Neither a Trait nor Wildly Fluctuating: On the Stability of Populist Attitudes and its Implications for Empirical Research

Abstract The literature on populist attitudes frequently makes one of two assumptions: populist attitudes are either stable or unstable. However, few studies have examined these diverging assumptions empirically. We use panel data collected over six panel waves between 2017 and 2021 in Germany to assess the stability of populist attitudes. Integrating inter-individual stability (variable-centred) and intra-individual stability (individual trajectories), we find that populist attitudes are neither fully stable (trait) nor fully flexible (state). For example, some respondents constantly changed their view on populism while the attitudes in one out of three individuals remained stable. We also explore empirical consequences and find that populist attitudes are more closely linked to vote choice when they are stable. Accordingly, we argue for a more nuanced understanding of the dynamics of populist attitudes, both at the variable and individual levels, where these attitudes are stable and consequential for only a subset of individuals.

Since the groundbreaking work by Hawkins and Riding (2010), Hawkins, Riding, and Mudde (2012), and Akkerman, Mudde, and Zaslove (2014), populist attitudes have become a fixture of public opinion and electoral behaviour scholars.Populist attitudes are often defined as referring to a thin ideology that 'considers society to be ultimately separated into two homogeneous and antagonistic groups, 'the pure people' versus 'the corrupt elite', which argues that politics should be an expression of the volonté générale […] of the people' (Mudde 2004, 543).The central ideas of populism form an attitudinal syndrome linked through a non-compensatory mechanism (Wuttke, Schimpf, and Schoen 2020), meaning that populism lies at the intersection of its dimensions, such as anti-elitism and support for popular sovereignty (Akkerman, Mudde, and Zaslove 2014;Castanho Silva et al. 2018;Elchardus and Spruyt 2016;Spierings and Zaslove 2017;Spruyt, Keppens, and van Droogenbroeck 2016).While broad agreement exists on these conceptual questions, there are diverging views among populism scholars about whether populist attitudes constitute a stable trait or a fluctuating state.
Pioneers Hawkins and Riding see populist attitudes 'not [as] a set of conscious issue positions or an ideology, but a worldview or mindset: a set of attitudes so fundamental that most people are unaware that they hold it ' (2010, 2).Subsequent studies propose a similar view of populist attitudes as deep-rooted, although less fundamental in that populist attitudes are not considered a worldview per se.Hawkins, Riding, and Mudde write that their 'findings support a model of populist attitudes as a latent disposition activated by political context ' (2012, 2).This implies that by and large -(non-)support for populist ideas is relatively stable and that populist attitudes are trait-like.To this point, Ardag et al. argue that evidence based on cross-sectional studies has 'shown […] remarkable stability of populist attitudes across countries, suggesting they are not volatile opinions ' (2020, 309).Further, supporting the stability view, multiple studies have explored links between populist attitudes and dispositional personality traits (Kenny and Bizumic 2023;Vasilopoulos and Jost 2020) and found some evidence for dispositional roots of populism (but see Fatke 2019).Yet, other studies adopt a dynamic view of populist attitudes, starting from the assumption that populist attitudes can systematically vary in response to external stimuli such as economic shocks or political communication (Hameleers, Bos, and de Vreese 2017;Hameleers et al. 2021;Rico, Guinjoan, and Anduiza 2020;Rhodes-Purdy, Navarre, and Utych 2021).These studies suggest that it is not uncommon for peoples' populist attitudes to change.These attitudinal changes may be rare and long-lasting, for instance, as a response to external stimuli.Alternatively, changes in populist attitudes may be short-lasting and may occur more frequently, similar to other highly fluid political attitudes.Regardless of the type of change, the dynamic view on populist attitudes assumes that a populist citizen today may be a non-populist citizen tomorrow.
The two main perspectives in the literature start from two fundamentally opposing positions of populist attitudes as either a trait or a state.Our objective is to provide an empirical assessment of the assumptions made about the stability of populist attitudes to inform studies investigating causes of change or stability and the consequences of populist attitudes.Understanding stability is crucial to determine the potential impact of a particular attitudinal construct on other attitudes and behaviours and its position in people's belief systems.For example, an attitudinal construct that is no more than a function of external forces is unlikely to be considered as powerful.By contrast, a construct with inertia that can withstand pressure from the outside may be considered a deep-seated predisposition (Converse 1970).The latter can decisively shape political information processing and behaviour (Howe and Krosnick 2017) and thus will be a factor of political significance, not just a by-product of other forces.What is at stake here are the fundamental questions that we can ask about populist attitudes regarding their origins and their consequences (see also Dennison and Turnbull-Dugarte 2022).
We start from the position that any given concept can be located along a trait-state continuum as most attitudes and behaviour can resemble states and traits simultaneously (Conley 1984;Kenny and Zautra 1995;Kustov, Laaker, and Reller 2021).We analyze individual panel survey data from Germany (GLES 2021), spanning six waves over nearly four years (2017-2021), during which people answered questions relevant to assessing their populist attitudes.Our findings suggest that populist attitudes fall close to the middle of the state-trait continuum, sandwiched between economic perceptions (highly flexible and more state-like) and political interests (highly stable and more trait-like).We discuss the implications of our findings and highlight how future research can incorporate and build on this paper's insights.

Data and Research Strategy
A proper test of the resistance to change of populist attitudes requires an extended period of study that includes environmental variation pushing for attitude change.Without such stimuli, opinions without any resistance to change may remain constant.To this end, we draw on multi-wave survey panel data from the German Longitudinal Election Study (GLES 2021), collected between 10 June 2016 and 12 March 2021.For the analyses presented here, the total N is 5,333. 1 The respondents for this study were drawn from a heterogenous online sample with quota sampling 1 This is the N of respondents for who we have valid populism scores across all six panel waves.The total N of respondents for all relevant sample components of the GLES 2021 study is 22,543.For more information on the panel, see Appendix 1.
(age, gender, and education) to approximate the German adult population with access to the internet.However, self-selection and panel attrition introduced biases not found in random probability samples.In particular, the fact that respondents tend to be slightly more involved in politics and more partisan than the larger population suggests that our analyses are likely biased towards stability (c.f.Gärtner, Gavras, and Schoen 2020).Thus, we proceed cautiously in our primary analyses and analyze respondents for whom we have valid scores in all panel waves and for all relevant variables. 2 Populist attitudes were measured in six panel waves: Wave 1 (17 August-28 August 2017), Wave 2 (27 September-9 October 2017), Wave 3 (15 March-26 March 2018), Wave 4 (20 April-May 6 2020), Wave 5 (3 November-17 November 2020), and Wave 6 (25 February-12 March 2021). 3By selecting these panel waves, the data covers a reasonably long period.The context also provided ample stimuli, which, in theory, could have caused changes in populist attitudesif they were malleable.The period includes the campaign leading up to the 2017 German Federal election on 24 September 2017, an unusually lengthy government formation period that followed the election's conclusion and the onset of the Covid-19 pandemic.These events might have caused changes in populist attitudes (for example, Amat et al. 2021;Gärtner, Gavras, and Schoen 2020).The selected case, period, and data allow us to examine the stability of populist attitudes in a context that favours the fluctuation of populist attitude levels.As such, the case selection reduces the risk of misinterpreting stable opinions as the absence of evidence for flexibility.

Measuring Populist Attitudes
To measure populist attitudes, we draw on a modified version of the validated populist attitude scale proposed by Schulz et al. (2018), which includes items capturing the three dimensions: antielitism, people sovereignty, and the homogeneity of 'the people'.The scale by Schulz et al. (2018) was the second most often used scale at that time (Marcos-Marne, de Zúñiga H, and Borah 2023) and features two of the dimensions that are also included in the most-used scale, the Akkerman, Mudde, and Zaslove (2014) scale, namely anti-elitism and people sovereignty, while also considering homogeneity as a third component of populist attitudes.The panel survey includes three items for each dimension (see Table 1).
For our analysis, we proceed in two steps.First, we average across the three items within each of the three dimensions. 4To ensure that any potential changes in attitudes towards populism over time are substantively meaningful, we tested for longitudinal measurement invariance to ensure that our variables carry 'the same meaning and the same scale over all time points and over all individuals' (Liu et al. 2017, 487). 5One particular challenge is the multi-dimensionality and the non-compensatory nature of populist attitudes, as populism lies at the intersection of its dimensions: anti-elitism, homogeneity, and sovereignty.By implication, we infer that our measure of populist attitudes is invariant if repeated measures of all three dimensions are invariant over time.Using this assumption, we test for measurement invariance in all three dimensions, following commonly applied strategies of adding restrictions to a series of CFA models (See Appendix 2 for details).The results show that all three scales are equivalent over time.The average scores thus accurately capture respondents' positions on each dimension in any given panel wave.We then 2 For more details on missing data, see Appendix 1.

3
In total, the survey panel included 15 waves at the time of the analysis.For simplicity, we use running numbers for the panel waves here, which correspond to the original panel waves as follows (running numbers=original panel wave): 1 = 5, 2 = 8, 3 = 9, 4 = 13, 5 = 14, and 6 = 15.See also Appendix 1.

4
We did not z-standardize items prior to averaging across them to preserve comparability of items over time and thus assume comparable measurement scales across dimensions.take the lowest score across the three dimensions as the aggregate populism score for each respondent.The minimum score across all three dimensions reflects the concept's essential noncompensatory characteristic and ranges from one to five (Wuttke, Schimpf, and Schoen 2020, 361).

Assessing Stability
We combine a variable-based approach to stability (rank-order stability) with the individual trajectory perspective (intra-individual stability).If we exclusively use a variable-based approach, we risk missing important heterogeneity in over-time stability within the population (Asendorpf 2015;Converse 1964;Hill and Kriesi 2001a;Hill and Kriesi 2001b).For example, it may turn out that the state view applies to some people while, for others, populist attitudes resemble traits that occupy central positions in people's belief systems and can shape other attitudes and behaviours.If this is the case, we should find inter-individual differences in the intra-individual stability of populist attitudes, consistent with the idea that we can place both populist attitudes as a concept and individuals along a continuum from trait to state.
To measure rank-order (inter-individual) stability, we correlate the values measured at the first panel wave with each subsequent wave.The higher the smallest correlation across the different measurement points, the more populist attitudes resemble a stable trait.This follows the basic idea that test-retest correlations can be described as a function of time and three different variance components.These components are: (i) a stable trait component that is stable across time, (ii) an autoregressive trait component that exerts some stability but changes over time due to underlying factors which themselves are partly stable, and (iii) a highly flexible state component that captures panel wave specific variance driven by, for example, any momentary factors at the time of measurement.The difference between the smallest correlation and zero captures the stable trait component, the range between the minimum and maximum correlation captures the autoregressive state component, and the range between the maximum correlation of a test-retest series and one captures the state component (Jansen, Lüdtke, and Robitzsch 2020;Kenny and Zautra 1995;Lüdtke, Robitzsch, and Wagner 2018).Thus, knowing time and correlations, we can deduce each of these components to place populist attitudes on a traitstate continuum.
We use two types of analyses to assess intra-individual stability.The first analysis separates respondents into three groups based on individuals' populist attitude scores: a value equal to or above four implies disagreement with populist attitudes, below three implies agreement, and the remaining respondents neither agree nor disagree.Based on this categorization, we label respondents 'stable' if they consistently agree or disagree across all six-panel waves or consistently

Anti-elitism
Politicians talk too much and take too little action.
The differences between the people and the so-called elite are greater than within the people.
Politicians care about what ordinary people think. [R]

Homogenous people
Ordinary people are of good and honest character.
Ordinary people all pull together.Ordinary people share the same values and interests.

People's sovereignty
The people should have the final say on the most important political issues by voting on them directly in a referendum.
The people, not politicians, should make our most important policy decisions.
The politicians in parliament need to follow the will of the people.
Note: [R] indicates reverse coded item.
report 'neither agree nor disagree'.The second group, which we label 'unstable', consists of the remaining respondents.These respondents cross 'opinion boundaries' (Hill and Kriesi 2001a;Hill and Kriesi 2001b) at least once, as they may, for example, agree with populist ideas at one point in time but disagree with them at another.As a second measure of intra-individual stability, we calculate absolute differences and standard deviations for each respondent for subsequent pairs of panel waves and average across the within-mean absolute differences.It is worth noting that these calculations follow the standard practice in the populist attitudes literature to consider respondents who 'neither agree nor disagree' to hold nuanced attitudes that are meaningfully located between the extremes of the scale.Alternatively, one could also conceive of respondents with a mid-point score as ambivalent or lacking a clear attitude (Freeder, Lenz, and Turney 2019;Hill and Kriesi 2001a;Hill and Kriesi 2001b;Kiley 2021;Zaller 1992). 6We follow the first approach but acknowledge that a thorough assessment to assess which of the two would provide a more realistic assumption, is out of this paper's scope.We reproduce the analyses assuming ambivalence/lack of opinion and find no results that would lead us to a vastly different conclusion (see Appendix 4).
To further contextualize our findings, we also compared populist attitudes to four other constructs that we would expect to vary in their stability (see, for example, Palmer and Duch 2001; Prior 2010): political interest, satisfaction with the government's work, retrospective sociotropic economic perceptions, and the respondents' positions on whether they prefer climate change protection measures over economic growth or vice-versa. 7We deliberately chose simple reference constructs for which many readers have an intuitive sense of stability.The caveat of these choices was that all the reference concepts are unidimensional, whereas populism at the individual level is multi-dimensional.The stability of populist attitudes may thus be sensitive to, for example, intertemporal variation on one of its dimensions, the choice of dimensions combined into an aggregate measure, or how we aggregate dimensions into an overall score.In other words, while the aggregate score can still accurately reflect the concept's overall stability, multi-dimensional concepts, by nature, have more potential sources of (in)stability.To this end, we conduct additional analyses, including analyzing stability at the level of populism's individual dimensions, using different combinations of dimensions, and using different aggregating mechanisms.

Results
Figure 1 shows that the sample average level of populist attitudes hardly changed throughout the period of observation.This finding is in line with work referencing little change in aggregate distributions across countries as support for assuming stability of populist attitudes (see, for example, Ardag et al. 2020).Yet, aggregate-level investigations may overshadow individual-level dynamics under the surface.
Figure 2 shows that the strength of the correlations decreases with time.The correlation of populist attitudes in the first survey wave with populist attitudes a few months later is 0.71.The correlation between the first and last survey wave decreases to 0.59 for populist attitudes.The positive correlation coefficient implies that, on average, people with higher levels of populist attitudes compared to other respondents in Panel Wave 1 also exhibit higher levels of populist 6 A cognitive pre-test of at least one of the items used here conducted in Germany suggests that respondents use the middle category as both a stand-in for 'don't know' (akin to the ambivalence/lack of opinion assumption) and as a nuanced opinion (Lenzner et al. 2016).

7
Political interest was measured during all six panel waves for which we have populist attitude measures.The other concepts were measured in Waves 1, 3, 4, 5, and 6.Political interest and retrospective sociotropic economic perceptions were measured on five-point Likert scales.Satisfaction with government performance was measured on an 11-point scale, which we recoded into a five-point scale.We retained the original seven-point scale for the climate change protection measure variable.The number of valid scores across all panel waves for the concepts are as follows: political interest (6,231), satisfaction with government performance (6,426), economic perceptions (6,381), and climate change protection (6,324).attitudes compared to other respondents in Panel Wave 6.Yet, as is also the case for the reference concepts, the binding power erodes over time, so today's differences in attitudes will have more in common with the inter-individual differences in populist attitudes next year than in the next two or three years.
Decomposing the fluctuation in populist attitudes in Fig. 2, we can say that approximately 29 per cent of the variance (1-0.71) is state-like in that it is highly unstable.Based on the distance between zero and the smallest correlation of 0.59, we also can say that 59 per cent of the variance is trait-like because it remains stable across the observed time period.The remaining 12 per cent include autoregressive changes over time, as inferred from the difference between the largest and smallest correlations measured across the nearly four years covered by the panel data.If populist attitudes fully resembled a trait-like construct, the correlations would be one or close to one over time.In short, populist attitudes are neither fixed dispositions nor erratic ad-hoc constructions but lie in the middle of a continuum between the two.
Political interest is the most stable of the concepts we selected for comparison, given that the trait component is roughly 80 per cent.This finding is consistent with the literature showing that political interest is relatively stable over time (Devine and Valgarðsson 2023;Prior 2010).By contrast, the position on the climate versus economic growth debate has the largest state component.Economic evaluations show the largest autoregressive trait component that captures partly stable variance (38 per cent).Between Panel Waves 1 and 4, there is a significant drop in correlations for economic evaluations.This decline coincides with the start of the COVID-19 pandemic, which likely explains the change in the autoregressive component.Lastly, the state-like component of government satisfaction is slightly smaller than the state-like component of populist attitudes (23 per cent compared to 29 per cent), but the trait component is almost equally large.What can we learn from comparing populist attitudes with these reference concepts?Populist attitudes are not as stable as political interests.However, they are slightly more stable than positions on climate protection and economic evaluations.Again, this evidence situates populist attitudes somewhere in the middle of the state-trait continuum.
Next, we move on to intra-individual over-time variations.In this analysis, we dichotomized populist attitudes and studied the occurrence of changes that involved crossing opinion boundaries.To ensure that the observed changes and subsequent classification of respondents were not the result of measurement error, we conducted the test proposed by Hill and Kriesi (2001a).The test reveals that the difference in changes across the two groupsstable and unstable populist attitude holdersvaries.This would not be the case if all observed changes resulted from randomly distributed measurement errors (See Appendix 3 for detailed results).
The results reinforce the general conclusion that populism at the individual level is neither wildly fluctuating nor fully stable.Across the six survey panel waves, 67 per cent of respondents switched their views on populism at least once, but only a handful of respondents changed their attitudes towards populism every survey wave (Table 2).Comparing populism with other concepts, populist attitudes fall between two extremes, exemplified by political interest and economic perceptions (Fig. 3).Whereas only 39 per cent of respondents provided inconsistent responses to the political interest item, the most stable of the selected comparison concepts here, 85 per cent of the respondents, crossed an opinion boundary on the economic perception item at least once.
Whereas the preceding analysis explored whether respondents changed their minds, our final test examined the magnitude of these changes (Table 3).It corroborates the previous findings.The only difference is that populist attitudes are the second most stable construct.However, across the three analyses, populist attitudes, satisfaction with government, and climate change all fall between the two extremes, political interest and economic perceptions, with only the magnitude and order changing.In our view, these minor differences do not take away from the central conclusion but merely highlight the advantage of using different approaches to stability, all of which come with their strengths and weaknesses (c.f.Freeder, Lenz, and Turney 2019).Note: We rescaled the climate change measure to range from 1 to 5 (original: 1 to 7).The populist attitude score is the minimum score across the three dimensions of populism.For satisfaction with government, economic perceptions, and climate change, the average absolute distance in the column 'W3-W4' refers to the absolute distance between Panel Waves 2 and 4, as there is no measure of either of the three concepts available for Panel Wave 3.
Considering the peculiarities and the multiple options for operationalizing populist attitudes (Wuttke, Schimpf, and Schoen 2020), we explored whether the results were driven by our choices of populist dimensions and the operationalization of populism as the minimum score of its constituent components.This may be a problem when most respondents score lowest on the same dimension, as in our data with a homogeneity dimension.We conducted three additional analyses.First, we analyzed each populist dimension separately.Second, we used multiplication as an alternative aggregating function.Third, we used a two-dimensional index that combines the most common dimensions, anti-elitism and people sovereignty (Marcos-Marne, de Zúñiga H, and Borah 2023) (Appendix 5).The results show that the homogeneity dimension is the least stable of the three dimensions, though all three dimensions show some instability.Based on correlations across time, we find that approximately 61 per cent of the variation in the homogeneity dimension is trait-like compared to 67 per cent in the anti-elitism dimension and 69 per cent in the sovereignty dimension.Regarding intra-individual stability, about 76 per cent of panel respondents held unstable attitudes toward the ideas of homogeneity compared to 61 per cent for sovereignty and 52 per cent for anti-elitism.Relaxing the non-compensatory rule and using multiplication to aggregate the dimensions or exclude homogeneity from the scale can result in a more stable construct.Yet, we also show that this is not a given; the outcomes can depend on the empirical distributions of the individual dimensions.Regardless, our central finding persists: populist attitudes are neither a fully stable trait nor wildly fluctuating.However, the extent to which the pendulum swings in one or the other direction is sensitive to the individual dimensions, including their empirical distributions.Displaying some change of mind is common; citizens may have good reasons to change their minds now and then (Kiley and Vaisey 2020).However, there can be consequences to disregarding the heterogeneity in stability for empirical questions about populist attitudes.

Empirical Implications
Does it matter whether populist attitudes are stable or not?To illustrate this point, we conducted two simple analyses of individual differences in the stability of populist attitudes.First, given the theoretical reasons for a negative association between populist and democratic attitudes (Wuttke, Schimpf, and Schoen 2023), we correlated the populist attitudes measured in Wave 5 with three items on attitudes towards liberal democracy. 8Fig. 4 shows that the relationship between populist attitudes and the endorsement of democratic principles differs in all three cases, depending on whether we examine citizens with stable or unstable populist attitudes (Appendix 6).
For the second test, we linked populist attitudes with (populist) party support, as is common in the literature (van Hauwaert and van Kessel 2018), based on the idea that only stable attitudes have the binding power to steer behaviour in specific directions.We focus on the vote for the Alternative for Germany (AfD), a party commonly described as populist (Arzheimer and Berning 2019).We analyzed whether populist attitudes played a more substantial role in shaping voting behaviour in the 2017 German federal elections among respondents with stable populist attitudes than respondents with unstable populist attitudes, focusing on those who constantly agree (1 per cent of our respondents) and constantly disagree (21 per cent of our respondents) with populist ideas.98 All three items were measured in Wave 6. Item wording (authors' own translation): 'Every party should have the chance to become part of the government' and 'Everyone should have the right to stand up for their opinion even if the majority has a different opinion'.A vibrant democracy is unthinkable without an opposition.Higher values indicate higher agreement with the statement.
The main takeaway is that respondents with stable populist attitudes were more likely to act following their populist orientations (Table 4).Thirty-one per cent of the respondents who agreed with populist ideas in the panel wave before the election, but who held unstable populist attitudes across all panel waves considered here, reported having voted for the AfD with their list vote.Among stable supporters of populist ideas, 54 per cent of the respondents reported having voted for the AfD.The share of AfD vote was just 7 per cent among those with stable antipopulist views.Among those who disagreed with populist ideas before the election but expressed more positive views towards populism at another time, 14 per cent voted for the AfD.We find similar results when comparing average absolute changes over time by populist attitude levels and self-reported votes or when using the switch count to capture stability (see Appendix 6).This analysis supports the theory that stable attitudes can influence political behaviour more than unstable ones.Thus, not taking stability seriously may lead to underappreciation of effects, depending on the distribution of stable and unstable populist attitude holders in the relevant population.

Discussion and Conclusion
Although it is essential to understand the nature and implications of populism at the individual level, the stability of populist ideas has scarcely been studied empirically.Instead, the literature  often posits one of two implicit assumptions.The predisposition view holds that populist attitudes are stable over time and may thus impact other concepts.In contrast, proponents of malleability conceive of populist attitudes as responsive to short-term stimuli rather than guiding politics.Assessing how stable populist attitudes are may help to contribute to our understanding of which role populist attitudes can play in political attitudes and behaviour.
Our analyses show that populist attitudes are not entirely erratic or simply reflect the current environment.Over more than three years, inter-individual differences in populist attitudes predicts future differences at r = 0.59.This aggregate-level metric masks tremendous variation between individuals.Analyzed as a categorical attitude, about less than a third of the respondents in our German sample held steady populist attitudes over multiple years: 21 per cent consistently disagreed with populist attitudes, 10 per cent consistently neither agreed nor disagreed, and 1 per cent consistently agreed.Most respondents were ambivalent or changed their opinion at least three of the five times when they were asked about populism.For these respondents, populist attitudes appear not to be a particularly stable and central construct.Hence, for some people, populism is a (permanent) way to view the world, whereas, for many others, the current level of populist attitudes depends on idiosyncratic or situational factors.
Against this backdrop, it is unsurprising that populism matters most (for other attitudes and behaviour) when it is stable and reflects a meaningful attitude.Our analysis suggests that fluidity in populist attitudes goes hand in hand with weaker associations between populist attitudes, support for regime principles, and voting for a populist party.The strong association between populist attitudes and relevant outcomes in the high-stability group is in line with the notion that populist attitudes serve as central political predispositions in that subgroup.The weaker association in the low-stability section of the citizenry fits with the view that populist attitudes do not play a strong role as independent forces in people's political belief systems.Overall, it does not seem fruitful to think about the stability of populist attitudes in either-or terms.Instead, both views about the stability of populist attitudes voiced in previous research appear to contribute to our understanding of populist attitudes and their implications as they apply to different subgroups.
Some qualifications are in order.Using panel data allowed us to study the stability of populist attitudes over time, both at the variable and individual levels.However, our panel participants differed because they were more involved and interested in politics than a random sample of the general population (Gärtner, Gavras, and Schoen 2020).Given the sample specificity, panel attrition, and panel condition, unbiased estimates would likely result in more respondents holding unstable attitudes towards populism than estimated here.This adds to the conclusion that populist attitudes function as a dominant worldview is limited to a minority of people, at least in our case.Since our study was limited to a single country, future studies will also show whether these findings travel to other contexts.In general, we would expect to find that populist attitudes in different contexts also fall on a continuum between trait and wildly fluctuating, leaning slightly towards the latter.For example, examining the relationship between economic perceptions and populist attitudes in Spain, Rico and Anduiza (2019) find that the latter is less stable than the former regarding inter-individual stability.Further, we might expect that more individuals hold stable populist ideas in countries where populist ideas are more salient in the political discourse.Future research must empirically assess the levels of stability, as populations and timing matter (Kiley 2021).However, as a priority, it seems fair to assume that populist attitudes fall somewhere in between some of the more stable and unstable concepts examined.
Previous research that examined the links between populist attitudes and personality traits (Fatke 2019;Kenny and Bizumic 2023;Vasilopoulos and Jost 2020) suggests that populist attitudes, to some degree, can have the same underlying causes that determine a concept's stability level.However, the question of who holds stable populist attitudes does not fall within the scope of this paper.Nevertheless, explorative analyses (Appendix 7) suggest that the usual suspects, such as attitude extremity or higher levels of formal education (Howe and Krosnick 2017;Xu et al. 2020;Zaller 1992), contribute to the stability of populist attitudes.Coupled with additional evidence, these findings might inform studies taking place in shorter time windows than ours, using, for example, two-wave panels or cross-section data.These do not allow identifying stability by looking at changes over time, but using markers such as attitude extremity can help identify individual differences.For example, it could be that experimental stimuli to change populist attitudes have smaller effects on those on the extreme ends of the populism scale, and are more likely to move respondents hovering around the mid-point.These initial results and future research along the same lines can inform empirical research regarding treatment heterogeneity.This consideration reiterates our main point that we should take a more nuanced perspective that does not assume populist attitudes to be fully stable, flexible, or homogeneous across individuals.
Altogether, populist attitudes have become an important concept and a potential puzzle piece to explain the ebbs and flows of populist success.However, without considering the dynamics of these attitudes, most conclusions we draw about the role of populist attitudes as the consequence or driver of attitudes and behaviour remain limited.This paper seeks to set the stage for a debate integrating a more nuanced view of the dynamics.Populist attitudes are neither predispositional nor are they fully flexible.Instead, they fall in between.Multiple groups in any given population differ in their level of attitudinal stability, which affects how they react to environmental stimuli and how much they draw on populist attitudes when making political decisions.
Supplementary material.The supplementary material for this article can be found at https://doi.org/10.1017/S0007123423000492.

Figure 1 .
Figure 1.Distribution of populism scores and means among the German internet population between August 2017 and March 2021 shows little changes in aggregate distribution over time.

Figure 2 .
Figure 2. Correlations between measures taken at the initial survey panel wave and subsequent waves.

Figure 3 .
Figure 3. Share of respondents with unstable populist attitudes over time in comparison to selected concepts.

Figure 4 .
Figure 4. Correlations between populist attitudes and items on liberal democracy by stability.

Table 1 .
Items to measure populist attitudes

Table 2 .
Share of respondents by the number of times they cross populist attitude opinion boundaries across the six-panel waves

Table 3 .
Absolute changes in attitudes and average absolute within-respondent change between panel waves and across pairs of panel waves (standard deviation in parentheses)

Table 4 .
Reported vote for AfD by agreement with populist ideas and populist intra-individual stability