The Stability of Immigration Attitudes: Evidence and Implications

Authors: Alexander Kustov, Dillon Laaker, and Cassidy Reller

Journal: Journal of Politics (2021)

Resources

Figures & Tables

Abstract

Do voters have stable immigration views? While any account of immigration politics must make an assumption about whether underlying attitudes are stable, the literature has been ambiguous regarding the issue. To remedy this omission, we provide the first comprehensive assessment of the stability and change of immigration attitudes. Theoretically, we develop a framework to explicate the temporal assumptions in previous research and find that most studies assume attitudes are flexible. Empirically, we draw on seven panel datasets to test the stability question and use multiple approaches to account for measurement error. We find that immigration attitudes are remarkably stable over time and robust to major economic and political shocks. Overall, these findings provide more support for theories emphasizing socialization and stable predispositions rather than information or environmental factors. Consequently, scholars should exercise caution in using changing context to explain immigration attitudes or in using immigration attitudes to explain political change.

Introduction

While a lot is known about what explains the contemporary variation in individual immigration attitudes and preferences, (Note: While some scholars draw a sharp distinction between (self-reported) attitudes and (revealed) preferences, we follow @Druckman2000 and define “preferences” as any rankings derived from comparative evaluations of (or “attitudes” toward) various policies or other objects.) the literature has so far been rather ambiguous regarding the reality of their stability or change. It is common to assume that these attitudes are malleable and easily susceptible to information and contextual factors. Some scholars, however, conversely assume that these attitudes are rather fixed. While it may be possible to find empirical support for either of these perspectives under temporal constraints, the question is whether voters actually change their mind on immigration over the long term. (Note: Despite the complexity of immigration policy [@Ruhs2013] and related attitudes, we follow the existing public opinion literature [@Hainmueller2014] and consider voters’ underlying comparative evaluation of immigration as a whole.)

The previous literature provides a variety of distinct perspectives on the stability and change of immigration attitudes. As an illustration of this variability, we reviewed an exhaustive list of 76 articles published between 1993 and 2018 (see Figure fig:lit). Of these papers, 75 percent explicitly or implicitly assume that immigration attitudes are flexible, while 25 percent assume attitudes are fairly stable. In short, this analysis suggests most research on immigration attitudes assumes that it is possible to change a voter’s mind on the issue in a robust way. At the same time, the few papers that assume otherwise usually treat immigration preferences as a black-box variable.

Figure. Assumptions of stability and change in immigration attitudes literature

The debate on the stability and flexibility of immigration attitudes is closely linked to literature in political socialization concerning whether early experiences persist throughout one’s life or whether individuals can consistently update their beliefs in response to contextual factors [@Alwin1994, Sears1983b]. Moreover, the stability of immigration attitudes has critical implications for several debates within the immigration literature. For example, if immigration attitudes are flexible, then theories emphasizing the importance of predispositions have little validity. Alternatively, if immigration attitudes are stable, then arguments centered on information primes and contextual factors have limited theoretical traction. Despite the fact that any theory concerning immigration attitudes must make an assumption about their stability, empirical assessments of the issue are surprisingly lacking in the literature.

To remedy this omission, we provide the first comprehensive theoretical and empirical evaluation of attitudinal stability and change on immigration. Using seven panel surveys that cover the 2008 recession, Brexit, and the refugee crisis, we find that individual views toward immigration are remarkably stable across time. Our results do not necessarily discredit theories that assume flexibility. However, they suggest the variation these theories attempt to explain is quite minimal. Specifically, exogenous shocks may change immigration attitudes, but this change is small and quickly reverts back to an individual’s long-term equilibrium. Thus, our results cast doubt on the explanatory power of theories emphasizing the role of contextual factors and challenge much of the previous literature on immigration attitudes. Finally, consistent with the political socialization literature, we find substantial variation in the stability of immigration attitudes across age. Younger individuals are more likely to change their views toward immigration compared to older adults.

This article makes several notable contributions. First, our study adds important insights to the burgeoning literature examining public attitudes toward immigration. Theoretically, we explicate the often implicit assumptions concerning the temporal stability of immigration attitudes in the literature. Our analysis highlights the importance of specifying and testing the long-term temporal implications of immigration theories. Empirically, our results suggest that scholars would benefit from examining how and when immigration attitudes develop. Broadly, our findings generalize recent studies that demonstrate information cues [@Hopkins2018], refugee crises [@Mader2018], and economic shocks [@Goldstein2014a] do not substantially influence immigration attitudes.

Second, this article contributes to the comparative politics literature examining the rise of populism. Our results indicate that changing immigration attitudes cannot explain the recent rise of anti-immigration parties [@Bonikowski2017]. Rather than changing attitudes toward immigration, exogenous shocks such as a refugee crisis, may simply increase the salience and issue importance of immigration. In other words, when immigration is a low-salience issue, voters with anti-immigration attitudes may still vote for parties with pro-immigration positions. When the salience of immigration increases, however, it becomes a key voting issue [@Mader2018]. Consequently, future research on the topic can benefit from focusing on attitude and issue salience [@Dennison2018].

Finally, this article adds to the broader scholarship that examines the stability and development of political attitudes [@Bartels1993, Gerber1998]. The previous literature often uses panel surveys that include at most five waves and only one survey item to test attitude stability. However, this approach requires unrealistic and untestable modeling assumptions. This article draws on panels that include more than five waves and multiple survey-items, which allow us to empirically test and relax many of these assumptions.

Persistence and development of political attitudes

Central to the question of immigration attitude stability is an extensive debate in the political socialization literature concerning the “persistence” of early life experiences versus the “lifelong openness” of individuals for belief updating [@Alwin1994, Sears1983b]. (Note: Several studies analyze the stability of political attitudes and party attachment [@Jennings1984, Sears1999, Green2002]. Others extend this research to ideology, group evaluations, political interest, and other issue positions [@Alwin1991, Alwin1994, Prior2010].) On one end of the spectrum, the persistence model emphasizes the importance of pre-adult experiences and the enduring nature of political attitudes. (Note: There is a rich literature on how college, school, and family affect the political socialization of individuals [@Jennings1984, Mendelberg2017].) On the other end, the lifelong openness model posits that individuals can change their beliefs throughout their lives in response to current events. Alternative models simply vary the probability of attitude change over the course of one’s life. For example, previous studies find strong support for the impressionable years model, which emphasizes the importance of late adolescence and early adulthood [@Alwin1991]. (Note: Bayesian learning models provide another approach that generates a similar hypothesis as the persistence model when the political system is relatively stable [@Bartels1993, Gerber1998].) This period is seen as consequential because young adults have limited political experience and are often just beginning to engage with political institutions.

Accordingly, if the lifelong openness model is correct, immigration attitudes should reflect current contextual factors, such as economic conditions or media coverage. Alternatively, the persistence model suggests that immigration attitudes should be highly stable over time. Finally, if the impressionable years model was correct, immigration attitudes should be unstable for young individuals and begin to crystallize as they age. (Note: Since stability is often a function of the specific attitude [@Sears1983b], previous research on other political issues provide little information on the stability of immigration attitudes. @Sears1983b suggests that attitudes will have high stability when they have consistent meaning over time and are politically salient.) The literature on public attitudes toward immigration has developed separately from this discussion. As a result, scholars have not seriously considered the theoretical implications and empirical reality concerning the stability or change of immigration attitudes. They have not theorized or identified when these attitudes are likely to develop [@Lancee2015]. This is perhaps disconcerting given that every account of immigration attitudes must make an assumption about their stability. (Note: Some theories are interested in the short-term change of immigration attitudes, which may still be important inasmuch it affects the immediate outcomes with long-term consequences [@Goodwin2017].)

Change and stability of immigration attitudes

Our analysis of the previous literature suggests that most studies assume that immigration attitudes are at least somewhat flexible. In this section, we explicate the often implicit assumptions in the previous literature. We do so by developing a framework that categorizes the temporal assumptions in previous studies into four groups along a continuum: “communication,” “environmental,” “intermediate,” and “predispositional” theories (see Figure fig:theories). Theories that emphasize framing, elite rhetoric, or media coverage, which we label “communication” theories, assume immigration attitudes are flexible. On the other end of the continuum, “predispositional” theories, which highlight personality, ethnocentrism, and cognitive biases, assume that immigration attitudes are highly stable. Near the left of center is “environmental” theories, which focus on economic conditions, demographics, and the political environment and assume immigration attitudes are somewhat flexible. Finally, “intermediate” theories, located to the right of the center, assume immigration attitudes are fairly stable and emphasize socialization, economic position, and group identity.

Figure. Assumptions of stability and change in theories of immigration attitudes

Expectations of relative volatility

We categorize studies that emphasize systematic communication and environmental factors as theories that assume immigration attitudes are relatively flexible. First, one obvious source of temporal variation in individual attitudes is related to changes in voters’ beliefs and information. We label this group of studies as “communication” theories, which assume that immigration attitudes can be easily changed depending on information and framing.

To that end, research shows that most people have little knowledge about politics and often misperceive the connection between public policies and their interests [@Druckman2014]. Likewise, the electorate seems to be ill-informed on the consequences of immigration. For instance, people tend to overestimate the number of immigrants, as well as their impact on national fiscal systems, wages, and employment [@Dustmann2005a]. Would people have different immigration attitudes, had they been better informed? According to the “enlightened preference” literature [@Gilens2001], the answer should be positive. After all, education is found to be one of the strongest predictors of immigration attitudes [@Hainmueller2014]. At the same time, this perspective also implies that, in line with the top-down elite approach [@Lenz2012], politicians should be able to easily manipulate voters preferences on immigration as on any other issue.

As the growing evidence demonstrates, however, informing voters may not work for immigration issues. Accordingly, @Hopkins2018 find that in seven distinct experiments correcting people’s misinformation on immigration numbers does not change their policy attitudes [@Johnston2016, Alesina2018]. @Barrera2018 provides evidence that fact-checking makes people update their knowledge, but does not change their attitudes or behavior. Of course, there is some evidence of successful change of individual attitudes due to information, (Note: @Grigorieff2016 and @Facchini2016 find that providing information may slightly improve opinion on immigration for those with already negative attitudes and these effects persist over a few weeks.) but few experimental studies have the capacity to test whether these effects are truly long-lived (over months and years).

A related way to change people’s opinion on immigration—where information cues may not be sufficient—is to reframe it by highlighting certain concerns that people already hold dear [@Chong2007]. There is some indication, for instance, that people are responsive to the appropriate moral arguments (e.g., “equal treatment”), general pragmatism, and the appeals to their national economic interest [@Gilliam2010]. Similar to information experiments, however, the longevity of these effects in an “informationally rich” competitive political environment is unclear. When it comes to media effects, for instance, @Flores2018a finds that elite cues can make people who are skeptical of immigration even more negative, but these effects are short-lived and thus require constant repetition.

A second group of theories that assume immigration attitudes are at least somewhat flexible focus on systematic environmental factors, such as economic conditions and demographics. For example, according to “group threat” theories people should negatively respond to the increase of an immigrant population. The overall evidence for this idea, however, is mostly observational and rather inconclusive [@Pottie-Sherman2017]. One the most ambitious experiments supportive of the “group threat” effects with a strong treatment (i.e., repeated exposure to ethnic outgroups on the train) shows that these effects are small and wane greatly after ten days [@Enos2014b]. @Hangartner2018 finds that exposure to refugees causes an increase in anti-immigration attitudes for natives. (Note: Though, the identification strategy using a non-longitudinal survey is perhaps problematic given that the Greek islands near Turkey have historically been a point of illegal immigration into the EU.)

The literature emphasizing the role of economic conditions also suggests that immigration attitudes can be rather flexible [@Kehrberg2007, Wilkes2008]. One widely held belief is that recessions tend to cause a spike in resentment toward foreigners. Broadly, as the economy declines (improves), anti-immigration attitudes should increase (decrease). Most of this research, however, focuses on cross-sectional and cross-time differences in aggregate public opinion. Using panel data on individual attitudes, @Goldstein2014a demonstrate the lackluster effect of the 2008 financial crisis on US immigration attitudes.

Expectations of relative stability

We define studies that emphasize socialization, economic position, group identity, personality, ethnocentrism, and cognitive biases as theories that assume immigration attitudes are relatively stable. We categorize these theories into two broad categories: “predispositonal” and “intermediate” theories. First, immigration attitudes should be durable if they are primarily driven by underlying psychological predispositions related to personality [@Gallego2014], ethnocentrism [@Kinder2010], altruism [@Kustov2019], or authoritarian and (anti-)egalitarian ideological motivations [@Cohrs2010]. Since these factors are likely very stable, immigration attitudes should rarely change.

The second group of intermediate theories cover various explanations, but tend to assume that immigration attitudes are at least somewhat stable. For example, we would expect stability if individual attitudes are driven by identity-protective motivated reasoning in a polarized political environment [@Lodge2013, Kahan2016a]. According to this perspective, informing voters about immigration may do little to change their policy preferences, especially if they perceive that their key political identities are at stake. Given that people rarely change their partisan allegiances, the prospect for attitudinal change regarding immigration without a significant political change are thus rather scant.

In this respect, we should also expect the underlying attitudes to be stable if the policy or the related political environment have not changed. While it is easy to think of prominent examples of crises (e.g., recent influx of refugees into Europe), international migrants have consistently accounted for only 3 percent of the world’s population over the last 100 years. At the same time, ever since the U.S. Chinese Exclusion Act and the introduction of the passport system, national governments have had capability to restrict any entry to their territory. We would also expect attitudes to be stable if the underlying social norms have been unchanged. (Note: The comparison to the historical stability and recent change in LGBT attitudes across advanced democracies is quite instructive here [@Tankard2016].) Accordingly, despite a few fluctuations and a rise of positive attitudes over the last several years, the Gallup poll shows little change in the aggregate US immigration attitudes since 1966. (Note: For details, see the latest reports from Gallup (2018).) In line with these perspectives, there is some evidence from panel data that immigration attitudes are robust to receiving education [@Lancee2015] and even an economic shock [@Goldstein2014a] or a refugee influx [@ Mader2018].

Immigration attitudes should also be somewhat stable if political economy theories are correct. Most prominently, if labor market competition influences immigration attitudes [@Scheve2001a], then attitude change should only occur when individuals (expect to) enter and leave the labor market, change their sector, or acquire skills. Furthermore, if public finance is an important factor [@Hanson2007], then change in immigration attitudes would be rare since average taxation and spending rates are quite stable.

In sum, every account of immigration attitudes must make an assumption about their stability. While some scholars are skeptical that individuals may even have stable policy preferences [@Converse1964], there are theoretical reasons to believe immigration attitudes—or at least the motivations behind them—are relatively stable. (Note: Some theories that emphasize the interactions between information, environment, and stable predispositions (e.g., “sociotropic politics”) are in principle consistent with either perspective on attitude stability depending on the relevant independent variables and their change (or its absence).) Understanding the stability of immigration attitudes provides important theoretical leverage to evaluate several key debates within the literature. Our research aims to resolve this disagreement by providing the most comprehensive empirical assessment of the stability question to date. (Note: Importantly, regardless of the stability of immigration attitudes, political outcomes may still be dependent on the change of related social norms [@Tankard2016] or the salience of immigration as a political issue [@Hatton2017]. Given the absence of high-quality longitudinal data, the examination of norm change is beyond the scope of this paper. At the same time, the evidence of immigration’s changing salience as a response to changing context is overwhelming. For instance, it has been shown that increases of local immigrant population may help politicize the issue [@Hopkins2010].)

Data Sources

To analyze the stability of immigration attitudes, we draw on seven high-quality, population-based panel surveys from the United States and Western Europe: the Netherlands’ Longitudinal Internet Studies for the Social Sciences (LISS) panel, British Election Study (BES) panel, Norwegian Citizen Panel (NCP), The American Panel Survey (TAPS), Ireland National Election Study (INES), Swiss Household Panel (SHP), and German Longitudinal Election Study (GLES). We select panels that conduct at least three waves and span at least two years. Table tab:paneldata provides a brief description of the panel surveys used and the specific questions. A detailed discussion of each panel survey can be found in the Appendix.

tables/table1

It is important to highlight some of the strengths of the specific panels used in this article. First, several panels cover major shocks, which many theories suggest should cause shifts in public opinion toward immigration. Specifically, the LISS panel, covers the financial crisis in Europe. The BES, NCP, GLES, and LISS panels cover the European refugee crisis. The BES also covers the 2016 referendum on EU membership in the UK, which caused substantial media coverage and public debate over immigration. If attitudes remain stable through a major economic contraction and inflow of migrants, they are unlikely to change.

Second, several panels include multiple survey items to elicit immigration attitudes and all panels include more than five waves. For example, the LISS panel includes nine waves and six questions that cover various elements of the issue, such as immigration levels and asylum applications. Previous studies that examine the stability of attitudes often only use single questions and rarely use panels that extend more than five waves, (Note: One exception is @Prior2010, who uses several panels with more than five waves to test the stability of political interest.) which requires unrealistic and untestable modeling assumptions. By increasing the number of survey items and time-periods, we are able to relax and test these assumptions.

One potential issue arises if the time periods covered in the panels have no variation in the key variables identified in “communication” and “environmental” theories. In other words, without variation in these variables, then we should also see stability in immigration attitudes. However, the time period covered by the panels includes substantial variation in these variables due to the global recession, Brexit, and the refugee crisis. For example, during the referendum on EU membership in the United Kingdom, coverage of immigration tripled and was overwhelming negative [@Moore2017]. (Note: Coverage increased faster than any other political issue and primarily blamed migrants for economic and social problems. It was the focus of 99 front pages compared to 82 front pages concerning economic issues) The refugee crisis caused similar patterns in the Netherlands, (Note: See @Jacobs2018 and @McAuliffe2017.) Norway, (Note: See IMEX (2018).) and Switzerland. (Note: See @McAuliffe2017) Moreover, Germany received over a million refugees after Angela Merkel opened its borders in 2015. In the Netherlands, there is also substantial variation in the unemployment rate, (Note: Based on data from the European Social Survey, there is also substantial variation in the satisfaction with the economy over time in Switzerland and the Netherlands.) ranging from a low of 2.75 percent in 2008 to a high of 7.41 percent in 2014. (Note: See FRED (2018).) Finally, the refugee crisis caused a substantial influx of migrants across Europe. Thus, these events likely provide a hard test for the stability of immigration attitudes. Even for the countries that did not experience the most severe economic turmoil or influx of migrants, these events still had substantial economic and political effects. Further, given the interconnectedness of the European economy and society, shocks in one country likely spill-over.

Empirical strategy and results

Our main empirical exercise is to determine whether individuals have the same immigration attitudes over an extended period of time. A critical issue to confront when evaluating absolute individual-level attitude stability is measurement error. When scholars develop theories, they often posit a relationship between abstract concepts. To test these theories, however, they must first specify concrete indicators to measure these concepts, which introduces measurement error. The concept of immigration attitudes is multi-layered and, therefore, selecting a specific question(s) to measure this concept is difficult. For instance, potential questions can capture general immigration attitudes or focus on a specific group, such as skilled immigrants, refugees, or migrants from certain countries. Furthermore, even for a particular question, there is still some flexibility with regard to question wording and the number of response categories. All of these factors will likely always introduce some amount of measurement error. Random variation can also be introduced by respondent inattentiveness or fatigue, the interview context, and simple typographical errors.

Measurement error is especially problematic when evaluating the stability question because it attenuates observed correlations across time toward zero [@Ansolabehere2008a]. Therefore, to assess stability in a compelling way, it is necessary to isolate true changes in immigration attitudes from this random variation. We account for this random response variation in our analysis in two major ways. First, we leverage the several panels that use multiple survey questions to elicit immigration attitudes. By simply averaging these survey items together, we are able to reduce the variance of the measurement error and, thus, are able to better estimate respondents’ underlying immigration attitudes [@Ansolabehere2008a]. Second, we estimate a measurement model (described below), which evaluates the relative stability of immigration attitudes. As @Prior2010 notes, while perfect relative stability may coincide with absolute instability when all individuals change by the same degree, we alleviate this concern by also demonstrating the stability of immigration attitudes at the aggregate-level (see Appendix).

Broadly, we follow a similar strategy as @Prior2010, who studies the stability of political interest. We evaluate the stability of immigration attitudes through three distinct approaches. First, we provide graphical evidence to establish a baseline for the stability of individual-level immigration attitudes. Second, we directly address potential measurement error using the approaches described above. Third, we explore potential variation in the stability of immigration attitudes. To preview our results, we find that immigration attitudes are remarkably stable, even across major political and economic shocks. We also find that young individuals are more likely to change immigration attitudes, which provides evidence for the impressionable years hypothesis.

In the Appendix, we analyze the stability of immigration attitudes through two additional approaches. First, in the spirit of @Achen1975,Feldman1989, we investigate the nature of response instability. The results are consistent with the random variation in the data being measurement error. Second, following @Prior2010, we estimate dynamic panel models to examine if and how fast immigration attitudes revert back to an individual’s long-term equilibrium after a shock. The evidence suggests that immigration attitudes quickly revert back to the long-term equilibrium.

Individual-Level Stability of Immigration Attitudes

One straightforward way to measure the stability of immigration attitudes at the individual level is to examine the share of respondents who give the same response at different times. The top panel in Figure fig:panelindividual reports the percentage of respondents who give the same answer in the first wave and in each subsequent wave. The bottom panel in Figure fig:panelindividual reports the percentage of respondents who change by less than two response categories. The solid line reports the percentage for respondents who completed the first wave and one additional subsequent wave. (Note: There is no substantial difference between respondents who completed all waves and respondents who completed the first and at least one subsequent wave, which suggests that panel effects do not exist in the data. In the Appendix we also test for panel effects in a systematic way.) Our results are similar to those in @Prior2010. For the top panel, the percentage providing the same response in each subsequent wave as in the initial wave ranges from 32 percent (INES) to 89 percent (TAPS). For the bottom panel, the percentage of respondents who do not change by more than one response category ranges from 71 percent to 94 percent.

Figure. Individual level stability of immigration attitudes

Three points are worth mentioning here. (Note: See @Prior2010 for a more in-depth discussion.) First, as the number of categories increase, the stability of immigration attitudes decreases. For example, in the top panel of Figure fig:panelindividual, TAPS, with only two response categories, has the highest percentage of respondents giving the same response in each subsequent wave while the INES, with seven categories, has the lowest percentage. As @Prior2010 notes, it is quite intuitive that the wide range of response categories allow respondents to report smaller changes in immigration attitudes. If elicited attitudes are influenced by the specific context of when the question survey is conducted, a higher number of response categories is able to capture that fluctuation. Though, if a respondent’s true immigration attitude is between two categories, they may alternate between the two categories across the waves. This would indicate a change in attitudes; however, this response variation was introduced by the number of response categories and is measurement error [@Prior2010]. Alternatively, with a wider range of response of categories, it may be harder for respondents to accurately specify their true attitudes. This especially becomes difficult across multiple years. A seven and eight on a ten point scale may represent the same immigration attitude during different years if the specific context changes the meaning of the values. Additionally, respondents who randomly answer survey questions have a higher probability of selecting the same category when there are fewer options.

Second, as the length of time since the first wave increases, stability in immigration attitudes decreases. However, this decrease is small. The probability of reporting the same answer in the LISS survey in 2008 and 2009 is 0.58. If this represented the true stability of immigration attitudes after one year, it should be expected that the probability of reporting the same attitude after n years is $0.58^n$ [@Prior2010]. Thus, the stability of immigration attitudes between 2008 and 2017 should be $0.58^9=0.007$. As Figure fig:panelindividual clearly illustrates, the empirical probability is much higher (0.50).

Third, the bottom panel of Figure fig:panelindividual shows a substantial increase in attitude stability. For the LISS panel, the percentage changing by less than two categories between 2008 and 2009 is 94 percent and this only decreases to 89 percent when comparing 2008 and 2017. This is a drastic improvement in stability when only 58 percent gave the same response in 2008 and 2009.

Immigration attitudes appear to remain stable throughout economic turmoil and the refugee crisis. While measurement error certainly may cause some of the variance in responses, it is also possible that the differences in the specific environment in which the survey question was asked is driving these differences. Nevertheless, the results point to stability around a central tendency. Though, a more rigorous analysis of the stability of immigration attitudes requires explicitly addressing measurement error.

Measurement error and the stability of immigration attitudes

An individual’s response to a survey question includes the true score (a latent variable), which is unobservable, and measurement error. To accurately estimate the stability of immigration attitudes it is necessary to isolate true attitude changes from this error. While there are numerous causes of this random response variation, under certain assumptions about the nature of the error, the effect of measurement error on stability estimates can be appropriately modeled. Our goal in this section is to distinguish the true change in immigration attitudes from variation that is introduced by measurement error. We account for random response variation in our analysis in two ways. First, we follow @Ansolabehere2008a and leverage panel datasets that use multiple survey items to measure immigration attitudes and simply calculate the correlation coefficients between the first and last waves. We are able to reduce the variance in the measurement error and better estimate respondents’ true attitudes by simply averaging these survey items together. Second, we estimate latent structural equation models to evaluate the relative stability of immigration attitudes.

Table. Spearman rank correlations between first and last waves for scales with multiple survey items and average individual items

First, following @Ansolabehere2008a we examine the stability of immigration attitudes by estimating simple correlations between the first and last waves. For each panel, we construct scales, which are simply the averages between the survey items for each wave. We then calculate the correlations between the first and last waves. Column 4 in Table tab:panelcor reports these correlations. In Column 5, we also report the average correlations between the first and last waves when using only a single survey item. Correlations in parentheses are estimates when only using respondents who completed all panel waves.

The correlations when using the scales are much larger compared to the correlations only using a single survey item. This result is consistent with measurement error being present in the data. Specifically, for the LISS panel, the estimated correlation between 2008 and 2017 when using the scale equals 0.720, while the average correlation when using each question separately is only 0.507. This represents roughly a 42 percent increase in the correlation estimate. Importantly, during this time period, the Netherlands and the EU experienced a significant financial recession and a refugee crisis. The differences between the correlation estimates are smaller for the NCP and BES; however, they are still meaningful. The NCP and BES also cover the refugee crisis and, again, the results still provide evidence of relatively high stability. Further, the BES panel spans the referendum on EU membership in the United Kingdom, where immigration played a prominent role. Overall, these results provide simple and persuasive evidence that immigration attitudes are very stable, even during periods where the previous literature would suggest large changes.

We now move to a more sophisticated analysis of the stability of immigration attitudes by estimating latent structural equation models for the LISS, NCP, and BES panels. An individual’s response to a survey question is a combination of their true unobservable attitude toward immigration and measurement error. More formally, let $x_t$ be the respondent’s answer to the survey question $i$ at time $t$ and is a function of their latent immigration attitudes ($Y_t$) and an error term ($_i,t$);The error term has mean of zero and a variance of $^2__i,t$.

x_i,t= _i,,t*Y_t + _i,t,

which represents the measurement component of the model. The relationship between the latent immigration attitudes at the different values of $t$ is the structural component and is the element of interest in the model. It is modeled as a lag-1 process, which implies that that immigration attitudes at $t$ are a function of the respondent’s immigration attitudes at $t-1$ and some disturbance term;

Y_t = _t-1*Y_t-1 + _t for\ t&=2,3,…,T \ Y_t = _t for\ t&=1.

After accounting for immigration attitudes at $t-1$, immigration attitudes at $t$ do not depend on earlier values. By including the disturbance term ($t$), the model implies that immigration attitudes at time $t-1$ do not perfectly predict immigration attitudes at time $t$.We assume that the mean of $_t$ is zero and estimate its variance $^2$. The $$ coefficients are the stability estimates and the main quantity of interest. Values closer to one imply a strong relationship between the underlying latent immigration attitudes across waves and values closer to zero imply a weak relationship.

Previous studies analyzing attitude stability often only use three-wave single indicator models developed by @Wiley1970. Though, using additional waves and multiple indicators allows us to relax some of the more problematic assumptions. Our approach is superior to much of the previous literature for two reasons. First, the three wave single indicator models are just identified, and thus, need to assume that error variances are constant and not correlated across panels. This assumption is problematic because error variances will decline if respondents become more familiar with the survey question after each wave. Further, errors may be correlated if respondents are consistently confused by the same elements of the question. By extending the number of waves and using multiple indicators we can test and relax these assumptions. Second, since the three wave single indicator model is just identified, it is not able to assess model fit. In contrast, models with more than three waves and multiple indicators are over-identified and can estimate model fit statistics.There are some additional assumptions that we must make. First, we assume that the disturbance terms are uncorrelated, $E[_t, _s] = 0$ for $t s$; second, we assume the disturbance terms are not correlated with latent immigration attitudes in the previous waves, $E[_t, Y_s] = 0$ for t$>$s; third, we assume that the error terms are uncorrelated with latent immigration attitudes, $E[_t, Y_t] = 0$; and finally, we assume that the error terms are uncorrelated with the disturbance terms, $E[_t, _t]=0$. In some models we also assume that errors terms within each period and across waves are not correlated; however, we test and relax these assumptions. To estimate the models, we also must constrain one factor loading ($_i,t$) to 1. We constrain the factor with the highest loading, so all other factors are below 1.

We assess model fit through four approaches; the Comparative Fit Index (CFI) and its extension the Tucke and Lewis Index (TLI), where values of 0.90 or higher indicate good model fit; the Standardized Root Mean Square Residual (SRMSR), where values less than 0.08 indicate good model fit; and finally the Root Mean Square Error of Approximation (RMSEA), where values less than 0.08 indicate good fit.

The results are reported in Table tab:semresults. In Columns (1), (3), and (5), we estimate models that assume the error terms are independent. The stability coefficients are all near 1.00, indicating high stability from one time period to the next. The fit statistics for the three panels suggest that the models can be improved. None of the models consistently meet the criteria for the goodness of fit tests. It is likely that the error terms for each question are related across time periods since these questions do not change. If respondents make an error on a specific question during one wave, they will probably make a similar error on the question during other waves. The error terms for the questions within the same time period may also be related. This would be the case if the specific context at the time of the survey influences respondents’ answers. We use Modification Indices and Lagrange Multiplier tests to examine potential violations in the samples. In Columns (2), (4), and (6), we estimate models removing these constraints. The stability coefficients are all still close to 1.00. Importantly, the fit of the models drastically improve. The CFI and TLI values are all 0.99 or higher. Additionally, the RMSEA and SRMSR values are all below 0.03. It appears the models fit the data very well.

Further, measurement models estimate the reliability of individuals’ latent immigration attitude, which is the true score variance divided by the total variance in the observed indicator. For all three panels, the reliability estimates are quite high, suggesting internal consistency. Specifically, for the LISS panel the estimates for each wave range from 0.79 to 0.82, for the BES panel the estimates are between 0.87 and 0.93, and for the NCP the estimates are between 0.69 and 0.83.

Table. Measurement models for stability in immigration attitudes

Overall, the results from the measurement models suggest that immigration attitudes are highly stable. The lowest estimated stability coefficient is only 0.93. Further, of the 40 stability coefficients estimated, only 10 have 95 percent confidence intervals that do not include 1.00. (Note: Measurement models do not always produce estimates near 1.00. See @Feldman1989 and Green (2004).) This stability is quite remarkable given the recession and the refugee crisis during this time period. Importantly, the previous literature that emphasizes “communication” and “environmental” variables would suggest that the economic downturn and influx of migrants should cause a change in immigration attitudes. Thus, finding stability in immigration attitudes during this period should strengthen our confidence that attitudes are stable when the economic and political climate are less volatile.

Variation in the stability of immigration attitudes

We examine potential variation in the stability of immigration attitudes using the LISS and BES panels. We focus on two hypotheses derived from the broader public opinion and political socialization literatures. We leave other potential heterogeneity in the stability of immigration attitudes to future research. First, a key debate in the literature examining the stability of political attitudes centers on how to conceptualize the random variation in survey responses. While we follow @Achen1975,Ansolabehere2008a and assume this random variation is measurement error, some scholars suggest that this variation is evidence of non-attitudes or, in other words, respondents randomly answering survey questions [@Converse1964]. Converse develops a “black-white” model, which posits that respondents can be divided into two groups: a minority, the politically sophisticated, with stable attitudes, and a large majority, the unsophisticated, with non-attitudes.

We follow @Ansolabehere2008a and test this hypothesis by examining temporal correlations of immigration attitudes across levels of political sophistication. We measure political sophistication in two ways: education and political knowledge. For education, we divide the population into two groups: a high-education group, which includes individuals with at least a college degree (high sophistication), and a low-education group, which consists of those with a high-school degree or less (low sophistication). For political information, the LISS and BES panels do not include similar measures. The BES panel includes our preferred measure, which is a series of 10 factual questions that asks respondents to identify the names and positions of several foreign leaders, to identify their district representative in Parliament, and to answer various questions about the European Union. We divide the respondents into two groups: a high-information group, which consists of respondents who answered at least six questions correctly, and a low-information group, which includes those who correctly answered fewer than six questions. (Note: The mean number of questions answered correctly is 4.77.) For the LISS panel we opt for an alternative measure based on self-reported political and news interest. (Note: The question wording for political interest is: “Are you very interested in political topics, fairly interested or not interested?” The question wording for news interest is: “Are you very interested in the news, fairly interested or not interested?”) The high-information group consists of individuals who indicated they were “very interested” in both the news and political issues and the low-information group includes all other respondents.

The second hypothesis we test is from the political socialization literature. Specifically, the impressionable years hypothesis posits that older adolescents and young adults are developing their core political attitudes. The key empirical implication from this hypothesis is that younger individuals should have lower correlations across time because they are responding to contextual factors. To test this hypothesis we divide the population by age. The older group consists of respondents that are at least 30 years old while the younger group includes respondents that are younger than 30. The results are reported in Table tab:variation.

tables/table4

There is some evidence of heterogeneity in the stability of immigration attitudes by political sophistication. The correlations are smaller for respondents in the low education and low political information groups compared to the groups with high education and high political information. Though, these differences are small. The largest differences for the LISS and BES panels are 0.095 and 0.058, respectively. While differences do exist, the correlations are fairly similar across levels of political sophistication. Importantly, it appears that measurement error is a much larger issue. For example, in the LISS panel, the differences between the correlations of the scale and individual survey items are both around 0.20, which is more than two times the difference between the high and low political sophistication groups. The results are similar for the BES panel. Thus, there is limited empirical support for Converse’s (1964) “black-white” model, at least concerning immigration attitudes.

Finally, we find support for the impressionable years hypothesis. Specifically, respondents below the age of 30 have substantially lower correlations between waves compared to older individuals. The difference between the older and younger group for the LISS panel when using the scale is 0.204. For the BES panel, the difference is 0.098. These differences are similar when using the individual survey items, which suggests the finding is not an artifact of measurement error.

Discussion and conclusion

The previous public opinion literature on immigration has largely ignored the empirical reality and the related theoretical implications of attitude persistence. While most scholars assume that attitudes toward immigration are quite flexible, there has been no comprehensive test of this of assumption. In this article, we extensively examined the stability of immigration attitudes using seven panel surveys and a variety of methodological approaches to account for measurement error. The evidence suggests that immigration attitudes are remarkably persistent and hard to change, even during economic and political shocks. Our results also indicate that stability does not vary across levels of political sophistication. Furthermore, consistent with the political socialization literature, we find the younger individuals have lower stability compared to older adults.

Overall, these findings have important implications for a number of theoretical debates in the literature that so far have been largely concerned with cross-sectional rather than temporal variation in immigration attitudes. At the same time, the existing experimental and quasi-experimental research has often focused too narrowly on immediate attitude changes, instead of the enduring effect of a particular shock. Inasmuch as attitudes do not change across time, despite changing economic conditions and demographic context, our results provide more support for explanations that emphasize the role of stable predispositions rather than communication or environmental factors. While we do not dispute the validity of previous results on the role of information and ever-changing environment, our results imply that these factors can only explain a small minority of the underlying variation in immigration attitudes. Future research would thus benefit greatly by explicitly specifying and testing the lasting temporal implications of proposed theories, including those focusing on economic and non-economic variables. (Note: The immigration literature, however, is not alone in failing to consider the reality and implications of attitude stability. Much of the attitudinal literature in international relations and comparative politics has ignored the extensive debates on the nature of public opinion.)

Our results also run counter to many theories that emphasize the role elite position taking plays in influencing the attitudes (rather than just issue salience) among the mass public [@Lenz2012]. At least when it comes to immigration, individuals may not be “following the leader.” Relatedly, our results indicate that the recent rise of populist and radical parties cannot be explained by alluding to the change of likely stable immigration attitudes [@Bonikowski2017]. Consequently, future research on the topic can benefit from focusing on attitude salience and issue importance [@Dennison2018].

Finally, the results also provide support to a broader literature emphasizing the importance of early life experiences in the development of attitudes. Consistent with the “persistence” and impressionable years models, we find evidence that younger individuals experience substantially more response variation in immigration attitudes. If these attitudes begin to crystallize when individuals are young, scholars would benefit from specifying the conditions during this period that influence the attitude development [@Laaker2018]. While our data indicates that young people are more susceptible to change, future research needs to more rigorously explore the specific period when these attitudes begin to develop.

Of course, our research is not without limitations. Most important, our conclusions on the stability of immigration attitudes after correcting for measurement error are potentially challenged by questions concerning the source and meaning of random error [@Zaller1992, Feldman1992]. (Note: Response instability may arise because individuals are influenced by the contextual factors at the time of the survey. Individuals may have an underlying distribution of potential answers that they draw from when responding to survey questions. Under this model, all respondents have some central tendency in their response, but also have variance. While measurement models treat this variance as random error, @Feldman1992 argue that it is substantively important variation that should not be ignored.) Nonetheless, we provide strong evidence that the existing random variation is in fact measurement error by showing that the stability of immigration attitudes increases when using multiple survey items. Even more important, this variation is rather small, suggesting that whether we call this random component “measurement error” or “meaningful variation” does not affect the conclusion of stability.

Furthermore, the available data do not allow differentiating between the underlying attitudes and social norms. For instance, stability may be a result of the constant legal environment regarding immigration in most countries [@Tankard2016]. While many of the shocks examined in the literature do no seem to have a lasting effect on immigration views, this may not apply to truly systemic changes such as related to major wars and regime changes. Therefore, our results do not at all imply that immigration attitudes cannot be changed in principle or be different under alternative political institutions. (Note: There is some evidence of the recent (positive) change in aggregate US attitudes over the last several years (see the recent Pew Research report). It is unclear, however, how robust these changes are. Consequently, future research can leverage more recent panel data when possible.)

Related to this, our work does not examine other changes that may be occurring at the individual level. For instance, changing party positions may play an important role in altering the political behavior of individuals. As parties’ stake out divergent positions on immigration, the impact of new policy stances may lead to changes in individual party affiliation similar to how changing positions on civil rights led many whites to defect from the Democratic party [@Carmines1990, Caughey2018b]. While some research exists [@Hajnal2014,Carsey2006,Mader2018], further examination of how enduring opinions can lead to partisan change may be warranted.

All these limitations notwithstanding, the study provides the most comprehensive evidence of the attitudinal stability on immigration with important implications for the literature. Most prominently, our findings imply that scholars should exercise caution in using communication and environmental factors (such as information or economic conditions) to explain immigration attitudes or using immigration attitudes to explain political change (such as the rise of populist parties). More broadly, our research suggests that public opinion and immigration scholars alike would benefit from examining the long-term temporal variation of individual attitudes.

References

See source bibliography.