Does Health Vulnerability Predict Voting for Right-Wing Populist Parties in Europe?
Yes and No, Depending on the Covariates
Abstract
Abstract:Introduction: A recent article by Kavanagh et al. (2021) in American Political Science Review suggested that health vulnerability predicts voting patterns for the populist far-right. Aim: We sought to distinguish socioculturally conservative from antielite voting, expecting health vulnerability to predict only the former. Methods: We combined data from the European Social Survey (round 7) with the Chapel Hill Expert Survey (2014). We preregistered several other hypotheses regarding associations between health and voting and predetermined the smallest effect size of interest (SESOI: r = .05). Results: Health vulnerability did not predict socioculturally conservative voting, and the results for antielite voting were mixed. To investigate why our result differed from Kavanagh et al.’s, we reran their analysis employing covariate specification curve analyses. The statistical significance and the direction of the association between health vulnerability and voting depended on which covariates were included. Across 1,000 models with randomly drawn covariate specifications, 59.6% showed a positive, 29.8% a zero, and 10.6% a negative association between health vulnerability and voting for far-right populist parties. However, all effect sizes were more consistently smaller than our predetermined SESOI. Conclusions: Our study illustrates the necessity of causally justifying and preregistering all covariates and predetermining an SESOI.
Attempts to explain the global rise of right-wing populism have increased in recent years. In addition to the usual suspects of economic insecurity and cultural backlash, Kavanagh et al. (2021) introduced health vulnerability as a potential explanation for why people vote for populists. They analyzed European Social Survey (ESS) data to show that the voter’s perceived health may meaningfully contribute to populist support, i.e., they argued that illness and physical disability increase an individual’s sense of personal vulnerability; they blame their misfortunes on the political establishment, desire “changing the political status quo” (Kavanagh et al., 2021, p. 1104), and vote for “parties that campaign for a fundamental restructuring of a ‘biased and broken’ system” (Kavanagh et al., 2021, p. 1104).
Kavanagh et al. (2021) used two authoritative lists (Inglehart & Norris, 2016; Norris & Inglehart, 2019) of European right-wing populist parties to determine whether respondents had voted for either a right-wing populist party or another party in their country’s last national elections. One list was based on party policy positions (Inglehart & Norris, 2016), the other on party antielite rhetoric and the salience of anticorruption in the party platform (Norris & Inglehart, 2019). Kavanagh et al. categorized the parties mentioned on either of these two lists as right-wing populists; that is, they categorized a party as a right-wing populist party either based on party policy positions or antielite rhetoric and the salience of anticorruption. This confounding of right-wing policy positions and general antielitism did not allow them to separate the “thick” and the “thin” aspects of right-wing populism (see Mudde, 2007). Based on social psychological research on terror-management theory (Greenberg et al., 1990) and on system-justification theory (Jost & Banaji, 1994), we preregistered several hypotheses building on the idea that it was right-wing ideological content and not populism per se that drew the votes of those who reported poor health.
Definition and Measurement of Populism
Although populism has grown into an animating force of contemporary European electoral politics, its substance remains elusive. In the literature, populism has alternatively been conceived as an ideology, a strategy, and a communication style (Van Kessel, 2014). Much of the work on populism follows Mudde and Kaltwasser (2017) in assuming that populism interprets politics as a struggle between the corrupt political elite and the pure people. The most popular measures of populism also tend to focus on this antagonism, including items such as “Politicians in parliament need to follow the will of the people” and “Political differences between the elite and the people are larger than differences among the people” (Akkerman et al., 2014). This critique of the elite is not confined to the political elite but can also be directed toward other elite groups – such as government officials, the media, scientists, and big business – who may all be seen as incompetent, arrogant, and selfish (Barr, 2009). Populism is commonly thought of as a “thin” ideology, in the sense that it always combines with different “thick” or “host” ideologies that provide populism with programmatic content (Mudde & Kaltwasser, 2017). This means that there can be both left-wing and right-wing forms of populism.
Kavanagh et al. (2021) relied on two studies by Inglehart and Norris (2016) and Norris and Inglehart (2019) for their classification of parties as right-wing populist. In both studies, Inglehart and Norris employed the Chapel Hill Expert Survey (CHES; Bakker et al., 2015), which asked experts knowledgeable about European political parties to estimate the ideological and policy positions of political parties in the country with which they were most familiar. The nine CHES items on which Inglehart and Norris (2016) constructed their first populism index were favoring traditional social values, nationalism, tough law and order, religious principles in politics, and supporting rural interests (but opposing liberal lifestyles), multiculturalism, immigration, and rights for ethnic minorities. They summed the items and standardized the resulting scale to 100, labeling it the CHES Cultural Scale. Parties scoring more than 80 points on this scale were labeled right-wing populist parties.
In their subsequent study, Norris and Inglehart (2019) took a different approach, employing two different CHES items to assess populism – the importance of antiestablishment and antielite rhetoric and the salience of anticorruption. These two items were summed and standardized to 100, and parties scoring over 80 were considered populist. Before combining the two lists for their definite classification of parties as right-wing populist or not, Kavanagh et al. (2021) excluded left-wing populist parties that had initially been included on the latter list (Norris & Inglehart, 2019), such as Syriza of Greece and Podemos of Spain.
The present research is based on the notion that it could be worthwhile to separate populism, understood as far-right policy positions (Inglehart & Norris, 2016), from populism understood in terms of the antagonism between the “people” and the “elite” (Norris & Inglehart, 2019). The latter approach to defining and measuring populism better reflects that populism is commonly considered not only an ideology but also a strategy and a communication style (Van Kessel, 2014). Moreover, it captures the idea that populism is a thin ideology (Mudde & Kaltwasser, 2017) that can latch onto parties on both the political left and right. To disentangle ideological content from antielitism and to allow for the inclusion of left-wing populism, we made the CHES ratings of antielite rhetoric our measure of populism. In contrast to Norris and Inglehart (2019) and subsequently Kavanagh et al. (2021), we did not include anticorruption salience in our measure of populism; research on the CHES ratings of anticorruption salience has shown this dimension to be strongly related to the national context and not that relevant in most European countries. We measured party ideology using CHES ratings of ideology and policy positions. Also in contrast to Inglehart and Norris (2016), Norris and Inglehart (2019), and subsequently Kavanagh et al. (2021), we did not dichotomize the CHES ratings but used the original 11-point scales to rate all items. There is compelling evidence that continuous indicators are superior to dichotomized indicators (DeCoster et al., 2009). Nor did we collapse the items into sum scores or combine them in any other way. There has been a recent shift toward individual items rather than summary statistics across several fields of psychology: To thoroughly describe and understand a complex phenomenon, specificity is more important than simplicity (Kertzer & Powers, 2020; Mkhitaryan et al., 2019; Mõttus et al., 2019, 2020; Schmittmann et al., 2013). This is especially true if, as in the present context, there is no set-in-stone structure of party policy positions to refer to when considering how to construct the summary variables.
Health Vulnerability and Voting Populist
Kavanagh et al. (2021) suggested that health vulnerability can predict why people vote for right-wing populists. However, the parties they identified as populist were defined not only in terms of antielitism but also in terms of having very right-wing policy positions (Inglehart & Norris, 2016). Thus, one could interpret the results of Kavanagh et al. (2021) to suggest that health vulnerability predicts voting for parties with extremely right-wing policy positions – not populism per se.
It is important to note that Inglehart and Norris (2016) – and subsequently Kavanagh et al. (2021) – do not refer to right-wing economic policy. In Europe, political ideology is structured by two independent dimensions (e.g., Feldman & Johnston, 2014): attitudes toward fiscal and economic policies (economic left-right, i.e., the extent to which the state should be involved in running the economy) and attitudes toward social or cultural issues. Following CHES (Bakker et al., 2015; Jolly et al., 2022), we refer to the latter dimension as “gal-tan,” in which “gal” is an acronym for “green, alternative, liberal” and refers to support for more expansive personal freedoms – greater civil liberties, same-sex marriage, a greater role for citizens in governing, etc. At the other pole lies “tan,” which refers to “traditional, authoritarian, nationalist,” rejects these ideas and favors law and order, family, religion, and customs, traditional morality, and a national way of life. The nine above-described policy position items employed by Inglehart and Norris (2016) to construct their classification of parties as right-wing populist fall under the gal-tan dimension, not the economic left-right dimension.
Building on social psychological research on terror-management theory (Greenberg et al., 1990) and on system-justification theory (Jost & Banaji, 1994), we hypothesized that it was the very right-wing policy positions (right-wing in the sense of tan over gal) and not populism (in the sense of antielitism) that drove the association between health vulnerability and populism Kavanagh et al. (2021) reported on. First, regarding terror-management theory, physical health problems and psychological distress are associated with more thoughts of death (Feifel & Branscomb, 1973; Fortner & Neimeyer, 1999). Experiments conducted within the framework of terror-management theory have shown that, after contemplating their own deaths, people report poorer opinions of those perceived as different (Routledge et al., 2010), provide them less resources (Tam et al., 2007), and punish them more harshly (Rosenblatt et al. 1989). This antipathy toward those perceived as different could translate into voting for tan parties over gal parties.
On the other hand, the literature on system-justification theory suggests that, when people feel a lack of personal control that threatens their overarching sense of order, they compensate by turning to and defending social systems (e.g., governments) that can reassure them that things are under control (Kay & Friesen, 2011). In terms of voting behavior, this implies that, when people feel vulnerable, they would be expected to vote for the status quo, not antielite parties seeking to overthrow the system.
In sum, we sought to dig deeper into the results reported on by Kavanagh et al. (2021), first and foremost by employing measures of party positioning that allowed us to disentangle voting far-right (tan over gal) from voting antielite. We hypothesized only the former would be associated with health vulnerability. Another difference was using several continuous measures rather than a dichotomous measure. Moreover, as we expected health vulnerability not to predict antielite voting, we needed to test for the absence of meaningful effects. That was our preregistered plan. However, the unexpected results set us on a more exploratory path, leading us to focus on investigating and showcasing the importance of selecting and preregistering control variables.
Research Questions and Hypotheses
Our first research question (RQ1; the numbering of the research questions and hypotheses corresponds with the preregistration) was whether health vulnerability was associated with the Chapel Hill-rated political ideological positions of the party for which one had voted. We expected no association with the economic left-right dimension (H1), but, regarding gal-tan, we expected tan voters to be higher in health vulnerability (H2). Second, we asked whether health vulnerability was associated with the Chapel Hill-rated salience of antielitism in the agenda of the party one had voted for (RQ2). We did not expect the salience of antielitism to be associated with health vulnerability (H3).
Regarding more specific party policy positions (RQ3), we expected health vulnerability to be associated with such policy positions that were related to the gal-tan dimension. Specifically, we expected health vulnerability to (1) predict support of tough measures to fight crime (H4), religious principles in politics (H6), a restrictive policy on immigration (H7), assimilation (H8), nationalist rather than cosmopolitan conceptions of society (H10); (2) predict opposition to liberal lifestyles (H5) and rights for ethnic minorities (H9); and (3) not be associated with policy positions on tax spending (H11), deregulation (H12), redistribution (H13), economic interventionism (H14), the urban-rural divide (H15), the environment (H16), regions (H17), and international security (H18). Regarding hypotheses H11 to H18, we expected no associations because these policy positions either align with the economic left-right dimension (H11-H14), for which we expected no association, or for which the associations, if any, are likely to be shaped by country-specific geographic policy (H15, H17) and the security environment (H18). Regarding the environment (H16), the CHES item opposes environmental protection – which was still supported across the political spectrum in 2014, when the data was collected (see, for instance, Gustafson et al., 2019) – with economic growth, making it more of an economic than social or cultural item.
Finally, we expected health vulnerability to negatively predict voting for antielite parties when the gal-tan position of the party was controlled for (RQ4, H19; those feeling vulnerable should, according to system justification, resist change).
Methods
Preregistration
The study was preregistered, and all data and materials are publicly available on the Open Science Framework (OSF). For the preregistered research questions, analysis plans, and hypotheses, see https://osf.io/drus4. Data, analysis, scripts, and other supplementary materials are stored at https://osf.io/epk24/.
Data
Following Kavanagh et al. (2021), we used ESS round 7 data collected in 2014 (Jowell et al., 2006) from 20 countries (n = 37,623; we excluded Israel because it is not included in CHES). All respondents who had indicated vote choice were initially selected for the analysis (n = 21,644), but we had to exclude a further 463 because of other missing variables (final n = 21,181).
Following Inglehart and Norris (2016) and Norris and Inglehart (2019), we employed the CHES wave from 2014. Moreover, at the time of writing, the 2014 CHES wave was the most recent wave, and it also was the wave that coincided well with the collection of ESS data.
For readers unfamiliar with the surveys, the ESS is a biennial multicountry survey carried out since 2002. Representative national samples from more than 20 European countries have participated in every ESS round. The CHES is an expert survey. Expert surveys are often used to measure complex phenomena that cannot be directly observed (e.g., to assess levels of democracy, positions on EU integration, or immigration policy). The experts the CHES relies on are professional political science researchers who have published on party politics or related themes. The expert ratings are collected from around ten experts in each of the participating countries. Extensive research on these ratings has shown them to be a reliable and valid source of information on party positioning on ideology, validated, for instance, against analyses of party manifestos, other expert rating surveys, and public opinion (Bakker et al., 2015; Hooghe et al., 2010) and party antielite rhetoric (Polk et al., 2017).
We report how we determined our sample size, all data exclusions (the criteria for data exclusion were established before data analysis, see the preregistration), all measures, and all analyses, including all tested models. We report exact p values, effect sizes, and 95% confidence intervals.
Measures
Following Kavanagh et al. (2021), we operationalized health vulnerability using two continuous self-reported health measures: (1) subjective general health, which the ESS elicited from respondents on a 5-point scale ranging from very good to very bad, and (2) being hampered in daily activities by illness, disability, mental health problems, or other infirmities, measured on a 3-point scale from none to a lot.
The party for which the person had voted was the party the participant reported voting for in the previous national election. For each such party, corresponding scores on economic left-right, gal-tan, 16 policy positions (e.g., redistribution, multiculturalism; for all policy positions, see Table 1), and antielite salience were obtained from CHES. For instance, the gal–tan item was: “Parties can be classified in terms of their views on democratic freedoms and rights. ‘Libertarian’ or ‘postmaterialist’ parties favor expanded personal freedoms, for example, access to abortion, active euthanasia, same-sex marriage, or greater democratic participation. ‘Traditional’ or ‘authoritarian’ parties often reject these ideas; they value order, tradition, and stability, and believe that the government should be a firm moral authority on social and cultural issues.” All items are fully described on the CHES website (https://www.chesdata.eu/). All scores ranged from 0 to 10 (e.g., scores on gal-tan range from 0 = Libertarian/postmaterialist, 5 = Center, to 10 = Traditional/authoritarian, and scores on antielite salience range from 0 = not important at all, to 10 = extremely important). These scores were scaled with mean and standard deviation values across all parties (see Table 1 for means and standard deviations).
We controlled for age, sex, years of full-time education, belonging to a minority ethnic group, and rurality of domicile. The research of Wysocki et al. (2022), who showed that controlling for relevant confounders can debias the estimated causal effect of a predictor on an outcome, guided our choice of covariates. However, this is true only when the controlled-for variables are causally before the predictor and the outcome. Controlling for inappropriate third variables – variables that function as colliders, mediators, or proxies (see Figure 3 in Wysocki et al., 2022, for an excellent illustration) – result in more biased estimates. Our five control variables are causally before both perceived health vulnerability and voting behavior. Furthermore, based on previous research, all five covariates can be expected to predict voting behavior (Ford & Jennings, 2020) and health vulnerability (Gkiouleka & Huijts, 2020; Mackenbach, 2019). This means that controlling for them should give a better estimate of the true causal effect of health vulnerability. We controlled for the fixed effects of country to adjust for differences in standard of living and evaluation of health across the countries in our sample.
Analysis Strategy
We examined the hypothesized associations from a set of survey-weighted regression models that used the voting variables one at a time as the dependent variable in distinct models. For each dependent variable, we constructed two models, one that predicted voting with subjective general health, the other that predicted being hampered by disability. All models included the five above-mentioned confounders as covariates. Following the analytical procedure used by Kavanagh et al. (2021), we used poststratification weights, including design weights provided in ESS in all analyses. This allowed correcting for differential selection probabilities, nonresponse, noncoverage, and sampling error (Kaminska, 2020).
To obtain regression coefficients corresponding to standardized regression coefficients, we scaled the health vulnerability variables with the standard deviation pooled across all countries (so as not to inflate SD with between-country variability). We standardized the expert ratings of party position to indicate scaled deviations from an average party in Europe.
One of our claims was that health vulnerability does not predict voting antielite. However, the large sample size of the ESS can produce statistical significance even for effects so small that they best be ignored. We employed equivalence testing to allow us to test for the absence of a meaningful effect (Lakens et al., 2018). The challenge with this approach lies in specifying the smallest effect size of interest (SESOI). We followed recent guidelines (Funder & Ozer, 2019), according to which an effect size r of .05 indicates a very small effect; effect sizes smaller than this are considered meaningless.
We used two types of hypothesis tests. For expected associations, we tested the statistical significance against the smallest effect size of interest, that is, against the null hypothesis b = .05. When we expected the absence of an association, we ran an equivalence test (Lakens et al., 2018) with equivalence bounds set at b = –.05 and b = .05. That is, we tested whether the regression coefficient b fell within this range.
Statistical Power
We had sensitivity to detect associations ranging, at minimum, from r = .073 to r = .096 as being significantly larger than the smallest effect size of interest, when setting both type-I and type-II errors at .05 (power = .95; see the above-linked open materials for reproducible calculations). For equivalence tests, we had a .94 to .99 power to identify zero effects between r = –.05 and r = .05 with type-I error set at .05.
Results
Confirmatory Results
Contrary to our prediction (H1), the association between health vulnerability and voting economic left-right, b = −0.05, 95%CI [−0.07, −0.04], did not fall within the equivalence bounds, p = .728. However, in absolute terms, it was also not larger than SESOI (testing against the null hypothesis b = −.05, p = .272). The pattern was similar when hampered by disability-replaced health vulnerability (b = −0.05, 95%CI [−0.06, −0.03], equivalence to [−.05, .05] p = .432). These results suggest an unexpected but weak deviation from zero.
Also contrary to expectations (H2), there was no association between health vulnerability and voting on the gal-tan dimension (subjective general health: b = 0.00, 95%CI [−0.01, 0.02], p (for b > .05) > .999; hampered by disability: b = 0.00, 95%CI [−0.01, 0.01], p (for b > .05) > .999). Regarding antielitism (H3), the results were mixed (subjective general health: b = 0.05, 95%CI [0.03, 0.06]; equivalence to [−.05, .05] p = .263; hampered by disability, b = 0.03, 95%CI [0.02, 0.05]; equivalence to [−.05, .05] p = .009).
The results regarding policy positions were generally in line with the above results, with health vulnerability associated with party policy positions on the economic left, such as support of higher taxes and more redistribution. By contrast, there were no associations between health vulnerability and more social or cultural policy positions, such as immigration policy and multiculturalism. The results regarding policy positions were thus generally opposite to what we had expected (H4-H10). The results supported only those hypotheses that did not expect an association (H11-H18). We also found no support for an association between health vulnerability and voting antielite when controlling for the gal-tan ideology of the party (H19). Table 2 and Table 3 report all results in detail. What generally stands out is the lack of effects. At the outset, we set r = .05 as the smallest effect size of interest. Had we demanded the slightest bit more, we could have ruled out the existence of any associations between health vulnerability and voting.
Exploratory Results
Given the lack of confirmatory results, we embarked on a more exploratory path to investigate possible reasons why our results differed from those of Kavanagh et al. (2021). One possible explanation for the disparate results was our reliance on different covariates. Wysocki et al. (2022) recently showed that controlling for inappropriate variables can severely distort the results of regression analyses; following their recommendations, we included only control variables we were certain were causally before the independent and dependent variables.
By contrast, Kavanagh et al. (2021) included 20 covariates, many of which (e.g., cultural attitudes, left-right self-placement, unemployment, income) are not unequivocally causally before health vulnerability or voting. We thus set out to investigate whether different covariates could help explain the disparate results. To investigate the robustness of the association between health vulnerability (the 5-point measure) and voting populist (the categorical measure employed by Kavanagh et al., 2021), we drew a random sample of 1,000 different covariate specifications to allow us to examine the covariate specification curve (see Simonsohn et al., 2020). We could not include all possible covariate combinations as the number of different combinations – and hence models – exceeded a million. In our random sample of 1,000 covariates, we always included country and survey year as covariates. In contrast, all other of the 20 covariates employed by Kavanagh et al. (2021) had a 50% chance of being included in a specific model. Our analyses were identical to those run by Kavanagh et al. (2021), except that we only used the first (of five) imputed datasets to restrict the multiverse of possible analysis. Figure 1 illustrates the results. Slightly more than half (59.6%) of the randomly specified models showed an association between health vulnerability and voting for far-right populist parties. The rest of the models showed either zero (29.8%) results or a negative (10.6%) association.
The direction of the association between health vulnerability and voting right-wing populist thus varied as a function of the covariates specified in the model. Regarding the strength of this association, the mean effect size across the 1,000 different covariate specifications was OR = 1.034, and the model with all covariates (pinpointed in Figure 1) showed OR = 1.057. Scaling these log-odd estimates with the standard deviation of health vulnerability gave a mean OR of 1.038, and an all covariates model an OR of 1.065. In correlation metrics, these correspond to r = .010 and r = .017, respectively. Figure S1 (see https://osf.io/epk24/) presents a specification curve for OR estimates scaled with the standard deviation of health vulnerability (this correction to the ORs was not done in Figure 1, as we initially wanted to facilitate comparison with the analysis run by Kavanagh et al., 2021). All but two correctly scaled OR estimates from the thousand randomly selected covariate specifications suggested no effect (equivalence to [−.05, .05]).
Discussion
At the outset, based on the results of Kavanagh et al. (2021), we believed there was an association between health vulnerability and voting for right-wing populists. We wanted to dig deeper into this association by disentangling party ideology, or programmatic content, from antielite posturing. We also employed an analysis strategy, i.e., equivalence testing, which would allow us to preset an SESOI and claim that antielite voting is not meaningfully associated with health vulnerability. However, our results led us to believe we were on a wild goose chase: There may not be any association between health vulnerability and voting.
The covariate specification curve analysis we ran shows that the results Kavanagh et al. (2021) reported are not robust but highly dependent on covariate selection. Null-hypothesis testing against zero in a very large dataset further accentuated the susceptibility of the results to covariate selection, that is, not just the statistical significance of the association but also its very direction.
The larger point that the specification curve analysis highlights is that covariate selection done at a stage at which the data is being analyzed gives the researcher far too much freedom: One could easily find negative, zero, or positive associations between health vulnerability and voting for the populist far-right. Bureaucracies such as preregistration are necessary to limit researcher degrees of freedom (Penders, 2022). Selecting and running a handful of robustness checks afterward does not help achieve this goal when the number of possible checks lies in the thousands, if not millions.
When the number of covariates is large, conclusions can vary as a function of covariate selection, even though everything else stays identical. The often unquestioned assumption that, when it comes to covariates, more is better may be as widespread as it is harmful; one must always consider how control variables are positioned in the causal chain that includes the independent and dependent variables, as this can strongly influence the estimates one obtains (Wysocki et al., 2022). That is, if the control variables are not confounders but either mediators (between the independent and dependent variable in the causal chain) or colliders (located downstream of both the independent and dependent variable), chances are that the estimates one obtains for the main variables of interest will be biased. The more control variables there are, the more complicated their roles in the causal chain become. For robust results, one must (pre-)select only causally justified control variables. Our approach was to select only such covariates we were certain were causally before health vulnerability and voting.
A related challenge stems from the use of large datasets. They are essential for the detection of effects that are of small magnitude. However, they contain the danger that one treats even very small deviations from zero as discoveries. Large data sets demand the assessment of SESOI. Defining SESOI also allows for testing the meaninglessness of associations (Lakens et al., 2018). Treating meaningless, non-null findings as meaningful discoveries may have stifled progress in many social science fields, especially those that use large datasets. The field would perhaps benefit if some lines of research were dealt a death blow by equivalence testing – testing for the absence of a meaningful effect.
Regarding the more substantial aspects of our research question – the implications of health for voting populist – the wisdom of hindsight suggests that the variable-centered approach we employed may not be ideal for the study of populism. Mudde and Kaltwasser (2017), in defining populism, noted that although “in theory, populism is an independent ideology unattached to any other ideology” (p. 12) and in practice it is “almost always combined with other ideological features” (p. 12). To take a well-known example, in the US presidential elections 2016, Bernie Sanders and Donald Trump, both of whom were considered populists (both were against the elites, for the people), campaigned on very different, even opposite, policy platforms (Staufer, 2021). To understand whether populism could help explain their appeal to voters, one must look at how populism combined with their other policy positions. A person-centered approach – which seeks to identify subgroups of people who share characteristics that differentiate them from other sub-populations – could be worthwhile in this pursuit. Those attracted to Trump’s populism could, in many ways, including health vulnerability, be very different from those attracted to Sanders’s populism.
The person-centered approach has been successfully applied in adjacent fields, such as research on prejudice (Dangubić et al., 2021) and conspiracy beliefs (Frenken & Imhoff, 2021) but not in the study of populism. The standard, variable-centered methodology that has looked for causes of the global ascendance of populism has focused on economic and sociocultural grievances. However, the empirical evidence for either of these competing explanations is at best mixed (for a recent review, see Berman, 2021). If the meaning of populism shifts from one context to the next, depending on the other ideological features with which it combines, the search for ubiquitous causal predictors may turn out to be fruitless. Controlling for the variables with which populism combines – the “host” or “thick” ideology – may not be an option. To do this, one would need to know how populism and the “thick” aspects of ideology (e.g., authoritarianism, climate-denialism, nationalism, etc.) are positioned in the causal structure, i.e., one would need to know how independent variables, colliders, mediators, and proxies are arranged (see Wysocki et al., 2022). That may not be possible. Moreover, how the variables are arranged may vary from context to context. In other words, given that populism always attaches itself to “thick” ideologies, it could be helpful to try a person-centered approach in future research on the predictors of populist voting.
We hope our paper can raise awareness of the importance of justifying and preregistering covariates as well as the need to determine and preregister the smallest effect sizes of interest. In the absence of such practices, the number of studies reporting statistically significant but meaninglessly small effects will continue to grow, luring others into pointless and time-consuming research paths. We were lured onto such a path by the purported discovery of a novel predictor of far-right voting. Had we not adopted the methodological rules we now prescribe, we could have ended up contributing to the proliferation of flawed research. Instead, we hope our paper benefits the field by raising awareness and showing what needs to change.
References
(2014). How populist are the people? Measuring populist attitudes in voters. Comparative Political Studies, 47(9), 1324–1353. https://doi.org/10.1177/0010414013512600
(2015). Measuring party positions in Europe: The Chapel Hill expert survey trend file, 1999–2010. Party Politics, 21(1), 143–152. https://doi.org/10.1177/1354068812462931
(2009). Populists, outsiders and antiestablishment politics. Party Politics, 15(1), 29–48. https://doi.org/10.1177/1354068808097890
(2021). The causes of populism in the West. Annual Review of Political Science, 24(1), 71–88. https://doi.org/10.1146/annurev-polisci-041719-102503
(2021). Understanding (in)tolerance of Muslim minority practices: A latent profile analysis. Journal of Ethnic and Migration Studies, 47(7), 1517–1538. https://doi.org/10.1080/1369183X.2020.1808450
(2009). A conceptual and empirical examination of justifications for dichotomization. Psychological Methods, 14(4), 349–366. https://doi.org/10.1037/a0016956
(1973). Who’s afraid of death? Journal of Abnormal Psychology, 81, 282–288. https://doi.org/10.1037/h0034519
(2014). Understanding the determinants of political ideology: Implications of structural complexity. Political Psychology, 35(3), 337–358. https://doi.org/10.1111/pops.12055
(2020). The changing cleavage politics of Western Europe. Annual Review of Political Science, 23(1), 295–314. https://doi.org/10.1146/annurev-polisci-052217-104957
(1999). Death anxiety in older adults: A quantitative review. Death Studies, 23(5), 387–411. https://doi.org/10.1080/074811899200920
(2021). A uniform conspiracy mindset or differentiated reactions to specific conspiracy beliefs? Evidence from latent profile analyses. International Review of Social Psychology, 34(1), Article
27 . https://doi.org/10.5334/irsp.590(2019). Evaluating effect size in psychological research: Sense and nonsense. Advances in Methods and Practices in Psychological Science, 2(2), 156–168. https://doi.org/10.1177/2515245919847202
(2020). Intersectional migration-related health inequalities in Europe: Exploring the role of migrant generation, occupational status & gender. Social Science & Medicine, 267, Article
113218 . https://doi.org/10.1016/j.socscimed.2020.113218(1990). Evidence for terror-management theory II: The effects of mortality salience on reactions to those who threaten or bolster the cultural worldview. Journal of Personality and Social Psychology, 58, 308–318. https://doi.org/10.1037/0022-3514.58.2.308
(2019). The development of partisan polarization over the Green New Deal. Nature Climate Change, 9(12), Article
12 . https://doi.org/10.1038/s41558-019-0621-7(2010). Reliability and validity of the 2002 and 2006 Chapel Hill expert surveys on party positioning. European Journal of Political Research, 49(5), 687–703. https://doi.org/10.1111/j.1475-6765.2009.01912.x
(2016). Trump, Brexit, and the rise of populism: Economic have-nots and cultural backlash (HKS Working Paper No. RWP16-026). https://doi.org/10.2139/ssrn.2818659
(2022). Chapel Hill Expert Survey trend file, 1999–2019. Electoral Studies, 75, Article
102420 . https://doi.org/10.1016/j.electstud.2021.102420(1994). The role of stereotyping in system-justification and the production of false consciousness. British Journal of Social Psychology, 33(1), 1–27. https://doi.org/10.1111/j.2044-8309.1994.tb01008.x
(2006). Measuring attitudes cross-nationally: Lessons from the European Social Survey. Sage.
(2020). Guide to using weights and sample design. Indicators with ESS data. Institute for Social and Economic Research, University of Essex. https://www.europeansocialsurvey.org/sites/default/files/2023-06/ESS_weighting_data_1_1.pdf
(2021). Does health vulnerability predict voting for right-wing populist parties in Europe? American Political Science Review, 115(3), 1104–1109. https://doi.org/10.1017/S0003055421000265
(2011). On social stability and social change: Understanding when system justification does and does not occur. Current Directions in Psychological Science, 20(6), 360–364. https://doi.org/10.1177/0963721411422059
(2020).
Foreign policy attitudes as networks . In A. MintzL. G. TerrisEds., The Oxford handbook of behavioral political science (online ed., Oxford Academic). https://doi.org/10.1093/oxfordhb/9780190634131.013.27 (accessed 15 February, 2024)(2018). Equivalence testing for psychological research: A tutorial. Advances in Methods and Practices in Psychological Science, 1(2), 259–269. https://doi.org/10.1177/2515245918770963
(2019). Health inequalities: Persistence and change in modern welfare states. Oxford University Press.
(2019). Network approach in health behavior research: How can we explore new questions? Health Psychology and Behavioral Medicine, 7(1), 362–384. https://doi.org/10.1080/21642850.2019.1682587
(2019). Personality characteristics below facets: A replication and meta-analysis of cross-rater agreement, rank-order stability, heritability, and utility of personality nuances. Journal of Personality and Social Psychology, 117(4), 35–50. https://doi.org/10.1037/pspp0000202
(2020). Descriptive, predictive and explanatory personality research: Different goals, different approaches, but a shared need to move beyond the big few traits. European Journal of Personality, 34(6), 1175–1201. https://doi.org/10.1002/per.2311
(2007). Populist radical right parties in Europe. Cambridge University Press.
(2017). Populism: A very short introduction. Oxford University Press.
(2019). Cultural backlash: Trump, Brexit, and authoritarian populism. Cambridge University Press.
(2022). Process and bureaucracy: Scientific reform as civilisation. Bulletin of Science, Technology & Society, 42(4), 107–116. https://doi.org/10.1177/02704676221126388
(2017). Explaining the salience of antielitism and reducing political corruption for political parties in Europe with the 2014 Chapel Hill Expert Survey data. Research & Politics, 4(1), 2053168016686915. https://doi.org/10.1177/2053168016686915
(1989). Evidence for terror-management theory: I. The effects of mortality salience on reactions to those who violate or uphold cultural values. Journal of Personality and Social Psychology, 57, 681–690. https://doi.org/10.1037/0022-3514.57.4.681
(2010). Adjusting to death: The effects of mortality salience and self-esteem on psychological well-being, growth motivation, and maladaptive behavior. Journal of Personality and Social Psychology, 99, 897–916. https://doi.org/10.1037/a0021431
(2013). Deconstructing the construct: A network perspective on psychological phenomena. New Ideas in Psychology, 31(1), 43–53. https://doi.org/10.1016/j.newideapsych.2011.02.007
(2020). Specification curve analysis. Nature Human Behaviour, 4(11), Article
11 . https://doi.org/10.1038/s41562-020-0912-z(2021). Donald Trump, Bernie Sanders and the question of populism. Journal of Political Ideologies, 26(2), 220–238. https://doi.org/10.1080/13569317.2020.1855777
(2007). Terror management among Chinese: Worldview defence and intergroup bias in resource allocation. Asian Journal of Social Psychology, 10(2), 93–102. https://doi.org/10.1111/j.1467-839X.2007.00216.x
(2014). The populist cat-dog: Applying the concept of populism to contemporary European party systems. Journal of Political Ideologies, 19(1), 99–118. https://doi.org/10.1080/13569317.2013.869457
(2022). Statistical control requires causal justification. Advances in Methods and Practices in Psychological Science, 5(2), 25152459221095823. https://doi.org/10.1177/25152459221095823
