Going beyond values versus self-interest: whose attitudes change after employment transitions?

Are self-interest or presumably stable value orientations and other predispositions the main drivers behind social policy attitudes? This article contributes to this debate by moving away from its binary discussion. It differentiates between attitude changes driven by self-interest that are in line with pre-existing predispositions and those that are not. Empirically, this article focuses on changes of labour market policy attitudes after employment transitions and job insecurity changes. More precisely, this article differentiates between attitude changes within three subgroups. (A) People whose self-interest after the employment transitions reinforces their prior predispositions. (B) People without strong prior predispositions, who are thus unconstrained by them. And (C) people whose self-interest after the employment transitions contradicts their prior predispositions. Panel analyses with fixed effects use German SOEP waves from 1997 and 2002. Main effects suggest an important role for selfinterest as they show significant attitudinal reactions after most of the transitions and perception changes. However, subgroup analyses result in a somewhat mixed picture. They show attitude changes within different subgroups after different transitions and perception changes. This mixed empirical picture suggests caution when interpreting attitudinal change or stability after changing material circumstances as a sign for the relative importance of self-interest or predispositions. ARTICLE HISTORY Received 2 July 2019 Accepted 6 August 2020


Introduction
Why do we want what we want or do what we do? To answer this question, a large part of the literature on political attitudes and behaviour centers on the influence of selfinterest vis-á-vis presumably stable predispositions. Examples for such predispositions are value orientations (Calzada et al. 2014;van Oorschot 2002), ideology (Jaeger 2006;Wiertz and Rodon 2019) or party identification (Lau and Heldman 2009) among others. This debate is also highly relevant for social policy attitudes (Owens and Pedulla 2014) including this article's empirical focus on labour market policy (LMP) attitudes (Margalit 2013). To be more precise, I focus on attitudes towards job creation policies (active LMPs) and unemployment benefits helping the unemployed financially (passive LMPs).
This article focuses on the question whether LMP attitudes change after employment transitions and changes of perceived job insecurity. Self-interest based arguments expect growing support for LMPs after job loss and other subjective or objective deteriorations of the individual employment situation. Arguments emphasizing the role of values, in contrast, are typically associated with the expectation of attitudinal stability. To move away from this binary 'self-interest versus values' discussion, this article contributes by differentiating between attitude changes that are consistent with pre-existing value orientations and other predispositions and those that are not. Thus, this article does not only infer the relation between self-interest and value orientations and other predispositions. Instead, this article observes their relation more directly.
By doing so, this article answers recent calls to better understand 'how prior leanings condition the response to the [economic] shock' (Margalit 2019, p. 290). More precisely, this article differentiates between three subgroups of people, who might experience self-interest driven attitude changes after employment transitions and job insecurity changes. These are (A) people whose predispositions already agree with those LMPs that serve their self-interest after these employment changes, (B) people who do not hold strong prior predispositions and thus are not constrained by these prior predispositions and (C) people whose prior predispositions contradict their later self-interest after these employment changes.
Empirically, I use German panel data from the Socio-Economic Panel (SOEP) from 1997 and 2002. The SOEP has not yet been used for analyses of dynamics between employment and LMP attitudes. Its advantage is that it allows for the observation of a sufficient number of employment transitions and job insecurity changes even when splitting the sample according to measures of prior predispositions.
When focusing on the whole sample most transitions and changes of perceived job insecurity are followed by attitudinal responses consistent with self-interest accounts. However, the subgroup analyses show a mixed picture. While a few attitudinal responses counteract pre-existing predispositions, many actually do not, as they reinforce pre-existing predispositions or happen in absence of strong predispositions.
In the following, I address accounts of welfare attitudes and specifically of LMP attitudes that focus on self-interest and accounts that focus on value orientations and other predispositions. After delineating the arguments for attitudinal responses within the three subgroups I discuss the SOEP data and variables and the panel fixed effects models used for the analyses. After that, the results will be discussed including a descriptive overview as well as several robustness checks. The article concludes with a summary and lessons for future research. The mixed findings caution against inferring the relation between selfinterest and predispositions from (in)significant main effects of changing material circumstances on attitudes. Furthermore, they caution against generalizing attitudinal reactions to one kind of changing material circumstances to reactions to other kinds of material circumstances.

Values versus self-interest and beyond
The basic assumption of self-interest based arguments regarding policy attitudes is that people support a policy because they benefit from it. This includes people who (1) objectively benefit from that policy now, (2) are objectively expected to benefit from that policy in the future, and (3) people who subjectively believe they will benefit from that policy in the future. These three versions of the self-interest argument are typically applied to the relation between social policy attitudes and income (Alesina and La Ferrara 2005) or unemployment. In the latter case, this application includes (1) those who are currently unemployed, (2) those who are in an objectively insecure employment situation like fixedterm employment (Marx 2015) or employment in an occupation with a high unemployment rate (Rehm 2016), and (3) those who subjectively worry about their job security for whatever 1 reason (Burgoon and Dekker 2010). All three groups of labour market 'outsiders' (Rueda 2005;Schwander and Häusermann 2013) are thus expected to demand government action for financial help during unemployment (passive LMPs) and measures to reduce unemployment (active 2 LMPs) among other things.
The rival argument par excellence comes from the 'values perspective', as it is referred to in the literature on attitudes towards social policies, redistribution, and the welfare state. Here, social policy attitudes are seen as determined by egalitarian value orientations, justice principles like equality, need or equity, and other value orientations not so closely linked with redistribution (Andreß and Heien 2001;Calzada et al. 2014 (Jaeger 2006;Pardos-Prado 2011). And to give some empirical examples from Germany with relation to LMP attitudes, research shows that spending less on LMPs is more acceptable for people who generally believe in everyone's own responsibility for their economic fate. Similarly, such cuts are less acceptable for people who believe the needy should to be supported by the state (Heinemann, Bischoff, and Hennighausen 2009). Also, identifiers with left parties are shown to support state responsibility for different social policies more than identifiers with conservative parties (Dageförde 2013).
Whether focusing on value orientations, ideology or party identification, all these different kinds of predispositions share several functions and characteristics especially in comparison to policy attitudes. Most importantly, they are assumed to be far more stable over time than policy attitudes and they are assumed to causally influence policy attitudes, not vice versa (Maio et al. 2003). Whether these assumptions are true is of course an empirical question and part of a longstanding debate. But these assumptions have certainly received too much empirical credit to ignore them (Carsey and Layman 2006;Weber and Saris 2014). Also, different predispositions can of course impact each other, as well, as e.g. Goren (2005) shows with regard to core values and party identification.
The self-interest perspective does not necessarily contradict such predispositional determinants of policy attitudes. Indeed, theoretical arguments relying on a self-interest logic often discuss the 'values perspective' before choosing to derive hypotheses from the self-interest mechanism (Häusermann, Kurer, and Schwander 2015;Rehm 2016).
Thus, if self-interest measures and value orientations were independent of each other the values versus self-interest controversy could be solved by simply acknowledging the existence of two different sets of predictors for political attitudes. However, they are not independent. Take the examples of attitudinal gaps based on age (Sears et al. 1980), formal education (Sears and Funk 1990) or employment related variables like subjective job insecurity (O'Grady 2019) or unemployment (Wehl 2019). On the one hand, these can be interpreted as a result of diverging interests of the young and old, the highly and low educated, or of labour market insiders and outsiders. On the other hand, however, they can be interpreted as a result of different socialization experiences of these groups in different school contexts, different political contexts (cohort socialization) or in more or less affluent parental households (O'Grady 2019; Sears et al. 1980;Sears and Funk 1990;Wehl 2019). Should we thus stop interpreting any socio-economic or demographic gaps in attitudes as evidence for self-interest (Chong, Citrin, and Conley 2001)? Or should we rather start to interpret all predispositional effects on attitudes as additional evidence for self-interest (Weeden and Kurzban 2017)? For some it basically becomes a matter of taste.
In contrast, a clear observational difference between self-interest based and value-based arguments emerges in a dynamic context where people move into unemployment, fixedterm employment, etc. and others move out of these insecure situations. The straightforward self-interest expectation is that people will support LMPs more strongly after their employment situation deteriorates. Similarly, their support for LMPs will decrease after their employment situation improves (Margalit 2013;O'Grady 2019). In contrast, the 'values perspective' emphasizes predispositional, and thus attitudinal, stability. Value orientations and other predispositions acquired during socialization are seen as rather stable afterwards in adulthood (Chong, Citrin, and Conley 2001;Sears et al. 1980). This focus on attitudinal change or stability after changing material situations including employment transitions is an increasingly used strategy against the self-interest versus values conundrum. The big advantage that comes with the focus on employment and attitudinal dynamics is that it eliminates potential confounding by past socialization experiences.
Studies applying such a dynamic focus to the U.S. context (Margalit 2013;Owens and Pedulla 2014) and to the Dutch context (Naumann, Buss, and Bähr 2016) find that job loss is related to higher demands for unemployment benefits, higher demands for assistance (including training) to the poor and the unemployed and higher demands for general redistributive measures. On the contrary, Stegmueller (2013) finds no evidence for relations between changes in unemployment and attitudes towards government responsibility for job creation in the UK context. Margalit additionally looks at changes of subjective job insecurity and finds evidence for higher demands for government assistance to the poor and unemployed after increasing worries about job insecurity. On the contrary, O'Grady (2019), looks at job insecurity changes in the Swiss context and finds a small increase in demands for social benefits but no increase in demands for taxing the rich. And while Margalit (2013) finds decreasing demands for assistance to the poor and unemployed after reemployment, Naumann, Buss, and Bähr (2016) find no evidence for attitudinal change after reemployment. However, they find even higher demands for unemployment benefits after temporary unemployment than after to job loss. Thus, not only the theoretical expectations from self-interest based arguments and arguments emphasizing values or other predispositions are contradicting. Empirical results are contradicting as well.
Who experiences self-interest-based attitude changes?
The theoretical and empirical ambivalence surrounding the self-interest versus values debate already led to some theoretical combinations of both. The starting point is to ask 'when self-interest matters', as Chong, Citrin, and Conley (2001) suggestnot just 'whether' self-interest matters. These authors argue that individuals' attitude changes are guided by material self-interest when their stakes are clear and high. Another example is the study by van Oorschot (2002) that asked people directly why they pay for social security and found that people are motivated by both, self-interest and value orientations.
Following these attempts to go beyond the simple dichotomy of self-interest versus value orientations, I suggest asking whose attitudes change after employment transitions and changes of job insecurity. More precisely, I suggest differentiating between three groups of people whose attitudes might change after such transitions as depicted in Figure 1. (A) People whose self-interest after the employment transitions reinforces their prior predispositions, (B) people without strong prior predispositions, who are thus unconstrained by them and (C) people whose self-interest after the employment transitions contradicts their prior predispositions. Of course, it is not possible to exclude that attitudinal reactions are observable in all three subgroups. That would be clear evidence for pure selfinterest arguments.
Why should there be attitude changes in the first subgroup, where prior predispositions already fit together with later self-interest? The literature offers at least two arguments for attitude changes after employment transitions or changes of subjective job insecurity in this subgroup. On the one hand, it is argued that self-interest affects how much someone cares about an issue, i.e. attitude importance (Boninger, Berent, and Krosnick 1995). Attitude importance in turn is related to attitude extremity (Boninger et al. 2009). On the other hand, this subgroup of people might experience attitude changes because they engage in some form of motivated reasoning (Taber and Lodge 2006). When confronted with new information (here rather experiences) people tend to incorporate information that is in line with their prior value orientations and other predispositions in their belief system as support for these predispositions and thereby strengthen these prior value orientations even more.
Several findings on the role of party identification in shaping attitudinal dynamics point to this direction. Thus, Brooks and Manza's (2013) study shows in the US context that after the great recession identifiers with the Democrats increased their demands for social policies, while identifiers with the Republicans even decreased such demands. The same is true for Peterson's (2016) finding that after winning the lottery Republican identifiers are less likely to move away from their Republican identification. Note that in this subgroup, attitude changes simply strengthen or reinforce individuals' prior predispositions. Thus, in this case, expectations from selfinterest-based arguments and arguments emphasizing the role of predispositions do not contradict each other but complement each other.
Hypothesis 1a: An attitude change in the direction predicted by self-interest arguments is observable after employment transitions and changes of perceived job insecurity for people whose predispositions already agree with those LMPs that serve their self-interest after these employment changes.
The second possibility is attitude change experienced by people who do not hold strong prior predispositions and thus are not constrained by these prior predispositions to borrow from the belief system terminology (Converse 1964). Or to put it differently, without strong values there is more room for self-interest effects.
Several varieties of this argument are well known in political science. People without strong predispositionsespecially without a partisan identityare often seen as the prototype of a good citizen, as e.g. Dalton (2012) argues when evoking the concept of cognitive mobilization. After all, they are assumed to be able to build their opinions and behave politically without reliance on their partisan identities as cues. Thus, e.g. de la Calle and Roussias (2012) show that Spanish independent voters were more likely than there counterpartspartisan identifiersto engage in retrospective voting. Similarly, Lachat (2007) shows that people with weaker predispositions are more prone to attitudinal changes. A somewhat similar argument is made in the context of motivated reasoning by Taber and Lodge (2006), who argue that weaker attitudes induce smaller biases when evaluating new information. Also, Peterson's findings focusing on self-interest-based effects on party ID are in line with this reasoning. Peterson shows that winning the lottery increases the likelihood of a Republican identification for previous non-identifiers. Note that in this case expectations from predispositional arguments and self-interest based arguments again do not contradict each other. However, in contrast to the first subgroup, this is neither a case supporting nor questioning the role of value orientations or other predispositions. After all, according to this hypothesis, the other two subgroups specified in H1a and H1c do not show attitudinal reactions precisely because of the predispositional influences.
Hypothesis 1b: An attitude change in the direction predicted by self-interest arguments is observable after employment transitions and changes of perceived job insecurity for people who do not hold strong prior predispositions.
Finally, attitude changes might also be experienced by people whose prior values and other predispositions do not agree with what serves their self-interest after the employment changes. In this subgroup, the contrast between previous value orientations or predispositions and later self-interest is sharpest. In that sense, this scenario could be seen as most likely case from a self-interest perspective (or vice versa as least likely case from predispositional accounts). Furthermore, it is quite likely that employment status changes induce experiences, that do not fit so well to the rest of people's belief system. i.e. employment status changes might induce cognitive dissonance (Festinger 1957). Attitude change in turn is a typical way to solve dissonances. Attitude reversal is in line with Margalit's (2013) result from the US context that the increase in social policy demands after experiencing an economic shock is higher for Republican than for Democratic identifiers. Note that in this case expectations from the 'values perspective' and the self-interest perspective clearly contradict each other. Effects occurring for this group of people would clearly highlight the importance of self-interest effects and question the role of pre-existing values and other predispositions in light of changing material circumstances.
Hypothesis 1c: An attitude change in the direction predicted by self-interest arguments is observable after employment transitions and changes of perceived job insecurity for people whose prior predispositions do not agree with those LMPs that serve their self-interest after these employment changes.

Data and methods
German data 4 from the SOEP (2016), the German Socio-Economic Panel (Wagner, Frick, and Schupp 2007) are analysed relying on fixed effects models. In addition to the following discussion of data and methods, detailed question wordings and recordings can be found in the appendix.

Measurement
LMP attitudes are the dependent variable in all analyses. The SOEP includes such items only in 1997 and 2002. Thus, despite the existence of yearly data from 1984 onwards, only those two waves are used for the analyses excluding respondents with missing values on the dependent variable in at least one of the waves. The SOEP items on LMP attitudes ask respondents for their position on government or state responsibility for LMPs. While government responsibility items are widely used, they are only suboptimal. Spending preference items could show an even closer link to respondents' material self-interest, especially when combined with trade-offs with cuts for other policy areas or higher taxes (Cavaillé and Trump 2015;Gallego and Marx 2017). The SOEP includes two items, one on passive LMPs and one on active LMPs (job creation), both measured on a 5-point scale. As discussed in the theory section, both kinds of LMPs can be assumed to be in the interest of unemployed persons and those with a high unemployment risk. Therefore, both LMP attitude items are averaged and jointly analysed. 5 A robustness check analyses them separately, however.
The main independent variables are employment status transitions and changing levels of job insecurity. First, these are transitions in and out of unemployment, i.e. job loss, reemployment, and temporary unemployment. Job loss and reemployment do not cover all transitions in or out of unemployment since transitions in or out of non-employment, like e.g. school-to-work transitions or retiring, are excluded. Temporary unemployment refers to persons employed in both analysed waves who went through an episode (≥1 month) of unemployment in between, which is measured using monthly employment records from the SOEP. Second, changes from a fixedterm contract to a permanent contract and vice versa are observable. These transitions between permanent and fixed-term contracts are only observable for employees and do not include transitions from or to self-employment or employment without a contract. As discussed below, due to the low number of respondents experiencing such employment transitionsespecially when splitting analyses by first wave attitudes or party ID −, job loss and switching from a permanent to a fixed-term contract are combined to 'deteriorations'. Similarly, reemployment and switching from a fixed-term to a permanent contract are combined to 'improvements'. Again, a robustness check analyses them separately.
Changing subjective perceptions of the employment situation are measured with items on job insecurity. SOEP respondents are asked on two 3-point scales how much they are concerned about their job security. The SOEP data additionally contains an item about respondents' subjective reemployment chances also measured on a 3-point scale. Each of these perception changes is converted into two binary variables indicating whether respondents worry more (or less) than in the first wave.
To differentiate between attitude changes within the three subgroups, analyses are split by respondents' first wave LMP attitudes and their first wave party identification. In the case of LMP attitudes 3 groups are built, namely a 'pro', a neutral, and a 'contra' government responsibility category. Thus, using the combined 6 measure of LMP attitudes, respondents with a score value between 1 and 2.5 are categorized as 'contra', respondents with a score values of 3 as 'neutral', and respondents with a score value between 3.5 and 5 as 'pro'. Also in the case of party identification three groups are built, namely identifiers with a 'left' party, i.e. social democratic, green, and far left identifiers, non-identifiers, and identifiers with a right party, i.e. christian-democratic, liberal, and far right identifiers. These three groups of party identifiers are built using the typical two-step survey-question about party identification asking respondents first, whether they feel close to any party, and in a second step, if so, to which one.

Analytical strategy
To analyse the dynamics of employment transitions, job insecurity changes and LMP attitudes, linear 7 fixed effects panel models (Allison 2009;Brüderl and Ludwig 2015) with cluster robust standard errors are applied. Such fixed effects estimators rely solely on intra-individual comparisons. Thus, all constant sources of heterogeneity like gender or past life course events are automatically controlled fora considerable advantage over results based on cross-sectional data (Brüderl and Ludwig 2015).
While it is not possible to include time constant variables as predictors, they can be included as moderator variables, as Hypotheses 1a-1c suggest. For most analyses, this is done by subgroup analyses with subgroups defined by respondents' first wave LMP attitudes or party attachments. However, these subgroup analyses are complemented with interaction models. These interaction models are used to test for significant differences in the effects of employment transitions and perception changes. Note that both modelling strategies can be exactly equivalent when all interactions which the subgroup analyses implicitly represent are explicitly modelled.
Importantly, results from fixed effects models are still threatened by omitted variable bias from time-varying variables. Therefore, indicators for further training, childbirth as well as changes of subjective health after the first analysis wave are added and thus factors, that might influence both, employment transitions and attitude changes. However, since the causal role of these time-varying variables is not clear and post-treatment bias is a danger their influence is considered in a robustness check only. One special case of a time-varying variable is time itself. In line with e.g. Owens and Pedulla (2014)  Predispositions, however, are observed beyond that, namely in the subgroup analyses. As discussed above, the different subgroups are based on respondents' first wave party ID and respondents' first wave LMP attitudes. Both measures have advantages and disadvantages. Party ID is one of the strongest political predispositionsespecially in comparison to LMP attitudes. The literature on motivated reasoning shows definitive evidence for that (Bisgaard 2015). However, the link between a left party ID and e.g. egalitarian values is weakened by the multidimensionality of political conflict. Non-economic issues obviously can and do play a role as well for individual predispositions like party ID or ideology (de Vries, Hakhverdian, and Lancee 2013). Therefore, subgroups are also built using respondents' first wave LMP attitudes. As discussed above, policy attitudes are seen as functionally different from predispositions. However, supporters of LMPs are likely to be general supporters of social policies due to their underlying (not measured) value orientations (Jaeger 2011). Figure 2 gives an overview of the three analysed subgroups. In the left panel we see that the overwhelming majority of respondents does not identify with a political party. In the middle panel, we see that an even bigger majority supports state responsibility for LMPs. In the right panel, we see the combination of both. As one would expect, within those who are sceptical about LMPs, there are more identifiers with parties of the right side of the political spectrum than identifiers with left parties. For those who support LMPs, it is the other way roundthere are more identifiers with left parties. However, probably the most obvious pattern is that in each of the three subgroups split by first wave LMP attitudes those without a party ID are the most frequent group.

Descriptive overview
So, on the one hand, party ID and LMP attitudes are related as one would expect. On the other hand, the overlap between support for LMPs and left party identifiers, scepticism about LMPs and right party identifiers, and being neutral about LMPs and not identifying with any party is clearly limited. Therefore, a crucial question for the following analyses is, whether the findings hold simultaneously for the subgroups split by LMP support and for the subgroups split by party ID. Figure 3 shows the employment transitions and perception changes that are observable in the SOEP data (1 st column of both figures) and their frequencies broken down by respondents' first wave LMP attitudes (2 nd column of Figure 3: sceptics '−', neutrals '+/-', and supporters '+' ) and their party identification (3 rd column of Figure 3: left 'L', no, right 'R' identifiers ).
Furthermore, the prevalence of support for LMPs, that was shown already in Figure 2, also becomes obvious when splitting the number of employment transitions and perception changes according to respondents' first wave LMP attitudes. For every employment transition and for every perception change positive first wave LMP attitudes outweigh neutral or sceptic attitudes and by far. Given those numbers statistically significant effects supporting Hypothesis 1a (1c) might be more likely to observe for employment and perception deteriorations (improvements).
Again, as before in Figure 2, the situation is very different when splitting the number of employment transitions and perception changes according to respondents' first wave party ID. Now, employment transitions and perceptions changes for people without a party ID clearly outweigh employment transitions and perception changes for party supporters. Given those numbers effects supporting Hypothesis 1b could be more likely to observe when analysing party ID subgroups. These different descriptives for the numbers of transitions and perception changes within party ID and LMP attitude subgroups again lead to an emphasis on results of the following analyses that can be observed for both kinds (LMP and party ID based) of subgroups simultaneously.
Finally, there are extremely small numbers for some of the employment transitionsespecially for sceptics of state responsibility for LMPs. Therefore, I combine job loss and the transitions from a permanent to a fixed-term contract within the category of employment deteriorations in the following analyses. Similarly, reemployment and the transition from a fixed-term to a permanent contract are combined in the category of employment improvements.

Main effects
Before looking at the effects of employment transitions and perception changes within the different subgroups and thus testing Hypotheses 1a-1c, Figure 4 (for tabular results see appendix Table A1) shows the respective main effects on LMP attitude changes. All positive subjective or objective changes (employment improvements, decreasing job insecurity, increasing reemployment chances) are accompanied by negative, significant coefficients. Thus, respondents withdraw to some degree their support for LMPs after such positive events or perception changes. Similarly, negative subjective perception changesbut not objective employment deteriorations or temporary unemploymentare accompanied by positive significant coefficients. Thus, respondents increase their support for LMPs when they worry more about their job insecurity or their reemployment chances.
The results so far firmly support the self-interest perspective and raise doubts about the relevance of arguments emphasizing the role of values or other predispositions. After all, respondents' attitudes seem to respond to employment transitions and worries exactly like the self-interest account predicts it. Respondents might nevertheless possess stable value orientations and other predispositions. These do, however, not stop respondents' attitudes from reacting to the employment changes.
One concern with such an interpretation is the question whether these coefficients actually represent causal effects and thus 'reactions'. 8 The concern I want to focus on here, however, is that -depending on who changes their attitudesthese changes do not necessarily as clearly contradict the idea behind the long-lasting impact of values and other predispositions as this interpretation suggests.  Looking at the left panel in Figure 5, we see that only those people, who already supported LMPs or adhered to a left party in the first wave, further increased their support for LMPs after employment deteriorations. Interestingly, this kind of attitude change, which simply reinforces pre-existing beliefs, occurs even though no main effect was observable. This attitude change among those who identified with left parties and supported LMPs is driven by the experience of a job loss as Figure A1 in the appendix shows. Thus, this result directly contradicts the evidence for the U.S. context presented by Margalit (2013) who showed that Republicans, but not Democrats, increased their support for social policies after a job loss. A similar increased support for LMPs among left party identifiers can be observed in the upper left panel in Figure 6 after an increased level of job insecurity. In contrast to the attitude changes after employment deteriorations, however, this effect only occurs when splitting respondents according to their party identification, not to their previous LMP support. Furthermore, left party supporters are not the only ones who increase their support for LMPs after perceiving higher levels of job insecurity.

Subgroup analyses
As can be seen in the upper left panel in Figure 6, increasing job insecurity leads to increased LMP support for those who are unconstrained by their prior predispositions, at least when judging from their previous LMP attitudes. Similarly, thoseand only thosewithout a party attachment react to employment improvements by decreasing their support for LMPs. But the strongest support for attitude changes among those unconstrained comes from attitude changes in response to less optimistically perceived reemployment chances. Here, both those without a party attachment and those with a neutral LMP attitude in the first wave increase their LMP support.
Finally, there is also evidence for attitude changes that contradict individuals' prior predispositions. In contrast to decreasing perceptions of reemployment chances, increasing perceptions of reemployment chances go hand in hand with lower support for LMPs among those who previously supported LMPs and those who previously identified with left parties. A similar drop in support among those who previously supported LMPs can be observed in response to decreasing worries about job insecurity. And a temporary episode of unemployment between the first and the second wave increased support for LMPs among those who did not support LMPs in the first wave. Figure 7 summarizes the results presented in Figures 4-6. Overall, there is some evidence for attitude changes in each of the subgroups and thus for each of the three hypotheses. However, for most of the employment transitions and perception changes there is only one subgroup for which an attitudinal reaction is observable. Thus, it is not possible to identify any constellation between pre-existing predispositions and self-interest as more likely than any other constellation to lead to self-interest based attitudinal reactions. 9 Figure 6. SOEP: results for job insecurity (FE-estimates with 90% and 95% confidence intervals).

Results summary and robustness checks
Furthermore, an interesting pattern emerges for employment improvements compared to employment deteriorations (and similar perception changes). It appears that attitude changes that simply reinforce predispositions are only present after employment deteriorations and more pessimistic perceptions. In contrast, attitude changes that challenge predispositions are present only after employment improvements and more optimistic perceptions. To be clear, this pattern was not expected for any theoretical reasons. It will thus be interesting to see, whether future research will be able to replicate it.
The presented and summarized results so far show in which subgroups significant effects of employment transitions and perception changes occur. An even harder test is, however, whether these effects differ significantly between the subgroups. After all the difference between a significant and insignificant effect might itself not be significant (Gelman and Stern 2006). Most confidence intervals shown in Figures 4-6 are overlapping. But overlapping confidence intervals are not a definitive sign regarding the (in)significance of the difference of effects (Austin and Hux 2002). Therefore, the shown subgroup analyses were replicated via interaction models including interactions between the subgroups and the employment transitions or perception changes (see Appendix Tables A8-A14). Based on these models, the differences between the average marginal effects of the transitions and perception changes between the subgroups are formally tested (see Appendix Tables  A15 and A16).
Doing so, it turns out that indeed the majority of the effect differences between subgroups are not significant with three notable exceptions. The first is the effect of employment deteriorations among left party identifiers, which is significantly bigger than the effect for people without a party identification (in ine with Hypothesis 1a). The second is the effect of increasing job insecurity among people with neutral LMP attitudes in the first wave, which is significantly bigger than for LMP supporters (in line with Hypothesis 1b). And finally, all differences based on LMP subgroups are significant for effects of a temporary unemployment episode (in line with Hypothesis 1c). Taken together, these tests thus corroborate the mixed picture of the results.
In the following, it will be tested how robust these results are, when time-varying covariates are included, when each employment transition is analysed separately (compared to the combined analysis of improvements and deteriorations) or when active and passive LMP attitudes are analysed separately.
Appendix Tables A1-A7 display the discussed results in tabular form and replicate all analyses adding controls for further education, childbirth and subjective health changes. This does not visibly alter the overwhelming majority of results. 10 The only exception are effects of employment improvements. After adding the time-varying controls it appears that both, first wave sceptics and first wave supporters of LMPs, become more sceptical about LMPs. This puts two findings to some degree into perspective. On the one hand, this is the first evidence for Hypothesis 1a after 'positive' employment or perception changes with LMP sceptics becoming even more sceptical of LMPs. On the other hand, together with the robust effect for people without a party ID after employment improvements this actually amounts to evidence supporting all three hypotheses simultaneously. This, however, is not true for both kinds (PID and LMP) of subgroups.
In a next step, appendix Figure A1 compares the results for the combined analyses of improvements and deteriorations of the employment situation with results for job loss, reemployment, and transitions between permanent and fixed-term contracts. As already mentioned above this shows that the growing support for LMPs of left party supporters and LMP supporters after employment deteriorations is actually only observable after job loss. However, effects of employment improvements are not heavily dominated by either reemployment or changes from a fixed to a permanent contract. Thus, the decreasing support for LMPs among people without a party ID after employment improvements is only observable after changing into a permanent contract. Contrarily, a new, i.e. now significant, effect becomes observable after reemployment when previous left party supporters grow more sceptical about LMPs.
And finally, I test whether attitudes towards active or passive LMPs heavily dominate the results and if a focus on the separate attitudes would lead to different conclusions regarding the hypotheses. Both concerns, however, seem unsubstantiated when looking at appendix Figures A2 and A3. First of all, separate analyses of active and passive LMP attitudes result in a similar mix of support for Hypotheses 1a-1c. And while the main effects are largely driven by changes in active LMP attitudes, subgroup results are not heavily driven by either active or passive LMP attitudes.

Conclusion
This article started with contradicting theoretical expectations and empirical findings from the self-interest and the 'values perspective': attitudinal change and attitudinal stability. Applied to the empirical context of the paper this lead to the question: do LMP attitudes change after employment transitions and changed job insecurity perceptions?
The main theoretical contribution of the article was to overcome this dichotomy from self-interest based and value-based arguments. Therefore, this article suggested to move away from empirical implications and to focus on empirical observations regarding the relation between self-interest and predispositions. To do so, I asked whose attitudes change in response to employment transitions and perception changes. To answer this question, the article differentiated between three subgroups of respondents. These subgroups were based on respondents' predispositions to support or question LMPs. More precisely, the article asked whether self-interest based attitude changes are more likely for respondents (a) with predispositions fitting that self-interest (b) without strong predispositions or (c) with predispositions contradicting that self-interest.
The answer to that question is crucial for connecting empirical results to the values versus self-interest debate. Scenario (a) would suggest that it is not 'values versus selfinterest' but rather that self-interest effects only occur when they are 'supported' by predispositional influences. Scenario (b) would suggest that self-interest effects occur only for people whose attitudes are only weakly influenced by predispositions. And scenario (c) would suggest that values and predispositions are indeed challenged by self-interest effects.
The analyses were based on German panel data from the SOEP in 1997 and 2002 and applied panel fixed effects models. Applying these analyses to the whole sample showed clear evidence in favour of self-interest based accounts of social policy attitudes. Respondents withdrew some of their support for LMPs after transitions improving respondents' employment situation and after worrying less about their employment situation. Similarly, respondents increased their support for LMPs after worrying more about their employment situation. Interestingly, this increased support could not be observed in the whole sample after transitions deteriorating respondents' employment situation like job loss.
However, the results of the subgroup analyses turned out mixed. Thus, while crucial for the interpretation of the effects, none of the three combinations between self-interest and predispositions turned out to be a condition for or against self-interest effects. More precisely, some self-interest effects were observed for people with predispositions supporting that self-interest. Other attitudinal reactions to employment transitions or changed feelings of job insecurity only affected people without strong predispositions. And yet other attitudinal reactions were only observed for people whose self-interest and whose predispositions pointed in opposing directions.
Even though the findings of the present article were mixed, they offer important lessons for future research and for interpreting existing findings. First of all, they show that evidence in favor of self-interest effects after employment changes as e.g. by Naumann, Buss, and Bähr (2016) or Owens and Pedulla (2014) is not simultaneously evidence against the role of predispositions. Similarly, failures to find main effects of unemployment or other objective or subjective employment risks on social policy attitudes as in the case of O'Grady (2019) or Wehl (2019) are not necessarily a sign for the absence of selfinterest driven effects. Additionally, the presented findings show that the interplay between prior predispositions and self-interest is not always as straightforward as the US-based results from Margalit (2013) suggest. This study found that increased social policy support after job loss is especially pronounced for Republicans, which challenges the influence of their party ID.
Second, the mixed findings also caution against generalizing findings on attitudinal reactions from one employment transition to another employment transition, perception change, or completely different changes in material circumstances. On the one hand, transitions and perception changes assumed to have similar self-interest effects turn out to result in different relations between self-interest and prior predispositions. Thus, to give an example, different subgroups adjusted their LMP attitudes to job loss and to increasing worries about job loss. On the other hand, the present article found that employment improvements lead to attitudinal changes challenging prior predispositions. On the contrary, employment and perception deteriorations lead to attitudinal reactions for people with supporting predispositions or without strong predispositions.
Finally, the question should be considered how context dependent the results of the present article might be. Keeping in mind the speculative nature of possible answers such speculations can go in two directions: the strength of predispositions in guiding policy attitudes and the material consequences of an employment transition.
Regarding the strength of predispositions, the literature on party identification has shown that its effects increase under high polarization (Druckman, Peterson, and Slothuus 2013). Germany at the turn of the century, however, has undergone a decline in polarization (Munzert and Bauer 2013) as well as a decline of the number of people identifying with a party and of the strength of party identification (Ohr, Dülmer, and Quandt 2009). The only moderate overlap between party identification and LMP attitudes shown in the descriptive findings fits rather well with this picture.
Regarding the material consequences of unemployment, the Hartz reforms introduced major changes to the German labour market. Given their implementation between 2003 and 2005 these changes all happened after the analysis period of the current paper. Among other things, the Hartz laws shortened the period for income-dependent unemployment benefits (60% of prior income) to twelve months and introduced several sanctions i.e. possibilities to cut unemployment benefits (Kemmerling and Bruttel 2006). Thus, during the period of analysis in this article unemployment (or the risk thereof) was still less hurtful than afterwards. Interestingly, the mentioned findings from Margalit (2013) from the US occur in a context of increasing polarization (Druckman, Peterson, and Slothuus 2013) and higher costs of job loss given the weaker social safety of a typical liberal welfare state (Ferragina and Seeleib-Kaiser 2011).
Analysing the role of these different context effects in driving changes of LMP attitudes after employment transitions is certainly an interesting avenue for future research. However, this article's basic idea is neither limited to employment transitions nor to LMP attitudes but can be transferred to other changes of individual material circumstances (e.g. income changes) or attitudes (e.g. redistributive attitudes or immigration related attitudes). In these (and similar) cases, it is possible and necessary for a sound judgement about the relevance of self-interest vis-á-vis predispositions to answer the question: whose attitudes change? Notes 1. Subjective worries about job insecurity can result from objective unemployment risks and from certain personality factors (Debus, König, and Kleinmann 2014;DeWitte 2005). Thus, it is not necessarily the case that people in an objectively insecure employment situatione.g. in fixed-term employmentperceive that situation as insecure (Schraff 2017). 2. Throughout the paper, the terms passive LMPs and unemployment benefits or financial help for the unemployed as well as the terms active LMPs and job creation measures are used interchangeably. Note, however, that especially active LMPs include many more measures. In cases of 'demanding' active LMP measures that e.g. implement sanctions against unsatisfactory efforts for job search one might even expect less support by labour market outsiders, as argued and shown by Fossati (2018). Such demanding active LMPs are, however, neither theoretically nor empirically the focus of the present article. 3. Of course this is not to say that party identification or political ideology are not used in social policy attitudes research, as evidenced by Jaeger (2006). 4. The SOEP data that support the findings of this study are available at doi:10.5684/soep.v31.1 after signing a user contract. 5. Using this combined measure of LMP attitudes additionally comes with the benefit that this measures allows for more potential changes than the original 5-point scale. 6. Building these subgroups could lead to ceiling or floor effects for the 'pro' and 'contra' subgroups. This problem is reduced to some degree by using the combined measure of LMP attitudes with its higher number of scale points to build the subgroups. 7. Given the subgroup analyses in presence of a dependent variable measured on a 5-point scale the choice of linear fixed effects models might seem odd especially in comparison to e.g. tobit models built to deal with ceiling / flooring effects. However, no established solution exists to implement these models with fixed effects, as estimators face the so called 'incidental parameters problem' resulting in biased point estimators or standard errors (Greene 2004;Cameron and Trivedi 2005, ch. 23.5) 8. To address these concerns, all analyses are repeated using several additional time-varying covariates. In the case of the main effects shown in Figure 4 this does not alter the results with the exception of more optimistically perceived reemployment chances. This negative main effect loses its significance after adding the time-varying controls. 9. The fact that the results are so mixed also reduces to some degree the concern, that results are affected by ceiling or floor effects. Of course, this does not exclude the possibility, that they are still present in the findings. 10. I control for these time-varying variables in two different ways. On the one hand these control variables are added to the subgroup analyses. On the other hand, these control variables are added to the interaction models mimicking the subgroup analyses. Both modelling solutions are not exactly equivalent, as the interaction models do not include interactions between the control variables and the subgroup indicators, i.e. respondents' first wave party ID or LMP attitudes. As fewer parameters are estimated in the interaction models, this could increase the efficiency of the estimates (of employment transitions and job insecurity changes). In fact, however, differences between both ways of including the control variables are hard to spot. In any way, both modelling solutions yield exactly the same conclusions as discussed here.