Explaining Populist Attitudes: The Impact of Policy Discontent and Representation

Recently, scholars of populism have increasingly started to theorise and capture susceptibility to populism at the individual level. Most of these studies, however, focus on the consequences of populist attitudes on political behaviour. Less attention has been paid to the question of which citizens have high levels of populist attitudes and why. While some scholars argue that populist attitudes more resemble an unchangeable personality trait, meaning that individuals may be more or less populist, others argue that it is a response to outside grievances or discontent. The latter suggests that levels of populist attitudes are dynamic and may change if grievances are addressed (or remain unaddressed). We contribute to this literature by asking how discontent fuelled by unfulfilled policy preferences affects the level of populist attitudes. Following the conception of populism as a thin-centred ideology, we argue that high levels of populist attitudes are not connected with certain issues per se. Rather, our argument is that people are more populist when they feel poorly represented on policy issues that they care strongly about. This argument provides an explanation for the observation that even voters of non-populist parties sometimes show high levels of populist attitudes. We test the impact of policy discontent on populist attitudes using data from the GLES 2021 Pre-Election Cross Section survey by combining information on citizens’ issue specific discontent with the perceived salience of respective issues. The results are in line with our expectations: Individuals with higher policy discontent are more populist.


Introduction
Populism is on everybody's minds and one of the most researched phenomena in recent years. Particularly since the Brexit referendum in June 2016 and Donald Trump's election in November 2016, societal and scientific interest in populism rose tremendously (Rooduijn 2019). While a broad corpus of academic literature is devoted to the analysis of populism among parties and political elites (e.g . Hawkins 2010;Mudde 2007;Van Kessel 2015;Zulianello 2020), there is less systematic research on populist attitudes among citizens. This is surprising given that, according to the ideational approach (Hawkins et al. 2019;Rovira Kaltwasser 2013, 2017;Mudde 2017), populism "exists as a set of widespread attitudes among ordinary citizens" (Hawkins et al. 2019, iii).
In contrast to other conceptualisations of populism, the ideational approach defines populism as a specific set of ideas and attitudes focusing "on the juxtaposition between a people-centred notion of political representation and the corrupt political elite" (Geurkink et al. 2020, 248). Based on this framework, populism can be measured at various levels, such as individuals, parties, or elites. Thereby, it has fuelled a vital strand of academic research that operationalises populist attitudes at the individual level and uses these measures in empirical research (e.g. Akkerman et al. 2014;Castanho Silva et al. 2018;Schulz et al. 2017).
Following this research, the ideational approach provides a both theoretical foundation and methodological toolkit for analysing how populist attitudes affect other attitudes or political behaviour, such as voting for populist parties (Akkerman et al. 2014; Van Hauwaert and Van Kessel 2018;Steiner and Landwehr 2018). Potential explanations of citizens' level of populist attitudes, however, have received substantially less attention (for a summary see Wuttke et al. 2020). This article contributes to the analysis of individual-level causes of populist attitudes and is thus interested in the question of why some citizens are more populist than others.
The literature thus far provides two sets of arguments about the roots of populist attitudes, which differ in their assumptions about the persistence and contextual nature of populist beliefs. The first argument states that populist attitudes are widespread among citizens and comparable to an inherent personality trait that lies "dormant until activated by weak democratic governance and policy failure" (Hawkins et al. 2019, iii). According to this view, populist attitudes are not inherently tied to a particular ideology or certain issues. Instead, they are conceptualised as a response of individuals that is activated once they feel badly governed and not sufficiently represented by political elites (Hawkins et al. 2012(Hawkins et al. , 2020. Populist attitudes are thus understood as a static characteristic of each individual that may be more or less pronounced and activated by contextual deficits. In other words, the level of populist attitudes remains rather constant and is activated by a (perceived) lack of representation.
The second argument holds that populist attitudes arise as a manifestation of specific grievances (Rico and Anduiza 2019) or discontent (Giebler et al. 2020;Rooduijn et al. 2016). 1 Given the strong relationship between populism and certain 'thick' ideological positions at the party level (Rooduijn and Akkerman 2017;Meijers and Zaslove 2020), proponents of this perspective assume that populist attitudes at the individual level are also strongly connected to certain host ideologies. For example, it seems reasonable to suspect that people with strong anti-immigrant attitudes are also supportive of populist parties ( Van Hauwaert and Van Kessel 2018), partly driven by the strong supply of such positions at the party level. In this view, populist attitudes are dynamic and may change if grievances are addressed (or remain unaddressed).
Following through on these conceptualisations of populist attitudes, we provide an integrated perspective on why some citizens are more populist than others. We combine the two aforementioned approaches by assuming that a) populist attitudes stem from a lack of (policy) representation and resulting discontent, but b) that populist attitudes are dynamic in that they can be mitigated if these policy concerns are addressed. Specifically, we argue that populist attitudes are induced by a combination of unfulfilled policy preferences in highly salient issue areas and are thus potentially caused by any issue and not just based on the usual suspect of immigration policy. This perspective is consistent with the conceptualisation of populism as a 'thin-centred ideology' (Freeden 1998;Mudde 2004) which can be combined with any host ideology. Moreover, the argument puts particular emphasis on the role of issue salience, highlighting that citizens have to care about an issue in order to feel poorly represented by political elites. In this regard, our argument follows the idea of, for example, Bélanger and Meguid (2008), that issue salience is an important moderator of how policy positions affect other political attitudes and political behaviour.
In sum, we expect that individuals are more populist when they strongly care about a certain issue but do not feel that their position is well represented in current politics. For right-wing respondents this is often the case with migration. Claims that immigration has been misgoverned by the Merkel government have been a core theme for the AfD. But for other respondents, it might be the economy or climate protection, two issues that are also frequently discussed in public debates. If people care about one or several of these issues, but feel that their voices are not being heard, they might consider that the political elite is just following their own will and moving away from the 'ordinary people'. Our argument, therefore, provides a potential explanation for why high levels of populist attitudes are often observed even among voters of non-populist parties.
To test these expectations, we rely on the GLES 2021 Pre-Election Cross Section survey data ZA770; GLES (2022). Specifically, we use citizens' evaluation of government performance in three potential issue areas: migration, economy and climate change. We combine this issue-specific discontent with the perceived salience of the respective issues. The empirical analysis demonstrates that policy discontent is indeed an important predictor for populist attitudes. Accordingly, individuals with higher policy discontent tend to also be more populist. The effect of policy discontent is substantial and has a strong impact on respondents' level of populist attitudes comparable to education or satisfaction with democracy. Moreover, we show that the effect of policy discontent is independent of vote choice, which helps to explain intra-party variation in populist attitudes and, thus, why even some voters of nonpopulist parties hold populist beliefs.
This manuscript is structured as follows: We start by discussing and reviewing the current literature on explanations of populist attitudes. We then make a theoretical argument by providing an integrated perspective on how issue discontent, issue salience and populist attitudes are related. In the subsequent section, we describe our research design and empirical strategy. We then report the results of our preregistered analysis as well as additional robustness tests and further analyses. Finally, we conclude by discussing our findings and suggesting endeavours for future research.

Literature review: What explains populist attitudes?
The rise of populist parties caused an explosion of research about the roots of populism. While most studies focused primarily on the supply side and the emergence of populist parties in response to a series of economic, cultural, and political crises (Kriesi and Pappas 2015;Van Kessel 2015), the question about the drivers of populist attitudes at the individual level has only recently received scholarly attention. Thus, the focus has started to shift from the supply side of parties to the demand side of citizens. The methodological backbone of this research is the development of scales for measuring levels of populist attitudes at the individual level that conceptualise populism as a thin-centred ideology or a set of ideas (Akkerman et al. 2014;Castanho Silva et al. 2018;Schulz et al. 2017). But precisely because research on populist attitudes emerged only recently, we know relatively little about their dependence on temporal and contextual factors. Moreover, it remains controversial whether and to what extent populist beliefs occur independently of ideological attitudes and policy positions.
Both the argument that populist attitudes are independent of particular policy positions and its context dependence were discussed at the outset by Hawkins and colleagues (Hawkins et al. 2012(Hawkins et al. , 2020. According to their understanding, populist attitudes are dormant in individuals and are thus comparable to an inherent personality trait that is activated by a given political context. This populist predisposition is most likely to be activated in a context "in which there are widespread failures of democratic governance that can be attributed to intentional elite behaviour" (Hawkins et al. 2020, 286). From this perspective, certain crises, such as an economic recession, may increase the likelihood that populist attitudes are activated, but they alone K are not sufficient to evoke the fundamental populist ideas of a corrupt elite or establishment blamed for the lack of popular accountability (Rico et al. 2017;Rico and Anduiza 2019). Rather, it is a belief in systematic flaws within the political regime that leads to alienation from politics and elected politicians as representatives of the 'true will' of the people. Populist attitudes are thus understood as a rather static characteristic of each individual, which may be more or less pronounced, and is activated by contextual deficits (Busby et al. 2019;Hawkins et al. 2020).
Another strand in the literature renders populist attitudes as a manifestation of specific grievances (Rico and Anduiza 2019) or discontent (Giebler et al. 2020;Rooduijn et al. 2016). Given that populism at the party level often tends to emerge at the margins of the political spectrum, proponents of this approach assume that populist attitudes tend to be held by more leftist and rightist individuals and are thus strongly connected to certain 'thick' ideologies. Populist attitudes are thus attributed to radical policy positions on various issues that result in dissatisfaction with political representation because these more radical positions are not represented in (government) policy. In this view, populist attitudes are dynamic and may change if grievances are addressed (or remain unaddressed). This view on the link between specific policies and populist attitudes is represented in a variety of different strands in the literature.
Kriesi and colleagues (Kriesi et al. 2006(Kriesi et al. , 2008(Kriesi et al. , 2012, in a set of influential studies, argue that globalisation has substantial implications for the political landscape by creating distributional conflicts in both economic and cultural terms. As globalisation processes create winners and losers, national political parties use this new conflict line to mobilise voters. Unsurprisingly, evidence suggest that populist parties "perform" crises (Moffitt 2015) and that both economic and cultural conflicts associated with globalisation have been used by populists in Europe (Kriesi and Pappas 2015;Habersack and Wegscheider 2021). Therefore, we could anticipate that citizens react to these processes, and as a consequence turn more populist due to an increase in discontent.
Similarly, Eatwell and Goodwin's (2018) controversial argument focuses on the absolute and relative economic deprivation related hardship, which may be utilised by populist actors. Specifically, they focus on four central concepts: i) distrust which fuels individuals' (national) populism; ii) destruction and iii) dealignment of personal and local identities in the process of globalisation, and finally, iv) deprivation, which is emphasised by inequality and the perception of 'neoliberal' economics. Hence, individuals who feel increasingly removed from their identities through accelerating processes of (economic) globalisation may be more susceptible to a national vision of populism. Norris and Inglehart (2018) do not focus on "national populism" but rather explain why there is a cultural backlash against globalisation (also see Schäfer 2021). According to their argument, as societies turn more liberal, citizens with strongly conservative and traditional-in essence, illiberal-stances perceive a lack of substantive representation in certain (mostly cultural) policy areas. For example, legislation liberalising same-sex marriage is in strong contrast to these individuals' worldviews. As a reaction to this perceived lack of representation, they turn to populist parties as they offer an authoritarian populist alternative that promises to navigate societies back to their heartland, a golden age of traditional values.
Finally, other authors have argued that populist parties use climate change to foster their anti-establishment positions and criticise established elites (Otteni and Weisskircher 2021). Following this argument, populist parties may use climate change to mobilise their voters (Vihma et al. 2021). Hence, we would anticipate that populist voters (who on average tend to be more climate sceptic, see, e.g. Huber 2020) may actively criticise governmental climate policy. At the same time, evidence suggests that some progressive left-wing populists may criticise governments for not doing enough on climate change (Huber et al. 2021).
Despite this evidence, a general problem applies to the literature discussing the impact of grievances on populist attitudes. While most of these studies argue that grievances influence populist attitudes, it is not entirely clear whether the causal relationship could also be reversed. In other words, whether developing stronger populist attitudes might result in having stronger grievances on certain issues. While some recent studies demonstrate that a perceived lack of representation does seem to influence the level of populist attitudes (Castanho Silva and Wratil 2021), it is beyond the scope of this article to provide an empirical test of the direction of the causal relationship, and we ask readers to keep this limitation in mind. Future research might address this question in more detail based on comprehensive timeseries data.

Theoretical argument: Discontent salience
The theoretical approaches described above share that they introduce a dynamic perspective on how citizens perceive representation and which stance they take towards politics in response to certain forms of discontent. Citizens are not populist per se, but rather their levels of populist attitudes are a reaction towards political developments and a feeling of not being represented by the political elite. Thus, these studies share a common implicit framework. Our core contribution is to make this relationship explicit. Individuals turn to populist parties or become more populist if they perceive a salient issue to be poorly represented and governed. Existing evidence suggests that this may work for cultural (Eatwell and Goodwin 2018; Norris and Inglehart 2018) as well as economic issues (Kriesi and Pappas 2015; Rico and Anduiza 2019), for globalisation-induced issues such as climate change (Lockwood 2018), and perceptions of democracy in general (Heinisch and Wegscheider 2020). Hence, we argue that the underlying argument of why people turn populist is a combination of issue representation and issue salience. Basically, we argue that dissatisfaction with democratic representation on a salient policy issue should induce populist attitudes and that, therefore, citizens who are dissatisfied with the current representation on a particular issue should be more populist. However, this effect should be more pronounced, potentially even visible only among those who perceive the misrepresented issue as salient. In the form of an equation: K In which P A j are an individual j 's populist attitudes, D ij captures the discontent of individual j on issue i and S ij represents the salience of issue i for individual j .
In light of this argument, it is essential to emphasise that populist attitudes are not a function of specific issue positions per se. Although, empirically, individuals who oppose immigration may often be fairly populist, this merely reflects-according to our argument-that these citizens care strongly about this issue and, at the same time, perceive the government to be poorly representing their preferences on immigration. In other words, they would potentially not be as populist if a substantially more immigration-sceptic government would be representing their policy positions. Likewise, individuals with strong anti-immigration sentiments are not expected to be more populist (even if they feel poorly represented) if the issue is not important to them. Moreover, our argument is not limited to the issue of immigration, but applies equally to other dimensions of the policy space, such as social welfare and redistribution or climate change. Based on the mechanism described above, we expect individuals to become more populist when they are increasingly dissatisfied with the representation on an issue that they deeply care about. From this discussion and the integrated perspective provided above, we derive the following hypothesis: Individuals who are dissatisfied with the government's work on salient issues hold stronger populist attitudes than those who hold no discontent or are only dissatisfied with issues that are not important to them.

Data
We test these theoretical expectations using the GLES 2021 Pre-Election Cross Section survey data (ZA7700; GLES 2022). The dataset and its documentation are accessible via the GESIS data archive after one-time registration. 2 This survey contains information of 5,116 interviews with citizens in Germany. The population captures all German citizens above 16 currently residing in private households in Germany. Respondents are selected following a multi-stage register sample, in which municipalities are selected first and then respondents are randomly selected from these municipalities. Responses are collected through either computer-assisted webbased interviews (CAWI) or paper-and-pencil interviews (PAPI).
We exclude missing data list-wise. Based on previous GLES pre-election surveys, we expect missing data to occur quite rarely. As described in detail in Appendix B, we run analyses testing whether missing values for the main variables in our analyses (populist attitudes and policy discontent) are predicted by other relevant variables (political ideology, political interest, satisfaction with democracy, education, region, age and gender). However, contrary to our expectation in previous versions of this manuscript, missing values do not occur completely at random (see Table B1 in the Appendix). When we regress the occurrence of missing values in the variables capturing populist attitudes (i.e. our dependent variable) and policy discontent (i.e. our main independent variable) on a set of control variables, we find the probability of item non-response for both variables to be slightly higher among women. Furthermore, we find that individuals are less likely to respond to one of the items necessary to calculate our policy discontent variable if they are more left-wing, less interested in politics, or more dissatisfied with democracy in general. However, as described in detail in the Appendix, the influence as well as the total number of missing values is rather small, so we do not consider this to be a major problem for the analysis and, therefore, do not impute missing values. It does imply, however, that our analysis is less representative, as the sample is mildly skewed towards male respondents and respondents who are more right-wing, politically interested and more satisfied with democracy.
We use two types of sampling weights to ensure robustness of the results. First, we use a design weight that corrects the oversampling for East Germany. Second, we use an adjustment weight that adapts the sample to socio-demographic (gender, age, education) and regional characteristics (regions, East-West) according to marginal distributions of the Microcensus 2019. We report the results of both weight types along with the unweighted results.

Dependent variable: Populist attitudes
Following our theoretical argument, the dependent variable is a respondent's levels of populist attitudes. We aggregate six survey items, which are designed to measure populist attitudes. The GLES includes the CSES Module 5 populist attitudes scale (Hobolt et al. 2016), which is loosely based on the scale suggested by Akkerman et al. (2014). It is designed to capture the two core elements of populism (antielitism and people-centrism) in a single dimension. Table 1 contains the exact item wording. All items are measured on five-point Likert scales.
How populist attitudes scores should be aggregated has been subject to recent research (Wuttke et al. 2020). These studies call into question whether using the mean or similar approaches for constructing a single score for multi-dimensional concepts are appropriate, arguing that some concepts are non-compensatory, i.e. lower/higher values on a certain dimension cannot be compensated by higher/lower values on other dimensions. In order to construct a non-compensatory index, however, a scale has to cover at least two dimensions, which is questionable for the items we are using to measure populist attitudes. Thus, our primary analysis is based on measuring populist attitudes based on confirmatory factor analysis where all items load on a single factor -as intended by the scale. To calculate the factor scores, we use the lavaan package (version 0.6-8) within the software R (version 4.0.5) and use the weighted least squares estimation with robust standard errors and mean-and-variance adjusted test statistics (WLSMV). In Table 1, we report the chi-square ( 2 ) and robust estimates of RMSEA, CFI and SRMR to evaluate the model fit according to established cut-off criteria (Kline 2005;Hooper et al. 2008 Items for the question 'Please say how much you agree or disagree with each of these statement'. Responses were originally captured on a scale from 1 (strongly agree) to 5 (strongly disagree), but we recoded the items from 0 (strongly disagree) to 4 (strongly agree) so that higher values indicate higher agreement. Loadings show completely standardised estimates of the confirmatory factor analysis. robustness of our results, we also use the unweighted mean as well as the minimum value of all items as additional aggregation methods. 3 Figure 1 shows the distribution of populist attitudes by vote choice. 4 The first distribution displays the populist attitudes across all respondents. The subsequent distributions are ordered by the median value of populist attitudes grouped by the respondents' vote choice. The distribution suggests high face validity, as AfD voters have the highest populist attitudes on average, while Green voters tend to be the least populist. At the same time, there are considerable intra-party differences in the levels of populist attitudes, which allows us to analyse variation for voters of all parties.

Independent variable: Policy discontent
Theoretically, we scrutinise how citizens' discontent with specific issues and their respective issue salience relate to populist attitudes. To this end, we need to capture two central variables: issue discontent and issue salience. Starting with the former, we utilise a measure capturing respondents' satisfaction with the German govern-3 The correlation between the populist attitudes measure estimated with the confirmatory factor analysis and the measure using the unweighted mean is 0.993, and with the measure using the minimum value of all items is 0.821. The correlation between the measure using the unweighted mean and the measure using the minimum value of all items is 0.810. 4 Visualising the distribution of populist attitudes based on vote choice was not explicitly pre-registered. We capture vote choice through two variables. We combine information on voting intention for the second vote in the upcoming election (q8ba) with information about vote choice for the second vote for those who have already cast their vote by postal voting (q10ba). We code respondents by their voting intention or mail vote for the SPD, CDU/CSU, Green Party, FDP, AfD and Left Party, as well as all other parties, and set all other values as missing (including errors in data, not specified, multiple choices, invalid choices and don't knows). Populist Attitudes

Vote Choice
Abb. 1 Distribution of populist attitudes. Note: Populist attitudes are estimated factor scores based on confirmatory factor analysis (see Table 1 for details). Electorates ordered by their parties' median score on populist attitudes ment's work on three specific issues. 5 To capture discontent with the policy areas of climate change and migration, we use the items on 'combating climate change' (q56a) and 'controlling immigration to Germany' (q56b), respectively. For economic issues, we use 'the promotion of economic growth' (q56e). 6 In addition, we use three variables to measure the respective issue salience by making use of questions that ask respondents about their perceived importance of specific policy dimensions (q41, q44, and q49). The specific wording of these questions is 'How important is the topic of [taxes and social services/immigration of foreigners/combating climate change] to you?' on a scale from 1 (very important) 5 Specifically, the question wording is 'When you think about the major issues that have dominated the political discussions in Germany recently. How satisfied are you with the work of the German government on ...?' and capture positions on an 11-point scale from -5 (not at all satisfied) to +5 (completely satisfied). We reverse the scale from 0 to 10 so that higher values indicate higher issue discontent. 6 The second economy-related item is 'the modernisation of infrastructures' (q56d). We use this item only as a robustness check, because we believe it measures quite different preferences than the economic growth item. In other words, we consider the economic growth item to be a better measure for economic preferences than the infrastructure item. However, we also acknowledge that measuring economic preferences and satisfaction is more challenging because the economic performance can be measured with regard to various different dimensions. In contrast, immigration and climate change policies are potentially more uni-dimensional measures.
K to 5 (not important at all). We reverse the order of the scale for the analyses so that higher values indicate a higher issue salience. Moreover, for reasons explained in more detail below, we recode the values to a scale from 0 to 4 instead of 1 to 5. While the items on climate change and immigration perfectly mirror the question on satisfaction with the government's work, the item for the economy does not overlap perfectly. We are aware that this might create a slight disconnect. However, we deem losing the economic dimension altogether as more problematic than the slight inconsistency between the salience and discontent measures.
Using these three main variables (policy discontent with the government) and three moderating variables (issue salience) can create problems in the empirical analysis, as it can be expected that all variables correlate with each other. 7 To reduce the model dependence of our analysis, but also to better reflect our theoretical argument, we use the described variables to create a single score which simultaneously measures a respondent's level of issue discontent weighted by the respective issue's salience. To compute this score, we first create a relative issue salience score, denoted by S, which measures for each respondent i the relative importance given to an issue j . Specifically: This also justifies recoding the issue importance variables to a range from zero to four, because assigning the weight of zero for a non-important dimension has conceptual advantages. 8 For example, a respondent who strongly cares about the climate change issue (s i;climate D 4) but not at all about migration and economic issues (s i;economy D s i;migration D 0) will have a relative weight of 1 for climate change, but 0 for the other two issues. A respondent who cares equally about all three issues, will have a weight of 0.33 for each issue.
To create the final score, we multiply each issue-specific discontent d i;j with the respective relative importance weight and create an overall score, D i , of policy discontent: It is important to mention that this variable has a deliberately chosen compensatory nature. Higher values of satisfaction with the government in combination with higher levels of issue importance can compensate for discontent in other issue areas. We have chosen this because we can expect that people become less populist when 7 Figure A1 in the Appendix to this paper provides a correlation matrix for the variables. The correlations between the individual measures of discontent are fairly high, whereas the salience measures do not substantially correlate. 8 On a scale from one to five, the lowest value of relative issue importance would be 1/15, while with our approach it is 0/12. Thus, if a respondent assigns no importance at all to an issue, it would be removed from the equation in the latter case but not the former.

Vote Choice
Abb. 2 Distribution of policy discontent. Note: Policy discontent is based on Eq. 3. Electorates ordered by their parties' median score on policy discontent they feel represented in at least one issue area. This also mitigates concern that discontent in the three dimensions may not be independent and thus highly correlated, creating problems of multi-collinearity. 9 Figure 2 shows the distribution of the policy discontent measure by vote choice. 10 First, we note that policy discontent is fairly normally distributed across the sample. Second, and similar to Fig. 1, we find that AfD voters on average show the highest level of policy discontent, while voters of the two governing parties CDU/CSU and SPD show the lowest policy discontent on average. At the same time, we again find significant differences within the electorates of each party. 9 Table C5 in the Appendix to this paper presents an analysis of the disaggregated interactions of issuespecific discontent and salience. Note that we use the absolute salience of an issue, since it is impossible to include all three relative salience measures. Running the analysis for each dimension individually risks an omitted variable bias, as the discontent and salience of dimensions correlate (see Figure A1 in the Appendix). The findings suggest that issue-specific discontent is the driver of populist attitudes, whereas salience plays a subordinate role. The interactions of both dimensions fail to reach conventional levels of statistical significance. Additionally, the findings suggest that migration-related, and to some extent economic discontent, are related to populist attitudes, whereas the estimate for climate change-related issue discontent is positive (as expected) but not statistically significant. 10 This visualisiation based on vote choice was not pre-registered.

Control variables
We include several additional control variables. Specifically, we adjust our estimation for two sets of variables. First, we include core socio-demographic variables. Thus, we add age (d2a), gender (d1), education (d7) and whether a respondent lives in the east or west of Germany (based on available meta-data). Second, we add core political variables. Specifically, we include left-right self-placement (q37), political interest (q1) and satisfaction with democracy (q4). The Appendix includes all variables used in the questionnaire's order and additional descriptive statistics (see Section A in the Appendix).

Analysis plan
We run OLS regression analyses with populist attitudes as the dependent variable. The main independent and control variables are described in the sections above. Education 11 , gender and East Germany are included as dummy variables. Ordinal variables are treated as if continuous. Robust standard errors are used. Our preregistration does not specify a certain expected effect size. However, we expected that the coefficient for policy-specific discontent (ˇ) is positive and 'statistically significant' at the p < 0.05 level. Additionally, we also report exact p-values to avoid arbitrary classification of effects as either 'significant' or 'insignificant' based on the p < 0.05 threshold.

Results
Our models include 4,089 respondents due to missing values in one of the included variables. We report the main results of the OLS regression analysis in Table 2. The table displays four models. Models 1-3 follow the pre-analysis plan by including all mentioned variables, and they only differ with regard to the sampling weights used. Model 1 does not use any weights, Model 2 adjusts for the over-representation of East German citizens in the sample, and Model 3 adjusts for more general socio-economic and regional imbalances. Model 4 replicates Model 3 but includes a squared term for left-right self-placement. This model specification was not pre-registered but reflecting on studies that describe the relationship between left-right ideology and populist attitudes as U-shaped, we considered it reasonable to additionally test for this pattern in the model (Rooduijn and Akkerman 2017;Meijers and Zaslove 2020).
The most relevant for our hypothesis is the effect of the Policy Discontent variable, which we expect to have a positive effect on populist attitudes. This is exactly what we find. A one-point increase in this variable leads to a 0.07 increase in the level of populist attitudes among respondents. The p-values in all model specifications are below 0.01. 12 Taken together, these findings provide strong support 11 We recoded the variable for education into a dummy variable so that we only distinguish between individuals with and without Abitur. 12 The exact p-values are around 0.005. for our expectation that policy discontent does indeed drive populist attitudes. The finding is also not affected by the inclusion of survey weights, as all estimates remain almost identical.
These findings can be put into perspective by comparing the effect strength of the policy discontent variable with other variables. To do so, we estimate 'first differences' between the third and first quartiles for all included variables. This means that we predict values of populist attitudes when the independent variable has the value of its third quartile and subtract the predicted populist attitudes value when the independent variable is at its first quartile. These predictions follow the approach by King et al. (2000), and we repeat it for each independent variable 5,000 times. Based on these 5,000 simulations, we estimate 95% confidence intervals. The approach allows us to estimate comparable effect strengths by comparing the change in the predicted levels of populist attitudes when the independent variable changes by the inter-quartile range. It should be noted that for dummy variables, such as gender or east-west, the displayed effect is the change from 0 to 1, as this is the p < 0.001; p < 0.01; p < 0.05. The dependent variable is populist attitudes based on a confirmatory factor analysis (see Table 1 for details). Entries are unstandardised coefficients based on ordinary least square regression. Robust standard errors are reported in parentheses.  Table 2 (N D 4,089). Estimations are based on the approach developed in King et al. (2000), and we use 5,000 simulations for estimating the 95% confidence intervals (horizontal lines). Points are medians of the 5,000 simulations only way the variable can change. In this case, the effect size is identical to the one displayed in Model 3 in Table 2. The results are displayed in Fig. 3.
Based on this comparison, policy discontent has quite a large effect size. For example, the effect of policy discontent is stronger compared to that of political ideology. However, this comparison is skewed due to the non-linear effect of left-right ideology, which is evident from Model 4 and displayed in Figure C1 in the Appendix. The effect of policy discontent is also larger compared to political interest, age and gender. Despite East Germany being discussed as a region with particularly high levels of populist attitudes (Pesthy et al. 2021), we find no impact of the East Germany dummy variable when controlling for socio-demographic and regional characteristics. There are only two variables with an effect size relatively larger than policy discontent. First, satisfaction with democracy shows a negative effect of -0.23, indicating that democratically satisfied respondents are substantially less populist. Second, high levels of education also have a strong negative effect on populist attitudes. However, given these results, it seems reasonable to conclude that policy discontent is an important predictor for populist attitudes.

Robustness analyses
In order to scrutinise the robustness of our findings, we conduct three tests in which we alter the operationalisation of core predictors or the outcome variable. Specifically, we substitute our latent measure of populist attitudes with a unweighted mean across all six items as well as an Goertz-based approach using the minimum value across all six items (Wuttke et al. 2020). Additionally, we capture economyrelated policy positions using a second policy item focusing on public investment in infrastructure (see also Footnote 6). Figure 4 summarises the results (also see Tables C1, C2 and C3 in the Appendix for additional information). The black crossed ranges display the main models and are identical with the information displayed in Fig. 3. The ranges with blue triangles and green dots provide the estimate for our alternative measure for the independent variable, specifically the unweighted mean (blue) and minimum value (green). The alternative measure for our independent variable is displayed in orange with squares. The result remains substantially the same, indicating that our findings are robust to alternative model specifications.

Non-registered analysis: The role of vote choice
In this section, we further examine the robustness and variation in the relationship between policy discontent and populist attitudes. The aforementioned results could reflect the supply of certain positions by specific political parties. To rule out the possibility that the results and the effect of policy discontent on populist attitudes are driven by a particular group of voters, we examine below whether the effect differs by vote choice. Invariance in slopes between different parties indicates that the findings are more likely to be driven by the demand from citizens regardless of parties' specific supply. As already shown in Figs. 1 and 2, both populist attitudes and policy discontent vary significantly between voters of different parties. Please note that the following analyses are not pre-registered. In addition to the main models in Table 2, we include an interaction between our policy discontent measure and vote choice. The results of this interaction are visualised in Fig. 5. We find the positive effect of policy discontent on populist attitudes across all voters of different parties. Thus, while we find that AfD voters are more populist overall than other voters, policy discontent explains the variation within different groups of voters. Accordingly, even among voters with the lowest average levels of populist attitudes, such as Green party voters, the individuals most likely to be populist are those with high levels of policy discontent. These results thus help to explain why even some supporters of non-populist parties also hold populist beliefs. If voters, regardless of whether the party is populist or not, are dissatisfied with policies in an area they deeply care about, they are more likely to have pronounced populist attitudes.  Table C4 in the Appendix

Conclusion
Research on populism is increasingly interested in the question of which factors contribute to populist beliefs at the individual level. Drawing on the ideational approach towards populism (Hawkins et al. 2019;Rovira Kaltwasser 2013, 2017;Mudde 2017), previous studies mainly focused on two arguments. While the first argument states that populist attitudes are comparable to an inherent personality trait that is activated when people feel poorly governed and represented (Hawkins et al. 2012(Hawkins et al. , 2020, the second argument focuses on populist attitudes as an expression of specific grievances that can change when those grievances are addressed (Giebler et al. 2020;Rico and Anduiza 2019;Rooduijn et al. 2016).
In this paper, we contribute to this strand of research by providing an integrated perspective and combine the two aforementioned approaches. Accordingly, we assume that a) populist attitudes are based on a (perceived) lack of policy representation and resulting discontent, but b) that populist attitudes might also change if these policies are addressed. Moreover, we assume that people are more populist when they perceive unfulfilled policy preferences on highly salient issues they deeply care about. Thus, individuals that are strongly dissatisfied with the government performance on salient issues have higher levels of populist attitudes.
Based on a pre-registered analysis plan using the GLES 2021 Pre-Election Cross Section survey data, we combine information on citizens' issue-specific discontent with the perceived salience of respective issues. Running multiple regression models and robustness tests with our combined measure of policy discontent, we find strong support for our assumption. Accordingly, people who are very dissatisfied with the political performance on a particular issue that is particularly important to them are more likely to have higher populist attitudes. Furthermore, we find that this effect is independent of vote choice, which explains why even some voters of non-populist parties sometimes show high populist attitudes.
These results have some important theoretical implications. Our findings support recent studies on populist attitudes being the result of a (perceived) lack of political representation (Castanho Silva and Wratil 2021). Accordingly, populist attitudes could be mitigated, at least in part, by ensuring that people feel better represented and that their policy positions are reflected in the government's actions. Our results thus speak at least partly against the first argument above, that populist attitudes are a rather static personality trait. Instead, our findings support the idea that populist attitudes change as a result of a shift in political representation.
Moreover, our results support the argument that populist attitudes among citizens cannot be mitigated by the participation of the favoured party in government alone. Like other studies before, we find some difference in the average level of populist attitudes between voters of government and opposition parties, yet there is a fair amount of variation within these electorates that can be explained, at least in part, by policy discontent. This means that even individuals whose favoured party is in government may have high populist attitudes because, for example, they are dissatisfied with the government's performance on a particular issue that is important to them. Overall, our measure of policy discontent thus offers a good way to explain why populist attitudes might change over time and the variation of populist attitudes between and within electorates.
Of course, these results should be interpreted with a certain degree of caution. A possible limitation of our study is that the causal direction is not entirely clear. Assuming the opposite causal direction, i.e. populist attitudes lead to higher levels of policy discontent, is also plausible. Therefore, more advanced research designs, especially the use of panel data, seem to be particularly useful to explore this question in more detail. Especially in the case of a change of government and/or a change of policy positions within governments, it would be interesting to analyse whether and to what extent these events affect citizens' policy discontent and populist attitudes. This would also help to better disentangle the causal mechanism.
A second limitation of this paper is the combination of various types of issue discontent and issue salience into a single measurement. While this score was intentionally created to provide a salience-weighted measure of issue discontent, it is also true that a deeper understanding of which issues are particularly strongly associated with populism is also relevant. While populist attitudes may be partly mitigated by better representation of certain issues, radical ideologies seem particularly connected with populist beliefs and, thus, deserve scholarly attention. Especially regarding more fine-grained measures of populist attitudes, the question arises as to what extent all sub-dimensions (i.e. anti-elitism, people-centrism, Manichean worldview) can be addressed through better representation, and whether other sub-dimensions are more due to a fundamental worldview that goes hand in hand with certain radical ideologies. Future research should therefore continue to attempt to examine in particular the ideological breeding ground of populist beliefs.
Notwithstanding these limitations, our paper offers a new perspective to explain populist attitudes. In particular at the individual level, populist attitudes seem to be prevalent among a wide range of respondents. Populist attitudes do not accumulate exclusively among certain respondents, such as the voters of populist parties. At the individual level, populist attitudes appear to be more cross-cutting than, for example, 'thick' ideologies. Therefore, it seems necessary to investigate this widespread variation in populist attitudes among citizens in more detail. Our theoretical argument and measurement of policy discontent complements recent studies addressing this topic, and future research can build on this insight for an even more comprehensive explanation of populist attitudes.
Funding Open Access funding enabled and organized by Projekt DEAL.