The concept of public opinion as a dynamic, self-correcting mechanism has been a cornerstone of democratic theory since seminal works by [41] and [38]. The “thermostatic model” of public opinion argues that citizens respond to policy changes by moving their preferences in the opposite direction, effectively maintaining a form of political equilibrium. These theories suggest that when political actors move policy too far from the median voter’s preferences, the public responds by pushing back, creating a self-regulating democratic process ([13]). However, as we argue in this paper, this mechanism does not always work as smoothly as originally conceived in highly visible and politically divisive issues.

Scholars have long debated the malleability of attitudes toward immigration, a policy domain deeply intertwined with cultural identities, economic anxieties, and national security concerns ([36]; [16]). These characteristics suggest that opinions may be harder to change, raising questions about whether the “thermostatic model” of public opinion even functions when views are strong and tied to political identities. Emerging research in Europe finds evidence of a “reverse backlash” suggesting that exposure to extreme anti-immigrant rhetoric or policies might trigger a moderating response among voters ([10]). However, it is still unclear whether this reaction happens across most people or only among certain groups.

The Immigration attitudes are contentious by nature but might become more entrenched under extreme political polarization experienced in the last decade in advanced democracies. Does the thermostatic model work under such conditions?

While partisan polarization in the United States surrounding immigration has been most visible among political leaders, as with other issues elite partisan polarization has also penetrated down to the level of general public opinion and the electorate. Republicans and Democrats at all levels, have increasingly taken consistent positions on policy issues along conservative (Republican) and liberal (Democratic) lines. The sharp increases in partisan divergence¹ ([21]) have occurred not only on immigration, but fully across opinions on many issues ([1]; [3]; [22]; [35]). The differences between the two main parties have raised the stakes, making this competition increasingly fierce and possibly disabling some of its thermostatic mechanisms of public opinion to revert to more centrist views.

Fragmented and personalized media environments facilitate selective exposure and within camp message consistency, reinforcing prior attitudes ([30]; [39]). Social media platforms and personalized news feeds employ sophisticated algorithms that curate content based on users’ prior engagement patterns, effectively creating personalized information environments that expose individuals primarily to attitude-consistent information ([30]) that may polarize attitudes further.

Affective polarization intensifies in party alignment and out party animus, promoting cue consistent responses and polarization when counter cues are salient ([18]; [12]). In the specific context of immigration attitudes, this suggests that partisan attachments may override potential moderating influences of a reverse backlash while at the same time, increase polarization among attentive or predisposed subgroups ([5]; [16]; [15]; [37]). Under intense polarization, traditional moderating mechanisms may be weakened; predisposed groups can become more extreme in response to partisan cues ([17]; [12]).

Large opinion shifts are more likely when (1) elites send consensual signals, (2) media framing is sustained and coherent, and (3) objective conditions align with messages ([29]). Therefore, in today’s environment, coherence often exists within partisan silos ([31]), reinforcing attitudes rather than producing cross camp change.

Conditions for attitude change at the individual level and then in the aggregate, as described by [29], are rare nowadays compared to the 1980s and 1990s, because political elites presenting conflicting interpretations of events or policies (as opposed to congruent messages) prevail nowadays, which would make public opinion stable or fragmented rather than shifting in a coherent direction in order to correct for extremism at the elite level. Also, overall conflicting media narratives are the norm now, which rarely produce coherent aggregate opinion change. However, algorithm curation and the digital revolution have increased media coherence for individuals within their ideological camps. As a result, people may receive consistent messages, but only within their own political side which might generate greater polarization, rather than reverse backlash on immigration issues.

Prior’s ([31]) research on media choice and selective exposure documented how the expansion of media options has fundamentally transformed the information environment, allowing citizens to construct entirely separate political realities based on their consumption patterns. This structural transformation of the information ecosystem means that the conditions [29] identified as necessary for aggregate opinion change—unified elite messaging and consistent media framing—may still exist within partisan silos, but they operate to reinforce rather than shift attitudes across the broader public.

The third condition explained by [29] regarding change (when personal experiences and objective conditions aligning with some elite and media messaging) may be the underlying mechanism behind some of the findings in the reverse backlash scholarship and may help moderate extreme anti-immigrant attitudes. American citizens could be more responsive given the tangible and observable realities in their economic and social environments after Trump’s election.

The aftermath of the 2024 presidential campaign has been far worse than the first term in terms of implementing harsh anti-immigrant measures. These measures have broader effects on personal experiences of a broader section of the electorate. [11] analysis documented massive economic disruptions across major cities, with police departments incurring over $6.4 million in overtime costs. There’s been a disturbing pattern of violence, including six deaths in ICE detention centers and two fatal public shootings in Minneapolis, with detention rates rising nearly 75% from 40,000 to 66,000 individuals ([2]; [33]). The administration’s approach dramatically increased arrests of individuals with no criminal records, using detention as a pressure mechanism to force immigrants to abandon legal cases. Perhaps most impactful, at a larger scale, is the crisis of the Transportation Security Administration workers in late March 2026 unpaid for over six weeks, who began calling out of work at rates exceeding 40% at some airports, leading to what TSA administrators described as “the longest security lines in history” ([9]). Some airports experienced near-paralysis as understaffed security checkpoints forced passengers to arrive at terminals as much as seven hours before flights. This crisis was elicited by the congressional Democrats’ decision to defund the Department of Homeland Security in response to ICE enforcement abuses.

[7] provided early evidence of how policies can shape immigration attitudes in the context of the first Trump administration’s Muslim ban. Their research suggested that exposure to extreme anti-immigrant policies might trigger a liberalizing response among some voter segments.

If thermostatic and reverse-backlash dynamics apply to immigration, we should observe moderation when policy or elite signals become more restrictive ([10]; [6]; [7]). However, it is not clear whether under current polarization conditions in the United States, democratic publics are acting as a moderating force, follow elite cues (i.e. become more polarized) or remain stable.

Recent U.S. research offers two alternatives to thermostatic/reverse backlash expectations. One emphasizes elite cue-driven change ([43]; [23]; [37]; [4]). The other highlights stability rooted in long term predispositions—such as ethnocentrism, nationalism, and cosmopolitan identity—with limited evidence of abrupt change prior to 2016 ([8]; [19]; [20]; [32]). Both challenge automatic re-centering after major anti-immigrant campaigns and policy shifts.

Elite cues from party leaders and aligned media can shift immigration views, especially among co-partisans; effects polarize when counter cues are salient and are conditioned by predispositions ([12]; [16]; [15]; [40]). This perspective anticipates polarization among strong partisans rather than broad moderation ([43]; [23]).

In contrast, aggregate evidence often shows gradual liberalization and stable structure: opposition declined while predictors remained consistent from 1996–2018 ([34]). Other work finds relatively consistent attitudinal structure across partisan lines ([26]; [15]). Micro level instability may not sum to large aggregate change ([42]). Cue effects may decay without reinforcement ([20]), and reverse backlash remains a plausible mechanism to test ([10]). Taken together, studies of immigration opinion document both responsiveness to salient frames and substantial attitudinal stability, with changes concentrated among attentive or predisposed subgroups ([5]; [16]; [15]; [37]).

These patterns motivate our panel design to distinguish transient cue uptake from durable shifts. We use ANES panel data to examine within-person dynamics: panel designs allow us to distinguish transient cue uptake from more durable shifts ([8]; [19]; [20]). These literatures jointly inform our modeling choices and the tests of the reverse backlash theories we conduct with individual ANES panel data.

If there is indeed a reverse backlash mechanism at the individual level guiding attitudes towards immigration, we would expect to see some evidence that once Trump was reelected, one of the candidates with the most aggressive anti-immigrant rhetoric in recent times some voters with extreme anti-immigrant views, moderated them. Whether this moderation occurs broadly or only among certain individuals remains an open empirical question we explore in this paper. Because we do not observe direct exposure to specific policies, media, or enforcement actions, we refrain from attributing the observed within-person changes to any single cause. We instead treat elite cueing and issue salience to the policy environment as plausible pathways for future testing of micro mechanisms in an experimental setting.

Public opinion scholarship on anti-immigrant attitudes raises crucial questions about the nature of democratic responsiveness or opinion change in an era of increasing political polarization. Are classical models of public opinion change somewhat useful to understand opinion change regarding immigration in contemporary polarized political contexts? How do partisan identity, the electoral success of anti-immigrant movements, and exposure to extreme political rhetoric interact to shape attitudes toward immigration? Specifically, does exposure to aggressive anti-immigrant policies produce moderation? Or, do opinions remain largely stable despite changes in the political environment?

This paper challenges emerging research with a more optimistic view of democratic self-correction in the context of immigration attitudes in contexts of extreme partisan polarization, like in the United States.

This paper argues and finds that under the extreme partisan polarization dynamics in the US, individuals are less likely to revise their opinions and more likely to remain entrenched in them. However, when personal experiences with harsh anti-immigrant policies become widespread, moderation should occur, at least for the more extreme or anti-immigrant sectors. Some voters moderate their extreme anti-immigrant opinions once they are confronted with aggressive anti-immigrant policies or programs in their immediate environment.

The empirical analysis relies on the American National Election Studies (ANES) Merged Panel File (2016-2024). This dataset follows the same respondents over time and combines information from the 2016, 2020, and 2024 ANES Time Series studies, along with panel maintenance surveys conducted between these waves. The panel design allows for the analysis of individual-level attitudes across multiple electoral cycles, improving leverage over cross-sectional data.

The merged file includes nine waves: pre- and post-election surveys for 2016, 2020, and 2024, as well as three intermediate panel maintenance surveys (2019, 2022, and 2024). Respondents were initially recruited in 2016 through face-to-face and web modes, and subsequently followed through internet and mail surveys. The dataset contains 2,839 respondents in 2016 (pre- and post-election), 2,839 in the 2020 pre-election and 2,670 in the post-election wave, and 2,171 and 2,070 respondents in the 2024 pre- and post-election waves, respectively.

The 2020 ANES panel wave is excluded from the main analysis because the theoretical focus of this study is on potential backlash dynamics following the electoral success of anti-immigrant political forces, which are commonly associated with far-right populist parties and candidates ([28]). Prior research has conceptualized backlash as a reaction to the implementation or reinforcement of restrictive immigration agendas by such actors ([10]). Within this framework, the election of Donald Trump in 2016 and his subsequent political trajectory provide a relevant context to evaluate whether exposure to anti-immigrant rhetoric and policies generates attitudinal moderation over time. By contrast, the victory of Joe Biden in 2020 does not fit within this theoretical and conceptual framework, as it does not represent the advancement of anti-immigrant forces nor contribute to the mechanisms under study. For this reason, the 2020 wave is not included in the main models, although the broader 2016-2024 period remains central to the study’s empirical scope.

A key advantage of this dataset is that it merges multiple ANES studies into a single harmonized file using a unique respondent identifier, enabling consistent tracking of individuals across time. These observations provide a clear comparison across a critical period in U.S. politics, allowing the study to capture changes in attitudes over time while maintaining comparability in key measures.

To measure attitudes toward unauthorized immigrants, this study relies on both affective and policy-based indicators drawn from the ANES. These measures capture different dimensions of public opinion, including emotional evaluations and support for restrictive or permissive immigration policies. The dependent variable captures affective evaluations toward unauthorized immigrants using the ANES post-election feeling thermometer. The question asks: “How would you rate: illegal immigrants,” with responses ranging from zero (very warm or favorable) to one hundred (very cold or unfavorable). The original ANES variable is V162313. Non-substantive responses were recoded as missing, and the scale was reverse-coded so that higher values consistently indicate more negative, anti-immigrant attitudes.

The key independent variable (main predictor) measures pre-election policy preferences toward unauthorized immigrants. This variable is based on the American National Election Studies (ANES) question: “Which comes closest to your view about what government policy should be toward unauthorized immigrants now living in the United States?” Respondents chose from four options: make all unauthorized immigrants felons and send them back to their home country (coded as 4); have a guest worker program that allows unauthorized immigrants to remain in the United States temporarily for work (coded as 3); allow unauthorized immigrants to remain and eventually qualify for U.S. citizenship coded as 2; and conditional on meeting requirements such as paying back taxes, learning English, and passing background checks, allow unauthorized immigrants to remain and qualify for U.S. citizenship without penalties (coded as 1). The original variable is V161192. After recoding missing values, the scale was inverted, as mentioned previously, so that higher values reflect more restrictive and anti-immigrant policy preferences, while lower values indicate more favorable and permissive positions.

The second independent variable measures attitudes toward birthright citizenship. The ANES question asks: “Do you favor or oppose changing the U.S. Constitution so that the children of unauthorized immigrants do not automatically get citizenship if they are born in this country?” The original variable is V161193. This variable was recoded into a binary indicator where respondents who favor ending birthright citizenship (as well as those who express a neutral position) are coded as one, and those who oppose the proposal are coded as zero. This coding ensures that higher values represent less favorable attitudes toward immigrants.

In addition to these core variables, the analysis incorporates standard demographic controls, including gender, age, income, and education. Party identification is also included and used to disaggregate the analysis, allowing for the examination of differences across partisan groups. Across all variables, coding decisions were implemented so that higher values uniformly indicate less favorable or more anti-immigrant attitudes. Each question was used in the version corresponding to its respective survey year.

To examine the relationship between pre-election policy preferences and post-election attitudes toward unauthorized immigrants, this study employs a series of ordinary least squares (OLS) regression models. The empirical strategy is designed to capture both cross-sectional differences within each electoral cycle and changes over time, allowing for a more nuanced understanding of how attitudes evolve.

Models 1 and 2 estimate the relationship between pre-election policy preferences and post-election affective evaluations in 2016. In Model 1, the dependent variable is the post-election feeling thermometer toward undocumented immigrants, and the key independent variable is pre-election policy preferences toward unauthorized immigrants. Model 2 follows the same structure but replaces the main independent variable with attitudes toward birthright citizenship. These models provide a baseline assessment of how policy views before the election are associated with affective attitudes after the election within the same year. Models 3 and 4 replicate this framework for 2024, enabling a direct comparison across electoral cycles. Model 3 examines the relationship between policy preferences and the post-election thermometer, while Model 4 focuses on birthright citizenship attitudes. This parallel structure across years introduces a comparative framing that allows the analysis to identify whether the strength or direction of these relationships changes over time, particularly across a politically salient period in U.S. politics. All models are disaggregated by party identification, estimating effects separately for Democrats, Republicans, and independents. This approach allows the analysis to uncover heterogeneous effects across partisan groups, which is especially relevant given the polarized nature of immigration attitudes.

Model 5 introduces an additional econometric layer by focusing on within-individual change over time. The dependent variable is the difference in the post-election feeling thermometer between 2024 and 2016, while the main independent variable is the 2016 post-election thermometer. This specification captures how prior attitudes predict changes in affective evaluations over time, effectively leveraging the panel structure of the data. By modeling differences, this approach reduces concerns related to time-invariant unobserved heterogeneity and provides a more rigorous assessment of attitudinal change.

Across all models, standard demographic controls are included, such as gender, age, income, and education. However, it is important to note that the effective sample size is smaller than the original ANES panel due to missing data in key variables. In particular, income exhibits a relatively high rate of non-response in 2016, which restricts the number of observations when included as a control. For this reason, in the case of Republicans in Model 5, the analysis is estimated both with and without income as a control to assess the robustness of the results. More generally, sample sizes across models are also affected by respondents who do not report a party identification or fail to answer specific survey items, leading to variation in the number of observations used in each specification.

First, we estimated weighted linear regression models using the ANES 2016 post-election panel weight (V160102) and the corresponding ANES design variables (strata and Primary Sampling Unit). We restricted the analytic sample to respondents with positive panel weights.

Then, to estimate group-specific effects, or partisan identity, while preserving the survey design, we used the subpopulation approach, rather than subsetting the data prior to declaring the design. Models control for gender, education, age, and income alongside the focal policy variable.

Lastly, as robustness checks we first re-ran the main model using alternative single unit options to assess sensitivity to strata with a single Primary Sampling Unit (PSU). We then estimated a psu-only specification when strata information was problematic, and ran unweighted ordinary least squares and OLS with heteroskedasticity-robust standard errors. Coefficient estimates were substantively similar across specifications; however, variance estimates and statistical significance varied when the number of strata or PSUs was small. Accordingly, inference is based primarily on the strata and PSU specification with single units centered.

Here we present aggregate trends by party, to see whether partisans in the public have become more polarized about undocumented immigrants. If they have, we would expect Republicans and Independents’ attitudes to become more negative towards undocumented immigrants while Democrats becoming less so.

When looking at aggregate data in Figure 1, the feeling thermometer, a scale goes from 0 to 100, where values above 50 mean less favorable attitudes toward undocumented immigrants shows a high level of stability between 2016 and 2024. Contrary to popular belief, in terms of what increasing partisan polarization might look like for these opinions, there are small increases in negativity for all partisan groups. Among Democrats, the average negativity towards undocumented immigrants went from 47 in 2016 to 49 points in 2024; among Republicans, from 72 to 74; and among Independents, from 59 to 61.

The main pattern when looking at aggregate anti-immigrant attitudes over time seems to be persistence rather than change. Republicans stay clearly above 50 in both years, which shows consistently negative views. Democrats remain close to the neutral point, around 50, which suggests more mixed attitudes. Independents are in the middle and show very little change over time. This means that partisan differences are large, especially between Republicans and Democrats, Republicans showing a dislike for undocumented immigrants above 70 percent, and Democrats a few points below 50 percent denoting some favorability. Both groups remaining very stable over time and post Trump’s second term.

Figure 1. Trends in post-election anti-immigrant sentiment toward undocumented immigrants by Party ID (2016-2024). Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016-2024.

Overall, these results provide limited support for a reverse backlash. Instead of seeing a strong reaction that changes attitudes, the evidence suggests that opinions were already formed and stayed mostly the same over time.

In this section we use panel data from the 2016 and 2024 elections in the US to gauge whether pre-election anti-immigrant attitudes become more moderate in the aftermath of a candidate exhibiting some of the most extreme anti-immigrant rhetoric in recent history.

As shown in Table 1 below, there is little evidence that pre-election negative sentiment was moderated in the aftermath of Trump’s election in 2016. The feeling thermometer is coded so that higher values reflect more anti-immigrant views, positive coefficients indicate that less favorable and warm attitudes towards undocumented immigrants before the election are positively correlated with less favorable attitudes after the election as well. The post-election feeling thermometer is regressed on the pre-election one to gauge how much pre-election attitudes correlate with post-election ones regarding undocumented immigrants. The expectation is that if there is indeed a reverse backlash, we would see the coefficient of the preelection thermometer be close to zero or negative.

Table 1. OLS regression of post-election anti-immigrant sentiment toward undocumented immigrants and pre-election policy preferences toward undocumented immigrants by Party ID (2016).

Standard error in parentheses; *p < 0.05, **p < 0.01, ***p < 0.001. Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016–2024.

Somewhat surprisingly, the coefficient for pre-election views on undocumented immigrants for Democrats is 8.25, although it does not reach conventional levels of statistical significance. A one-unit increase toward a more restrictive position before the election is associated with an increase of about eight points in the post-election feeling thermometer, that is, a more negative evaluation of undocumented immigrants. In substantive terms, this suggests that a Democratic identifying individual who already held a negative position towards undocumented immigrants before the election tends to maintain—and potentially reinforce—that same orientation after the election, rather than moderating it, although this pattern should be interpreted with caution given the uncertainty around the estimate.

A similar pattern appears among Independents, where the coefficient of policy preferences is 12.55 and statistically significant, also showing an increase in negativity towards undocumented immigrants rather than moderation. Here, the magnitude is even larger than among Democrats, suggesting a strong positive association between pre- and post-election attitudes within this group. Among Republicans, the coefficient also shows a positive and statistically significant increase in negativity post-election, 8.23, indicating a consistent alignment between pre-election preferences and post-election sentiment.

Compared to the original specification, the differences across partisan groups are now less pronounced in directional terms, as all three groups display positive associations, though they vary in magnitude and statistical precision. This is possibly due to the constant or baseline opinion (holding constant all independent variables at 0) the average post-election thermometer would be around 66 for Republicans, 49 for Democrats, and 23 for Independents, suggesting meaningful differences in baseline levels of post-election sentiment across groups. Especially compared to the other two groups.

As shown in Table 2 below, there is limited evidence of reverse backlash, particularly among Republicans. Pre-election negative sentiment towards birthright citizenship for undocumented immigrants’ children is not associated with moderation in the feeling thermometer towards undocumented immigrants in the aftermath of Trump’s election in 2016. As in the previous model, the post-election feeling thermometer is regressed on pre-election attitudes towards undocumented immigrants to gauge how much pre-election attitudes correlate with post-election ones.

Table 2. OLS regression of post-election anti-immigrant sentiment toward undocumented immigrants and pre-election attitudes toward birthright citizenship by Party ID (2016).

Standard error in parentheses; *p < 0.05, **p < 0.01, ***p < 0.001. Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016–2024.

The constant now shows substantially higher baseline negativity across all partisan groups, with Democrats at 61, Republicans at 88, and Independents at 49, suggesting that post-election sentiment remains relatively unfavorable toward undocumented immigrants even when holding other variables constant. The coefficient for pre-election views on birthright citizenship for undocumented immigrants’ children for Democrats is 19.50 and statistically significant. A one-unit increase toward a more restrictive position before the election is associated with an increase of approximately nineteen points in the post-election feeling thermometer, that is, a more negative evaluation of undocumented immigrants. In substantive terms, this suggests that a Democratic identifying individual who held a negative position towards undocumented immigrants before the election is associated with maintaining and potentially intensifying that position after the election, rather than moderating it.

A similar pattern appears among Independents, where the coefficient remains positive (7.56) but does not reach conventional levels of statistical significance, indicating a weaker and more uncertain association between pre-election preferences and post-election sentiment. For Republicans, the coefficient is also positive (3.28) and statistically insignificant, suggesting little clear evidence of either moderation or reinforcement in this group. However, this is possibly due to the constant or baseline opinion (holding constant all independent variables at 0) the average post-election thermometer would be around 88, indicating a very high level of baseline negativity, which may limit observable variation in post-election attitudes.

Overall, compared to the earlier specification, the evidence points less clearly toward reverse backlash and instead suggests persistence—and in some cases reinforcement—of pre-election attitudes, particularly among Democrats.

Turning to Table 3, coefficients for pre-election policy preferences regarding deportation for all undocumented aliens remain positive and statistically significant across all three groups, on the dependent variable (post-election feeling thermometer towards undocumented immigrants). Among Democrats, the coefficient for pre-election views on removal of all undocumented immigrants in 2024 when regressed on the feeling thermometer is 7.67 and statistically significant, slightly larger than the 6.79 reported in 2016, suggesting no clear evidence of moderation relative to prior estimates. In substantive terms, this means that a Democrat who adopted a more restrictive position pre-election (i.e. comparing those who say immigrants should be granted a path towards citizenship with those who say they should be fined and removed) is more likely to express negative feelings towards undocumented immigrants post-election. In other words, even in 2024, individuals who already held more negative views tend to maintain them, with a magnitude that remains substantively meaningful rather than attenuated.

Among Republicans, the coefficient of pre-election views on undocumented immigrants is positive and statistically significant, 8.87, which is higher than previously reported (7.10), indicating a somewhat stronger association than in the earlier specification. The coefficient for this variable is also positive and significant for Independents, where a one-unit increase in pre-election anti-immigrant attitudes is associated with an increase of 13.58 points (compared to 13.03 in 2016) in the feeling thermometer (where 100 means the most negative feelings towards undocumented immigrants).

Table 3. OLS regression of post-election anti-immigrant sentiment toward undocumented immigrants and pre-election policy preferences toward undocumented immigrants by Party ID (2024).

Standard error in parentheses; *p < 0.05, **p < 0.01, ***p < 0.001. Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016-2024.

In contrast to the 2016 results presented in Table 1, the coefficients do not appear uniformly smaller; instead, they remain comparable or slightly larger in magnitude across groups, suggesting persistence rather than attenuation in the relationship between pre- and post-election attitudes.

Lastly, as shown in Table 4, there is, in contrast to 2016, limited and less consistent evidence of reverse backlash among Republicans. Pre-election negative sentiment towards birthright citizenship for undocumented immigrants’ children does not appear to be associated with clear moderation in the feeling thermometer towards undocumented immigrants in the aftermath of Trump’s election in 2024. The coefficient for pre-election views on birthright citizenship for undocumented immigrants’ children for Democrats is negative (−1.60) and statistically insignificant, suggesting no clear association between pre-election attitudes and post-election sentiment in this group.

Among Republicans, the coefficient is positive but small and statistically insignificant (3.33), indicating no association between moderation or reinforcement in post-election attitudes. Only among Independents do pre-migration negative attitudes on birthright citizenship remain substantively and statistically significant, with a coefficient of 21, indicating a strong positive association with post-election negativity toward undocumented immigrants. This represents the largest effect across the three groups in this model, suggesting that pre-election attitudes are most strongly carried over among Independents in 2024.

Table 4. OLS regression of post-election anti-immigrant sentiment toward undocumented immigrants and pre-election attitudes toward birthright citizenship by Party ID (2024).

Standard error in parentheses; *p < 0.05, **p < 0.01, ***p < 0.001. Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016-2024.

Compared to the 2016 results, the pattern shifts notably: rather than observing signs of moderation among Republicans, the results point to weak or null associations for both Democrats and Republicans, with stronger persistence concentrated among Independents.

Overall, models in 2016 and 2024 show that pre-existing attitudes toward undocumented immigrants are generally positively correlated with these attitudes in the post-election period, although the strength and consistency of this relationship varies across specifications and groups. There is evidence of stability over time, with limited indications of moderation confined to specific subgroups and measures rather than a systematic pattern. Importantly, the absence of statistically significant negative coefficients continues to provide little support for a reverse backlash in attitudes toward undocumented immigrants.

In this context, even amidst the strong anti-immigrant rhetoric during Donald Trump’s 2016 and 2024 campaign, the results tend to suggest that attitudes either remain stable or become more negative, especially for Independents, in the post-election period, while Democrats and Republicans show more inconsistent patterns in this specification.

We estimate first difference models (2024-2016) with the 2016 thermometer as the main predictor; negative coefficients imply greater moderation among those most negative in 2016. If reverse backlash dynamics operate over time, we should observe such negative associations. This approach provides a more robust test of reverse backlash, because it allows us to observe whether individuals adjust their attitudes after experiencing anti-immigrant policies in the first Trump administration.

As reported in Table 5, the dependent variable captures the change in the anti-immigrant feeling thermometer between 2024 and 2016 (coded so that higher values reflect more negative attitudes toward undocumented immigrants). We included the 2016 feeling thermometer as the main predictor. Negative coefficients in this predictor indicate that individuals who held more negative views in 2016 tend to moderate by 2024. Our expectation is that there would be some evidence of reverse backlash as individuals experienced anti-immigrant policies, rather than only the election of a candidate running on an anti-immigrant platform. To assess whether the results could be driven by regression to the mean, we conducted an additional check in which 2024 attitudes are modeled while accounting for respondents’ baseline levels in 2016. This exercise reveals a strong and statistically significant association between attitudes in 2016 and 2024, indicating substantial persistence over time. These results suggest that, although individual attitudes remain relatively stable, the patterns observed in the main analysis are unlikely to be explained solely by mechanical reversion from extreme initial values.

Consistent with this expectation, the results from the panel analysis (2016–2024) provide evidence of reverse backlash in anti-immigrant attitudes, although the magnitude varies across groups. Results show negative, statistically significant slopes across groups (Table 5), indicating that individuals starting more negative in 2016 tend to moderate by 2024. Among Republicans, the slope is largest in the model without income (−0.69): each one point increase in 2016 negativity predicts a 0.69 point decrease by 2024. Democrats (−0.45) and Independents (−0.62) show similar moderation; the full sample is −0.51. Table 5 illustrates stronger moderation among Republicans: those starting highest in 2016 are expected to move downward more by 2024, consistent with a pull toward the center. Although these shifts are temporally aligned with high-salience policy periods, future research should test whether this is causally related to greater exposure of extreme anti-immigrant policies or candidates.

These associations are consistent with reverse backlash dynamics, though magnitudes are modest. Using the Republican (no income) model, the predicted change is for a Republican with a 2016 score of 90, −9.85, yields a predicted 2024 score ≈ 80 (holding other covariates at reference values). The design does not identify mechanisms; future work should incorporate direct exposure measures to test variation among those who were personally exposed to immigration policies or experienced them through the media, although we would argue most individuals (unless the were in remote locations without access to any traditional or electronic media) were exposed to anti-immigrant campaign rhetoric and policies.

Table 5. OLS regression of changes in the feeling thermometer toward undocumented immigrants (2024-2016).

Standard error in parentheses; *p < 0.05, **p < 0.01, ***p < 0.001. Source: Authors’ calculations based on panel data from the American National Election Studies (ANES), 2016-2024.