2022. Again, people with negative impressions of inflation or unemployment were no more likely to change candidates, nor did changes over time in perceptions of national economic conditions or changes in perceptions of personal finances influence vote switching. These null findings are especially surprising given that within-subject tests of this kind are capable of statistically identifying very small effects.

Importantly, cross-sectional analyses of these same data yield significant relationships between vote choice and perceptions of the severity of Inflation, Personal Finances, and (in model 3) the National Economy, respectively (SI Appendix, Table S4). This type of evidence regularly leads observers to potentially inaccurate conclusions about the sources of change in voting from election to election. In 2022, supporters of Republican candidates believed economic conditions were worse and that inflation was more severe than supporters of Democratic candidates. These results aptly illustrate the problem of interpreting election outcomes based on cross-sectional findings. Relationships such as those in SI Appendix, Table S4 are unlikely to indicate that the economy or inflation changed people’s votes; they are more likely to be rationalizations based on preexisting political preferences.

Americans were widely aware of mounting inflation when they went to the polls in 2022. Why did these economic conditions have no electoral impact? In Fig. 1, we illustrate our respondents’ attributions of responsibility for inflation, which are either starkly partisan or completely nonpartisan. Democrats were most likely to blame Republicans or neither political party. Republicans were overwhelmingly likely to blame Democrats, and Independents blamed “both parties.” Perhaps most interestingly, well over half of this representative national probability sample—roughly 55%—held either “neither party” or “both parties” responsible. This pattern obviously blunts inflation’s potential electoral impact. People either blamed the opposing party or they did not assign responsibility to any single party.

Fig. 1.

Partisan attributions of responsibility for inflation, by party identification. Note: Respondents were asked, “Regardless of which political party you favor, which political party do you see as most at fault for rising consumer prices?”

The Influence of Abortion Preferences.

To put the null effects of these economic factors into a broader perspective, we address a second political issue believed to be salient in 2022: opinions on the legality of abortion. To this end, we analyze the electoral impact of individual attitudes toward abortion, operationalized in two different ways, coinciding with the two ways in which issue positions can influence vote switching. The first is a 2022 measure of opinions on abortion, an issue that became more salient that year due to the Dobbs decision. We expected that individuals who favored making abortion illegal (legal) in 2022 should be more likely to switch their votes toward (away from) Republicans from 2020 to 2022. A second test of whether abortion may have affected vote switching relies on a proxy: the change in an individual’s attitude toward the Supreme Court from before to after the Dobbs decision. Because we had both 2020 and 2022 measures of favorability toward the Court, we hypothesized that individuals who increased (decreased) support for the Court post-Dobbs should be more (less) likely to switch their votes toward Republican congressional candidates. Although there are certainly other reasons people may have changed their opinions of the Court, Dobbs was by far the most widely covered and well-known decision during the period from 2020 to 2022.

In Models 1 and 2 of Table 2, we test the hypothesis that to the extent that people favor making abortion illegal (legal), they should be more likely to change their votes in the direction of Republican (Democratic) candidates from 2020 to 2022. Like Table 1, the results shown in Table 2 are odds ratios derived from fixed effects logit regressions. As indicated by the odds ratios greater than 1.0 corresponding to Wave × Abortion opinion, among those individuals who opposed (favored) legal abortion, changes in voting between 2020 and 2022 shifted significantly toward Republican (Democratic) candidates. As shown in SI Appendix, Table S5, the results based on the larger sample in an ordinary least squares fixed effects model are similar.

Table 2.

Effects of abortion opinions and changes in favorability toward the Supreme Court on change in congressional vote choice, 2020 to 2022

	Model 1	Model 2	Model 3	Model 4
Wave (2020 to 2022)	0.075	0.089	0.244	0.239
	[0.005, 1.055]	[0.006, 1.311]	[0.018, 3.299]	[0.017, 3.356]
Wave × Inflation (2022)	1.156	1.158	1.136	1.157
	[0.714, 1.871]	[0.710, 1.887]	[0.691, 1.868]	[0.699, 1.916]
Wave × Unemployment (2022)	1.228	1.136	1.185	1.087
	[0.912, 1.653]	[0.833, 1.551]	[0.880, 1.597]	[0.796, 0.483]
Wave × Abortion opinion (2022)	1.491*	1.438*
	[1.072, 2.074]	[1.023, 2.022]
Supreme Court favorability			1.500**	1.485*
			[1.110, 2.029]	[1.093, 2.017]
Biden Thermometer	0.856**	0.854**	0.851**	0.850**
	[0.787, 0.931]	[0.784, 0.930]	[0.781, 0.928]	[0.778, 0.927]
National economy		1.129		1.107
		[0.927, 1.377]		[0.909, 1.348]
Personal finances		1.062		1.174
		[0.812, 1.389]		[0.901, 1.528]
Log likelihood	−136.074	−133.262	−136.797	−133.216
Number of respondents	214	212	216	214
Number of observations	428	424	432	428

Note: Entries are odds ratios with 95% CIs in parentheses, based on fixed effects logit analyses of individual-level congressional voting changes from 2020 to 2022. **P < 0.01, *P < 0.05.

As a second, somewhat stronger empirical test of abortion’s impact, Models 3 and 4 in Table 2 estimate the impact of individual-level change in favorability toward the Court from 2020 to 2022 on change in individual voter preferences. As shown, increases (decreases) in support for the Court systematically increased the likelihood of shifting votes toward Republican (Democratic) candidates. In all of these models, inflation, unemployment, and changes in general economic perceptions continue to have no bearing on vote switching. Table 2 thus confirms that whether measured directly or by a proxy (i.e., changes in favorability toward the Supreme Court), opinions about abortion altered voter preferences in 2022 while economic issues did not.

A key question for purposes of interpreting the impact of this issue on the election outcome is whether, on balance, the abortion issue generated greater gains for Republicans or Democrats. To address this question in simplest terms, the left side of Table 3 shows a tabulation of those voters who believe abortions should be illegal versus legal, crossed with those who changed votes in either direction. Since the overwhelming majority of citizens cast ballots consistent with their previous voting behavior, Table 3 focuses strictly on those who shifted preferences.

Table 3.

Vote switching by 2022 abortion and inflation opinions

	2022 opinions on abortion		2022 perceptions of inflation
	Legal in all or most cases, %	Illegal in all or most cases, %	Decreasing or staying the same, %	Increasing, %
Changed away from Republican congressional candidate	59.6% (36.0%)	39.3% (15.6%)	41.7% (2.2%)	51.6% (48.9%)
Changed toward Republican congressional candidate	40.4% (24.4%)	60.7% (24.0%)	58.3% (3.08%)	48.4% (45.8%)
Percentage of changers	100.00% (60.4%)	100.00% (39.6%)	100.00% (5.3%)	100.00% (94.7%)

Note: Cell entries are column percentages. Percentages in parentheses indicate the proportion of all vote changers’ opinions on abortion by direction of vote change. Abortion χ² = 8.82, P = 0.003; Inflation χ² = 0.45, P = 0.50. As shown in SI Appendix, Tables S6-1 and S6-2, there was no significant relationship between vote switching and change in views on other issues, including gun control, spending on social safety nets, climate change, and views on immigration. In addition, using the same analyses as above, in SI Appendix, Table S7 we show that these issues were not primed in 2022, and thus issue positions in 2022 bear no relation to vote switching.

The percentages in Table 3 illustrate the proportion of those holding specific abortion views who shifted votes in Republican or non-Republican directions as a percentage of each column. At first glance, it appears that abortion encouraged about as much switching toward Republicans as it did away from Republicans. Roughly 60% of vote switchers who thought abortion should be legal shifted their votes away from Republican candidates, and about 60% of vote switchers who thought abortion should be illegal shifted toward Republican candidates.

However, largely because of the much larger proportion of the population favoring legal abortion, most vote switching stemming from this issue benefited Democratic candidates. As shown by the percentages in parentheses in Table 3—representing the portion of all vote changers who fell into each category—Americans who favor legal abortion (60.4%) far outnumber those who think abortion should be illegal (39.6%). Due to this stark difference, only about 48% of all vote changers shifted toward Republican candidates, while almost 52% of them shifted toward a Democratic candidate, thus giving Democrats a small net boost.

Based on the results in Table 2, Fig. 2 illustrates the size of the effects of abortion opinions on the odds of vote switching in the 2022 elections. Setting all other variables at their means, we estimated the predicted probability of shifting toward a Republican candidate by specific abortion views. As shown in Fig. 2, the likelihood of shifting votes toward Republican (Democratic) candidates grows steadily larger as views on abortion become more (less) restrictive.

Fig. 2.

Predicted probability of shifting to a Republican congressional candidate, 2020 to 2022, by abortion opinion. Note: Based on fixed effects logit analysis in Table 2, Model 2, setting all other variables at their means. Due to the use of fixed effects logit, the predicted values illustrated above are strictly among those who switched vote preferences in one direction or another from 2020 to 2022.

Consistent with the population skew toward favoring legal abortion, far more people also experienced a drop in favorability toward the Supreme Court from 2020 to 2022 than an increase in favorability (32% versus 18.5%). Models 3 and 4 of Table 2 indicate that a one-point increase (decline) in our five-point measure of the favorability of voter opinions of the Supreme Court predicts an increase (decrease) by a factor of 1.5 in the odds of shifting toward Republican (Democratic) candidates. Notably, these predictions are based on those voters who changed preferences, not among the electorate as a whole. Whether directly examining how abortion opinions conditioned vote shifting, or when using support for the Supreme Court as a proxy, these are sizable effects for a single issue.

Is the abortion issue unique? As shown on the right side of Table 3, as well as in Table 2, inflation had no comparable relationship with vote switching. Further, as shown in SI Appendix, Tables S6-1 and S6-2, no other issue opinions in 2022 for which we were able to obtain data had a comparable impact, although increased disapproval of spending taxes on social safety nets also significantly predicted increasing (decreasing) support for Republican (Democratic) candidates. In addition, as shown in SI Appendix, Table S7, there is no evidence that any of these other issues were primed in 2022. Further, SI Appendix, Table S11 shows that after taking into account the impact of changing perceptions of the economy, perceptions of economic well-being were no more likely to cause changes in vote choice in 2022 than in 2020, thus countering the idea that the economy had been primed to have a greater influence on voting in 2022. Consistent with recent evidence on declining incumbency advantage (34), the inclusion of dummy variables indicating the presence of a Republican or Democratic incumbent congressperson also had no influence on the direction of vote switching.

The Impact of Voter Turnout.

Since our dependent variable combined both vote switchers and individuals who did not vote in 2020, but did in 2022, we do not know to what extent these findings are driven by changes in turnout from 2020 to 2022, as opposed to vote-switching among habitual voters. Because 2022 was a midterm election year, turnout was predictably lower than in 2020, a presidential election year. This difference complicates a direct comparison. But were new voters in 2022, or voters who did not vote in 2022 but did in 2020, distinctive in their views on abortion or in the extent to which they changed their views of the Supreme Court?

To answer this question, we compared abortion views and changes in favorability toward the Supreme Court among three turnout groups: 1) those who voted in both years, 2) those who dropped out of the electorate from 2020 to 2022, and 3) those who were new voters in 2022. As shown in SI Appendix, Table S8, these three groups did not differ in their views on abortion and only marginally differed in the extent to which they changed their views of the Court from 2020 to 2022. Thus, there is virtually no evidence that views differed systematically across voters who dropped out of the electorate in 2022, those who dropped in, and persistent voters.

Because comparing the likelihood of voting in a presidential year and a midterm year is not ideal, we conducted a parallel analysis drawing on turnout data for these same individual respondents in the 2018 midterm elections (SI Appendix, Table S9). Again, we found no differences in abortion opinions or in change in views of the Supreme Court among any of these three groups of voters. There was no indication that abortion proponents or opponents were more likely to be new voters in 2022, or that abortion proponents were any less likely to drop out of the electorate than abortion opponents (SI Appendix, Table S9).

Despite the null effects of voters’ opinions on changes in turnout, abortion could have been a particularly salient issue for voters who did not vote in 2020 but entered the electorate in 2022. To address this hypothesis, we estimated fixed effects regressions of the impact of 2022 abortion views on voting and, separately, the impact of change in Supreme Court favorability on voting. To test whether new voters were especially likely to rely on abortion-related views in forming vote preferences, we created a dummy variable representing whether a respondent was a new voter in 2022, and then interacted it with the appropriate indicators. As shown in Table 4, Model 1, abortion opinions influenced voting to a much greater extent among new voters in 2022 than they did among persistent voters. The estimated coefficient of Wave × Abortion Opinion × New Voter (0.192) is more than ten times the size of the coefficient of Wave × Abortion Opinion (0.016). Although individuals who turned out to vote in 2022 but did not vote in 2020 were no more likely to be abortion advocates than opponents, abortion as an issue weighed more heavily on voting among those new voters. This finding suggests that new voters in 2022 were highly motivated by abortion, even if their presence did not skew the distribution of voter opinions on abortion.

Table 4.

Additional impact of abortion views and change in favorability toward the Supreme Court among new voters on shifts toward Republican candidates, 2020 to 2022

	Model 1	Model 2
Wave (2020 to 2022)	−0.045*	−0.043
	(0.018)	(0.028)
Wave × Abortion opinion × New voter (2022)	0.192** (0.054)
Wave × Abortion opinion (2022)	0.016*
	(0.008)
Supreme Court opinion		0.005
		(0.009)
Wave × Supreme Court opinion		0.012
		(0.009)
Supreme Court opinion × New voter (2022)		0.275** (0.073)
Wave × New voter (2022)	−0.021	0.905**
	(0.125)	(0.249)
Intercept	0.392**	0.355**
	(0.005)	(0.029)
Log likelihood	1,654.684	1,667.441
Number of respondents	2,061	2,081
Number of observations	4,122	4,155

Note: Fixed effects regression analysis of congressional voting from 2020 to 2022. The sample includes respondents who shared vote intent in both waves. **P < 0.01, *P < 0.05.

Similarly, the results in Table 4, Model 2 indicate that changes in favorability toward the Supreme Court weighed more heavily in the voting decisions of those who newly turned out in 2022 than among persistent voters. The estimated coefficient of Supreme Court opinion x New voters (.275), is much larger than that of Supreme Court opinion (.005), and the former is statistically significant whereas the latter is not. Taken together, the findings in Models 1 and 2 strongly suggest that abortion views and changes in attitudes toward the Supreme Court weighed more heavily in the views of new voters than in the views of persistent voters, both abortion advocates and opponents alike.

Conclusion

The Supreme Court’s Dobbs decision increased the salience of abortion as an election issue. As a result, citizens’ views on abortion policies influenced voting in the 2022 congressional elections more than in the past. Since a larger share of Americans favor legal than illegal abortion, this priming effect was a boon to Democratic congressional candidates. However, more surprisingly, inflation and other economic perceptions did not affect vote switching, despite the severity of inflation and the many prognosticators who thought it would be crucially important to midterm voting.

This study illustrates what makes elections so difficult to forecast and what makes post hoc interpretations of them prone to mistake. The problem is not that voting is fundamentally difficult to predict; in fact, most votes are known long in advance of the election. Because switching from voting for a candidate of one party to a candidate of the other party is quite rare, studies seldom have the statistical power required to identify such effects.

The fact that our study does, nonetheless, identify such influences in a panel design speaks to the robustness of abortion’s impact. Based on cross-sectional relationships between issue positions and voter preferences, observers would conclude that inflation did, indeed, affect voting in the 2022 midterm elections. But one cannot determine which issues influenced change in candidate choice based on cross-sectional associations. Scholars can only assess which issues help versus hurt a candidate’s prospects by examining longitudinal data. Such evidence must be based on relationships between vote-changing and either issue opinions that have changed or that have been primed. In this case, our evidence supports the argument that opinions on abortion became more relevant to the direction of shifting voter preferences in 2022 after the Dobbs decision. However, we find no differences in turnout among those of differing views on abortion. Instead, new voters in 2022 were more likely to weigh abortion heavily in their vote-shifting calculus, regardless of whether they were for or against legal abortion.

Abortion was not a salient issue in preceding elections, which is why we do not have a measure of people’s opinions in 2020. If such a measure were available, we could examine whether opinion change on this issue also influenced voting. However, opinions on abortion appear to have been highly stable between 2020 and 2022 (35). Interestingly, by 2023, some polls suggested that the public had grown more supportive of abortion rights (36); if so, both opinion change and priming of this issue could precipitate vote shifting in future elections.

Based on other studies, we can be certain that Americans’ perceptions of inflation became more negative from 2020 to 2022 (37). But the electoral relevance of this issue and of changes in economic perceptions more generally was muted by attributions of responsibility that either coincided with preexisting candidate preferences, or that attributed responsibility to both or neither party.

Commonly cited evidence about the specific issues that people report as most important to them when voting are clearly not a safe basis for inferring the electoral impact of issues. Moreover, any issue-candidate preference association reflects both an issue’s positive impact for some and its negative impact for others. Position issues such as abortion are likely to simultaneously benefit candidates among some voters while hurting their prospects among others. For valence issues such as inflation or the economy, priming and negative changes in people’s perceptions are likely to co-occur because issues become more salient when they deteriorate.

Although we cannot determine which individual congressional races were won or lost based on abortion preferences, we can conclude that, on the whole, abortion views yielded greater support for Democratic than Republican candidates. Inflation and other economic indicators, on the other hand, had no systematic influence on vote changing. Our results further illustrate the perils of assuming that economic issues will trump all others in elections. Evidence for this claim has not been very convincing for congressional races. Nonetheless, the primacy of the economy lives on as a “conventional wisdom” in American politics and political journalism.

Finally, and perhaps most optimistically, our data provide evidence that voters shifted their congressional voting preferences in expected directions based on their issue opinions and the heightened salience of the abortion issue. Despite the highly stable nature of voting behavior, and a predictably small percentage of vote changers, it was possible for us to directly observe statistically significant effects on vote-changing based on the increased salience of abortion views, and indirectly based on their impact in changing views of the Supreme Court.

Many studies assume that policy issues have little bearing on voting, but that the economy has a substantial impact, especially in congressional elections. Further, the received wisdom is that perceptions of the economy “have a grounding in reality and an impact on elections that policy attitudes do not” (38). Yet from 2020 to 2022, congressional voting preferences changed in fundamentally rational ways based on abortion views, thus suggesting evidence of democratic accountability with respect to this particular issue. In contrast, accountability for the adverse economic effects of severe inflation was diffused by differing attributions of responsibility and partisan rationalization. Although varied attributions of responsibility are valid arguments about the economy in some respects, they are inconsistent with democratic accountability as envisioned in theories of retrospective economic voting.

Materials and Methods

The panel survey used in this study was a probability sample selected using address-based random sampling with face-to-face follow-up recruitment for nonresponses. Respondents were compensated by the National Opinion Research Center at the University of Chicago for survey interviews. Details on sampling and representativeness can be found in SI Appendix, Table S1.

The central outcome of interest in this study was how respondents voted in the 2022 congressional elections. In both election years, respondents who planned to vote were shown the names and party identification of candidates on their district ballots, as they would appear on the actual ballots, and asked for whom they would vote in October shortly before the November election. These dichotomous outcome variables were coded 1 for those who voted for the Republican candidate, and 0 otherwise, for both years. Those who did not vote in 2022 were excluded from our sample. Since postelection measures of voter preferences often reflect knowledge of the election outcome, we used pre-election vote intentions from immediately before the election.

In 2022, as in all contemporary US elections, votes were determined primarily by preexisting preferences that had little to do with economic conditions in 2022. The best predictor of current voting is always past voting, so much so that only a very small percentage of voters are expected to change their preferences from one party’s candidate to another in successive elections. For this reason, a fixed effects two-way panel regression was ideal because it could identify the predictors of change in voter preferences, rather than mere correlates of preexisting voter preferences.

To tap public Perceptions of the Severity of Inflation, respondents were asked, “Over the past year, have the prices of consumer products generally increased, decreased, or stayed about the same?” Responses were provided on a five-point scale. To tap Perceptions of Unemployment, they were asked on an identical scale, “Over the past year, has unemployment increased, decreased, or stayed about the same?” Both of these items were asked in 2022, when unemployment was low and inflation was high. For Perceptions of the Severity of Inflation as well as Unemployment and Abortion views, the interaction with Wave represents the extent to which those who thought inflation was higher in 2022 were more likely to switch their preferences toward the Republican candidates.

In addition, our statistical model includes changes in perceptions of national economic conditions, as well as respondents’ assessments of their personal finances in both 2020 and 2022. To the extent that those declined over time, we expected greater switching toward Republican candidates.

Using a two-way fixed effects individual-level panel analysis that incorporates a dummy variable for wave, coefficients for time-varying variables are to be interpreted as effects of change in the independent variables on change in the direction of voting for a Republican candidate, the hypothesized direction of inflation’s impact. Interactions between wave and opinions measured in 2022 indicate the extent to which those holding those views in 2022 were more (less) likely to change toward Republican candidates. Finally, we also include a range of other issues as robustness tests for our central findings (SI Appendix, Tables S6-1, S6-2, and S7).

IRB approval was received from the University of Pennsylvania, Protocol # 826006. Informed consent is administered to panel participants when they initially join the NORC panel.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Survey data have been deposited in Harvard dataverse: https://doi.org/10.7910/DVN/VVVQA4 (32).