Bringing Emotions into the Study of Responsiveness: The Case of Protective Policies

2.1 A Brief Recap on the Insights from the Policy Responsiveness Literature

‘Democratic responsiveness’ is what occurs when the democratic process induces the government to form and implement policies that the citizens want. When the process induces such policies consistently, we consider democracy to be of higher quality. Indeed, responsiveness in this sense is one of the justifications for democracy itself. (Powell 2004, 91)

Democracy as a linkage between preferences of the citizens and policies – as summarized in the quote by Powell above – has been at the centre of an important research agenda studying the quality of democracy. Empirically, responsiveness has often been seen as the core of “substantive representation” (Pitkin 1967) and studied as policy responsiveness, which occurs when a government “adopts policies that are signalled as preferred by citizens” (Manin et al 1999, 9).^[4] To empirically test this relationship between public opinion and policy adoption (for an overview, see Burstein 2003, 2010; Wlezien & Soroka 2021), studies have analysed (1) “congruence”, that is whether ideologies or public preferences on certain issues as measured in surveys overlap with policy outputs (Jones & Baumgartner 2004), as well as (2) “responsiveness” which occurs when politicians react (respond) to opinion changes during the policymaking process by adjusting policies in the direction preferred by citizens (Beyer & Hänni 2018, Stimson et al 1995). While studies on congruence often use aggregate measures of positions, such as left-right-scales (Golder & Stramski 2010), an important body of work interested in policy responsiveness emphasizes the need to differentiate between specific policies and issues when studying responsiveness through policy adoption (Rasmussen et al 2019). Scholars in the tradition of the thermostatic model have, for instance, looked into how changes in public opinion on certain policies (such as foreign aid, education, welfare or defence) are reflected in changes in public spending, and vice versa (Wlezien 1995, 2017, Soroka & Wlezien 2010). Similarly, studies based on the analysis of party manifestos have looked for correlations between public opinion, issue emphasis of parties in manifestos, and policy outputs – trying to model the entire democratic linkage from public opinion, political parties and policies (Dalton et al 2011).^[5] And as part of the “Comparative Agendas Project”, an international team of political scientists has developed and applied issue-specific codes that enable them to link the content of legislative bills, parliamentary speeches and public opinion in order to see to what extent the public agenda is linked to the political agenda (in parliament) and, eventually, policy outputs (legislation) (Baumgartner et al 2019). The core of the data collection is the detailed coding of issue-specific policy domains, which can then be used to match the dynamics in the public and the political realm.

Finally, an important mostly US-based body of research has followed Stimson’s work on public opinion in Stimson (1991), which argues that a promising way to capture citizens’ preferences is to focus on the policy mood – a general concept that can be related to different policy issues such as economic, cultural or crime-related issues. Stimson describes policy mood as a “shared feelings that move over time and circumstances” but links it to issues, arguing that “publics see every public issue through general dispositions” (Stimson 1991, 20). Hence, scholars from this tradition argue (and empirically find) some sort of “underlying preference for government action” (Wlezien & Soroka 2021), which can be aggregated based on different policy issues (Stimson 2012). Indeed, research finds clear indications of thermostatic responsiveness based linking this policy mood to government activity (Caughey & Warshaw 2017; Erikson et al. 2002; Stimson et al. 1995), although policy preferences do not always move together over time (Wlezien 2004).

Admittedly, this line of research has also not remained unchallenged. Critics have argued that most citizens are politically uninformed, do not care about what is written about certain policy issues in party manifestos and have not much knowledge about which policies have been decided (Jensen & Zohlnhöfer 2020). However, empirical analyses indicate that new information about policy is used even by less political sophisticated citizens to update their political views (Enns & Kellstedt 2008). Moreover, political communication researchers have pointed out that taking policy preferences as starting point of the chain of responsiveness is a heroic assumption, as preferences themselves are malleable and have shown to be influenced by political communication such as framing or priming (Druckman 2014). And similarly, theoretical work on representation has emphasized recursive processes that model the circularity between preference formation and policies (and vice versa) (Mansbridge 2018, Disch, van de Sande & Urbinati 2020). To address this issue, more elaborated perspectives such as the thermostatic model do indeed account for such endogeneity and propose statistical analyses using time lags model policy responsiveness (opinion-to-policy) and public responsiveness (policy-to-opinion) in an integrated framework (Soroka & Wlezien 2010, Esaiasson & Wlezien 2017).

Hence, the upshot of this rich literature is that – holding other important influences on policy-making constant – (1) policy outputs mostly do follow changes in public opinion and that (2) the extent of this responsiveness depends on context such as political institutions or the salience of the issue at stake (Wlezien & Soroka 2012).^[6] In most of these studies, though, opinion-policy linkages are modelled as issue- or (area-)specific, differentiating, for instance, between criminal justice, social welfare or foreign policy (see for instance the key monographs by Soroka and Wlezien (2010) or Baumgartner and colleagues (Baumgartner & Jones 1993, Baumgartner et al 2019)), or by a more general aggregation of different issues into a more general indicator of “policy mood” (Stimson et al. 1995).

2.2 Beyond Policy Areas and Issue Linkages

As we have seen, although differences in degree can be observed when it comes to how empirical studies analyse policy responsiveness – e.g. whether they use some aggregation of issues in policy areas (“foreign domains”, “domestic domains”), analyse linkages between explicit and precise issues and policy outputs (“higher taxes”) or aggregate issue preferences up into an indicator of “policy mood” – the empirical backbone of most of the studies remains the general idea of more or less specific issue-related linkages between preferences or opinions on the one and public policies on the other hand. While policy preferences clearly are an important part of democratic representation, some scholars have argued that we may overlook parts of the story when we generally take preferences as starting point. Protective policies are a case in point. In fact, we know from political psychology that while feelings of insecurity produce psychological processes resulting in an increased demand for protection by the state, it very much depends on framing which policies are actually considered as fulfilling this need of protection (Albertson & Gadarian 2015). A good example for such a cross-cutting relationship is for instance the finding that governments use welfare chauvinist social policies to respond mainly to issues of cultural insecurity (i.e. threats to the traditional way of life) and only secondarily to economic and social insecurity (van der Waal et al. 2010). Hence, responsiveness in this case seems to be less preference-related than one going back to subjective feelings that citizens seek to be taken care of by politics.

Theoretically, Albertson and Gadarian argue that the proposed mechanism behind insecurity and threats on the one and the demand for protection on the other hand is linked to emotions, and mainly anxiety (Albertson & Gadarian 2015, 19–26). This general expectation can be traced back to psychological mechanisms from terror management theory (Pyszczynski et al 2015) and insights from research on political conservatism (Jost et al. 2007). And indeed, using experimental evidence on cases such as a (smallpox) epidemic, terrorist attacks, climate change and migration, the study finds evidence that protective policies are indeed supported when people are anxious and fearful – depending on whether a threat is “out there” (i.e. unframed) or framed by political actors. Similarly, researchers working on fear of crime have not only shown that fear is correlated with more punitive attitudes (Singer et al. 2019), but also that people living in generous welfare states are less fearful of crime, probably because the “higher the capacity of the welfare state to cushion social and economic risks, the less existential insecurity should be left to be projected onto crime” (Hummelsheim et al. 2011, 331). Hence, social as well as penal policies seem to be correlated with fear of crime – a finding which is also reflected by the association between economic vulnerability and fear of crime (Vieno et al 2013). Hence, if such affective states generate demand for (protective) policies, research on responsiveness may be missing parts of the story by mainly considering preferences as the starting point.^[7]

Summarizing these insights from the literature, it seems therefore straightforward that studies of responsiveness would also benefit from reaching beyond linking preferences and policies and include feelings, emotions and affect^[8] – especially in cases linked to protection when emotional responsiveness through policy measures in different domains can be expected (see Figure 1). More concretely, I therefore expect that emotions of individuals will be systematically related to policies, in the sense that negative affective states and anxiety will go hand in hand with a demand for more protection (H1). In addition, reflecting the possibility of issue-unspecific responsiveness, I expect that negative affective states and anxiety will lead to an increased demand for protection by the state across different policy domains (H2).^[9] Evidently, this does not mean that preferences are irrelevant – they clearly are. I simply argue that responsiveness research needs to be open to the fact that, sometimes, policies are adopted to respond to emotions of citizens, such as anxiety, which in turn can be the result of rather different policy preferences. Interestingly, this idea of subjective feelings, affect and emotions as key ingredients of responsiveness is not too far away from the idea of policy mood in Stimson’s 1991 book on public opinion in America, where he describes it as connoting “shared feelings that move over time and circumstance” (Stimson 1991, 20). However, whereas Stimson – and many other researchers following his work – have looked at preferences as basic entity of responsiveness (be it issue-specific or generalized), I argue that it may actually be worthwhile to look at emotions and affect directly and unrelated to policy preferences.

Figure 1:

Summarizing the theoretical argument.

3.1 Data and Methods

I explore these relationships using wave 3 and wave 6 of the European Social Survey. I chose these survey waves, because they both include a short version of the widely used PANAS affect-scale, which measures positive and negative affect by asking respondents how they felt during the last weeks (see below for an overview of the data). In order to test the hypotheses above, I chose negative affective states, and anxiety in particular, as key independent variables in my analyses. For the aggregate affective state variable, I follow the re-analysis of the ESS-scales by Sortheix and Weber (2023). They find that five out of six standard negative affect items load on the same dimension and exclude “you felt that everything you did was an effort” from the final measurement model (Sortheix & Weber 2023, 5). The negative affect scale is therefore made up of five indicators. I use the mean of the respective variables, which is the usual procedure in research using these affect scales (Breyer & Bluemke 2016). In addition, I use the variable of “felt anxiety” as an alternative second independent variable, as it taps in specifically on the emotion which is theoretically most relevant to the demand for protection (see the discussion above). As dependent variables, I explore two different policies that related to protection. The first policy is related to economic insecurities and social welfare, namely policies aiming at redistribution of income. In contrast, the second policy is linked much more to cultural insecurities and related anxieties and proposes a harsher stance toward migration. Here, I use two different measures of those dimensions of protection.^[10] Table 1 summarizes the variables and respective measurements.

To analyse the data, I present descriptive statistics and results of regression analyses. The regressions also include a number of confounders, namely socio-demographics and variables related to economic status (see Table 1). Although the dependent variables are, strictly speaking, ordinally scaled (4 point or 10 point likert scales), I present results from OLS regressions for ease of interpretation. Estimations from ordinal logit models do not deviate substantially from the OLS results. As I am mainly interested in the associations on the individual level independently from cross-country variance or differences over time, I estimate a two-way fixed effects model that soaks up all variance between countries and waves. I estimate robust standard errors.

3.2 Empirical Results

To get a sense of the data and the Figure 2 presents the three dependent variables linked to protective policies as country-wise arithmetic means for both rounds of the ESS analysed here (ESS round 3 and 6). While it shows considerable cross-country variance in all three cases, the graph also indicates that the differences between nation-states seem to be a bit more pronounced for attitudes toward migration, that is policies that can be interpreted as protecting from cultural insecurity. Although the aggregated means do not show the variance on the individual level within countries that we are most interested in, the graph nevertheless already indicates that citizens’ demands of protection by the state do vary considerably in Europe.

Figure 2:

Attitudes toward protective policies in 30 European countries.

Moving on to bivariate associations on the individual level, Table 2 displays simple bivariate correlations between our dependent variables, that is how strongly respondents in the respective countries demand protection from the state in the case of economic insecurity and cultural insecurity on the one hand, and the emotional states that the respective persons feel when asked in the survey. Although correlation coefficients are not overwhelming in size, the pattern of association is clear-cut: Negative affect in general and anxiety in particular are positively correlated with demands for protection on both policies analysed here, namely policies directed at economic insecurity and cultural insecurity. This is at least a first indication that emotional states are linked to a general demand for protection, independent of the actual policy domain.

Importantly, the regression analysis confirms the results of the bivariate correlational pattern. The models presented in Table 3 are structured by dependent variable and consist of three models per dependent variable, which include, first, the negative affect scale and, second, the specific item that taps into feelings of anxiety. From the table transpires that even when controlling for confounders that may be related to both the dependent variable and the independent variables, such as the left-right position, income, age, education level and gender, I still find a pattern that points to an association between negative affect on the one and individual demands for protective policies on the other hand. The signs of the coefficients are similar to what I found in bivariate correlation, indicating that people in a negative affective state and those feeling more anxious are in favour of more protection (more redistribution, less migration). However, it is important to note that the associations are more significant and robust in the case of protection of cultural insecurity than for policies aimed at protection of economic insecurity. Hence, anxious people seem to be slightly more inclined to ask for protection of a country’s culture and their way of life than for reducing income differences. Still, for anxiety, the coefficient meets the lowest threshold of significance even for economic insecurity (Model 2 in Table 3).

Turning to the control variables it is also important to note that especially high economic insecurity and low education (as well as high age) are significantly (and highly) correlated to both protective policies (i.e. all three dependent variables) – a pattern in line with research on welfare chauvinism (Careja & Harris 2022). In contrast, gender and attitudes on the general left-right scale differ between protective policies related to economic and cultural insecurity. This finding calls for more research into the relevance of economic hardship on anxieties that may feed into demand for protection – relationships that are difficult to uncover with the limited cross-sectional data at hand.

In sum, these results indicate that emotions and affect seem to be associated on the individual level, with people experiencing negative emotions being more inclined to support policies that provide more protection. Hence, the first hypothesis formulated above is supported by the data. Moreover, with regard to our second hypothesis, the regression estimations also point to the fact that feelings of anxiety push people to demand protection in quite different policy domains – such as economic policies and migration policies (confirming H2). While it is true that these results do not provide direct evidence for the idea of emotional responsiveness in the sense that politicians respond to affective states with different policies (or discourses), the regression estimates can be read at least as a first indication suggesting that different protective policies may indeed respond to one and the same affective states. This points – on a conceptual level – to the possibility that responsiveness does not only relate to policy preferences, but to affective states and emotions.