Preferences, vote choice, and the politics of social investment: Addressing the puzzle of unequal benefits of childcare provision

Abstract Research on the politics of social investment finds public opinion to be highly supportive of expansive reforms and expects this support to matter for the politics of expanding social investment. Expanding social investment, it is argued, should be particularly attractive to left-wing voters and parties because of the egalitarian potential of such policies. However, few studies have examined to what extent individual preferences concerning social investment really matter politically. In this paper, I address this research gap for the crucial policy field of childcare by examining how individual-level preferences for expanding childcare provision translate into voting behavior. Based on original survey data from eight European countries, I find that preferences to expand public childcare spending indeed translate into electoral support for the left. However, this link from preferences to votes turns out to be socially biased. Childcare preferences are much more decisive for voting the further up individuals are in the income distribution. This imperfect transmission from preferences to voting behavior implies that political parties could have incentives to target the benefits of childcare reforms to their more affluent voters. My findings help to explain why governments frequently fail to reduce social inequality of access to seemingly egalitarian childcare provision.


Introduction
Over the past few years, the notion of a social investment welfare state has received increasing scholarly attention. The underlying idea of this work has been that a recalibration of the welfare state from passive compensatory transfers to social policies aimed at creating, mobilizing, and preserving skills (Garritzmann et al., : ) has the potential to strengthen social equality, welfare state sustainability, and economic growth (see also Morel et al., ; Hemerijck, ). Yet, compared with the potential benefits, the implementation of social investment reforms has so far been rather piecemeal. In some countries, governments have expanded social investment over recent years whereas in others governments have even cut back on it (Hemerijck, ; Morgan, ; Bouget et al., ). In addition, socially disadvantaged individuals are often found to benefit less from social investment policies than individuals in the middle classes (Pavolini and Van Lancker, ; Bonoli and Liechti, ). Those unequal benefits put into doubt whether social investment policies are in fact able to unfold their egalitarian potential.
This discrepancy between potential benefits and realized reforms has motivated research on the politics of social investment. Generally, scholars expect expansive social investment reforms to be attractive to governments operating under severe fiscal constraints. Left-wing parties in particular might be able to use social investment as a means of "affordable credit claiming" by enacting widely popular, egalitarian, and relatively inexpensive social policies in times of severe fiscal constraints (Bonoli, ). The growing number of studies on individual-level preferences towards social investment is in line with such reasoning, finding overall high levels of public support for expanding social investment (for example, Garritzmann et al., ). Studies diverge more regarding the amount of class conflict surrounding social investment reforms. While some studies expect the more affluent middle classes to be most supportive of social investment (Häusermann and Kriesi, ), others find support to be highest in the lowest income groups (for example, Chung and Meuleman, ). In most studies of individual-level preferences towards social investment, including my own, the relevance of studying preferences is motivated by the argument that preferences matter as a potential factor shaping the politics of expanding social investment (Beramendi et al., ; Häusermann and Kriesi, ; Busemeyer and Neimanns, ; Chung and Meuleman, ; Neimanns, ; Neimanns and Busemeyer, ). However, those studies hardly pay attention to the question of the extent to which those preferences in fact translate into politics.
In this paper, I address this research gap by examining how individual preferences with regard to early childhood education and care (hereafter, childcare) translate into real political behavior. Childcare is a crucial test case for social investment more generally, given that it is often considered a central element of a social investment approach (Morel et al., ; Hemerijck, ), and that social inequality of enrolment is often particularly pronounced (Van Lancker, ; Abrassart and Bonoli, ; Van Lancker and Pavolini, ). Using original survey data for eight European countries (INVEDUC, ), I examine how preferences for increasing public childcare spending affect party vote choice.
My central claim is that the link from childcare preferences to voting behavior, which can be seen as a necessary condition for individual-level preferences mattering politically, is less trivial than it is often portrayed. My findings show that, on average, preferences for increasing childcare spending are indeed related to voting behavior, with supporters of expanding childcare being more likely to vote for left-wing parties. This association holds when controlling for preferences regarding social spending and social value orientations, which takes into account that childcare preferences are embedded in broader latent attitudes.
However, it turns out that the transmission from preferences to votes is socially biased. For lower-income individuals, childcare preferences are unrelated to voting intentions. The higher up the income distribution individuals are, the more tightly voting behavior connects to preferences towards childcare. Thus, although lowerincome individuals might be the strongest supporters of additional public childcare spending, political parties of the left and the right have incentives to target reforms to more affluent voters because those voters' preferences translate more directly into actual votes. Taken together, my findings help us to understand why expansive social investment reforms are implemented less than one might expect and why they are often implemented in a socially stratified way. More broadly, these insights also connect to recent debates about unequal political representation (Bartels, ; Gilens and Page, ; Elsässer et al., ).
The politics of social investment: Individual-level preferences and voting behavior Most studies of individual-level preferences regarding social investment and childcare policies argue that preferences matter as a potential factor shaping the politics of expanding social investment and childcare provision (for example, Beramendi et al., ; Häusermann and Kriesi, ; Busemeyer and Neimanns, ; Chung and Meuleman, ). Generally, social investment policies are found to be widely popular among the public, with large majorities supporting expansion (Garritzmann et al., ). Besides assessing the overall level of support for social investment policies, analysis of the extent of class conflict in preferences has received central attention in this literature. Häusermann and Kriesi () argue that the highly educated middle classes benefit most strongly from social investment and should therefore be the strongest supporters of such policies (see also Beramendi et al., ; Garritzmann et al., ). Evidence from Garritzmann et al. () supports this line of reasoning. Income is found to matter less as a determinant of preferences regarding various social investment policies compared with passive transfer policies. Yet, various studies that more directly examine preferences concerning childcare tend to find lower-income respondents to be significantly more supportive of a strong role of the state in childcare provision (for example, Borck and Wrohlich, ; Chung and Meuleman, ).  None of these studies has examined the extent to which preferences in fact translate into politics, however.  Thus, it remains unclear whether public support for expanding childcare originates primarily from the lower or from the middle-income classes, and how this class-related patterning of preferences affects the politics of expanding childcare.
Why should preferences regarding childcare matter politically? To matter politically, preferences need to translate into political behavior, with voting behavior being probably the most direct and most widespread manifestation       of political behavior. The transmission from preferences to votes rests on certain informational requirements (Hellwig, ). Individuals need to identify a political party's position on a specific policy issue, and they need to be able to identify differences in positions across parties. The embeddedness of issues in broader societal cleavages and parties´programmatic identities usually help individuals to overcome such information problems (ibid.). In today´s postindustrial societies this cleavage structure is usually found to be two-dimensional, with preferences for expanding childcare provision being expected to correlate with economic positions and social values leaning more to the left (Beramendi et al., ; Häusermann and Kriesi, ; Garritzmann et al., ). To matter politically, childcare preferences would need to be a determinant of vote choice, independent of latent attitudes along the two dimensions of economic position and social values. If childcare preferences were only a subset of latent attitudes, parties´incentives to respond to voters´preferences to expand childcare could be rather low. Voters could be expected to be less likely to cast their vote based on their specific childcare preferences, but more on their broader latent attitudes on socio-economic issues and social value orientations. Parties, in turn, could find it more attractive to expand other areas of social spending or address other aspects of gender egalitarian positions in order to satisfy voter preferences.
The cognitive identification of partisan differences furthermore needs to translate into political behavior. First, childcare provision would need to be sufficiently important for relevant electoral groups to base their vote choice upon their preferences concerning childcare. Value change among highly educated middle-class women has been identified as a central factor leading to partisan de-alignment; in other words, to the erosion of ties between the established parties and those voter groups (Morgan, ; Blome, ; Schwander, ; Abou-Chadi and Wagner, ; León et al., ). The established parties have for a long time paid only limited attention to issues of work-family life reconciliation. Religiosity used to be an important factor in ensuring the political viability of this neglect by mainstream parties. Because sizable voter segments, and, above average, women, cast their votes for center-right parties out of mostly religious motivations, party competition about progressive family policies has often been limited (Emmenegger and Manow, ). With the declining importance of religion, also the strength of partisan attachments decreased. Those swing voters should be highly supportive of expanding childcare provision, and, it is hypothesized, childcare provision should often be sufficiently important to them to affect their vote choice.
As a second requirement for preferences to have a political impact, this partisan dealignment would need to have consequences for party competition; in other words, it would need to open up the potential for partisan realignment. With their future careers depending on continuing electoral success, professional politicians are regularly found to be highly attentive to changing preferences in their electorates (Soroka and Wlezien, ). They often devote considerable resources to inform themselves about public opinion, relying on the news media (Sevenans et al., ), their professional and social networks (Herbst, ), or public opinion surveys (Hager and Hilbig, ). For the case of childcare provision, more specifically, women´s groups within the political parties have often been relevant in raising awareness for the importance of childcare expansion to support gender equality and families that are increasingly not adhering to the male breadwinner model (Morgan, ). Concerns about demographic change in many Western societies further contributed to the perceived importance of modernized family policy in supporting the reconciliation of work and family life (ibid.; Fleckenstein and Seeleib-Kaiser, ). The combination of a growing salience of the issue of childcare, and of partisan dealignment contributing to heightened electoral competition (Hobolt and Klemmensen, ) render it plausible that parties have been particularly attentive to public opinion on childcare and have adjusted their positioning to these new societal and political demands.
The incentives to attract votes by expanding public spending on services targeted to electorally relevant groups are expected to be particularly pronounced for left-wing parties (Häusermann and Kriesi, ). Expanding childcare spending provides an affordable, potentially redistributive means of continued state engagement in social policy in mature welfare states and it responds to value change towards support for a more gender-egalitarian distribution of work and family responsibilities. With regard to voter preferences along a distributive and a social values dimension, left-wing parties are likely to find wider support and less opposition among their core electorates to expand childcare compared with right-wing parties (Häusermann and Kriesi, ; Garritzmann et al., ; Schwander, ).  Thus, it should be easier and electorally more rewarding for left-wing parties to adjust their programmatic positions to societal value change.
In line with these expectations, empirical findings show, first, that left-wing governments are more likely to expand childcare provision and, second, that expanding childcare provision pays off electorally. Studying the effects of government partisanship on public childcare spending, Hieda () and Busemeyer and Seitzl () find left-wing governments to be more likely to expand childcare. In a similar vein, studies focusing on the electoral consequences of parties´programmatic positioning find that if left-wing parties emphasize childcare and related policies more strongly, their vote share increases (Nelson and Giger, ; Abou-Chadi and Wagner, ). On average, left-wing parties seem to be able to promote childcare expansion and to reap the electoral benefits of such programmatic positioning.
While the studies discussed above suggest a positive link between individuallevel preferences regarding childcare and voting for the left, to date no study has directly examined whether this link really exists. Studies that have examined the electoral consequences of value change have focused primarily on value change across different electoral groups, based on factors such as gender or occupation.       As a consequence, this literature remains disconnected from studies on the determinants of childcare preferences, which come up with distinct expectations regarding the popularity of childcare expansion and the associated potential of class conflict. Building on the literature discussed above, I expect: Hypothesis : The more individuals support an expansion of public childcare spending, the more likely they are to vote for left-wing parties.
The expectation spelled out in Hypothesis  leaves us with an important empirical puzzle. If childcare preferences tend to go along with support for state involvement and redistribution, and if left-wing parties are often the central promotors of expansive social investment reforms, why have recent childcare reforms often failed to address the social upward-bias in childcare enrolment (Van Lancker, )? Across Western countries, more affluent families are overrepresented among the users of childcare because lower-income families are to a disproportionate extent negatively affected by problems of affordability (Abrassart and Bonoli, ), availability (Pennerstorfer and Pennerstorfer, ), or a combination of factors (Van Lancker and Ghysels, ; Lewis and West, ).  In the following, I argue that the link between preferences and voting is likely to be in itself socially upward biased as well.
I expect factors on the input and output side of the political process to matter for the conditioning impact of individuals' socioeconomic position on the preference-voting nexus. First, with regard to the input side, lower-income individuals are more likely to have lower political efficacy (Marx and Nguyen, ). The experience of economic problems is emotionally and cognitively absorbing and reduces individuals´willingness and capacity to stay politically informed. Because low-income individuals are less likely to believe that their vote could make a difference or to be aware of parties´programmatic offers, their preferences towards childcare should matter less for their vote choice. In addition, low-income individuals are particularly likely to abstain from voting, which further reduces their weight in the political process (Schäfer, ). Political efficacy might be particularly important for the case of childcare because of its investment character, with some of its benefits materializing only at some distant point in the future. Whereas individuals in a low socioeconomic position might assign more weight to their current consumption needs relative to future investment, those in a more privileged socioeconomic position are better equipped to take a more long-term perspective (Beramendi et al., ).
Second, on the output side, lower levels of political efficacy find their reflection in socially unequal political representation. A growing number of studies document that the preferences of affluent individuals are better represented in political reforms, whereas the preferences of lower income groups often hardly influence policy output (Bartels, ; Gilens and Page, ; Elsässer et al., ). There can be a variety of reasons for unequal political representation, including reliance on private campaign donations, unequal political participation, or unequal descriptive representation in parliament (ibid.). These factors help to understand why governments could be more responsive to more narrowly defined private interests, neglect preferences of nonvoters, and might neglect preferences of their lower income constituencies, if their assessment of public opinion is biased by their own socioeconomic background and their specific social and professional networks (Carnes, ; Schäfer, ). Issues that are important for individuals with lower incomes are often also less salient in the public discourse which reduces the pressure for governments to address those issues (Busemeyer et al., : ). The limited influence on reform processes is likely to further reduce the motivation of lower-income individuals to base their vote on their specific preferences towards childcare policy.
In the case of childcare more specifically, unequal policy benefits are reflected in socially unequal enrolment patterns (Van Lancker, ). Political debates about expanding childcare have often been driven by the goal of facilitating the combination of work and family life for working parents: in particular, working women (Fleckenstein and Seeleib-Kaiser, ; Naumann, ; Morgan, ). In contrast, aspects related to access to childcaree.g. for non-employed parents, or support for disadvantaged children, which might be of particular importance for lower-income constituenciessometimes received a relatively low weight in the political process (Lewis and West, ). Socially stratified access to childcare might lead lower-income individuals to expect that they would not benefit from expansive childcare reforms (Neimanns and Busemeyer, ) and therefore might not let childcare preferences affect their vote choice. For those who already enjoy privileged access to childcare it might appear more plausible that they would benefit from additional public money spent on childcare.
Taken together, the factors related to the input and the output sides of the political process should contribute towards a socially biased link between childcare preferences and voting for the left. Lower-income individuals should have fewer incentives to base their vote choice on their specific childcare preferences, whereas for more affluent individuals the opposite should be the case: Hypothesis : The strength of the association between support for expanding public childcare spending and voting for left-wing parties increases with income.

Methods and data
The empirical analysis builds on data from an original public opinion survey (INVEDUC, ) which we collected as part of a larger collaborative       research project. In this survey, we asked respondents for a range of preferences regarding education and social policies. The survey was conducted by a professional survey company using computer-assisted telephone interviews in early  in eight European countries (Denmark, France, Germany, Ireland, Italy, Spain, Sweden, and the United Kingdom). The random probability sample is representative of each country´s adult population and comprises a total of , individuals (for details, see Busemeyer et al., ).  To examine how childcare preferences translate into voting behavior, I run multinomial logistic regression models with vote intention as the dependent variable ("If there was a general election tomorrow, which party would you vote for?"). The list of major parties in their country was read out to respondents and several residual categories for those without clear party preferences were mentioned as well. The mentioning of additional parties was recorded via open-ended questions. I include only individuals with citizenship of the respective country, who should be eligible to vote in the national elections. For the analysis, I assign parties to party families, following the coding by Döring and Manow () (see Table A. in the appendix). Several further operationalization steps are necessary to avoid the number of observations for the individual parties becoming too small. First, I concentrate on left-wing and center-right parties because the number of observations is very small for some of the other parties in some countries. Second, on the left, I pool "communist/socialist," "social democratic," and "green/ecological" parties. On the center-right, I combine "Christian democratic" and "conservative" parties into a single category. Preliminary models show that pooling is appropriate. Childcare preferences do not matter as a determinant for vote choice between these three left-wing party families, on one hand, and the two center-right party families, on the other hand (see Table A. in the appendix). I refrain from including liberal and farright parties in the right-party category, because those parties tend to be less coherent in their positions on social policy across countries compared with the other party families (Busemeyer et al., ; Röth et al., ).  I use the residual category of having no determined voting intention for a specific party (which includes on average approximately  percent of respondents) as the reference category in the multinomial logistic regression models.
Focusing on voters of left and center-right parties implies that evidence of limited political influence of individuals from a lower socio-economic position should be interpreted as conservative estimates. The reason for this is abstention from voting, which is usually much more widespread among individuals from a low-income background (Schäfer, ). This pattern is also present in the data used for this analysis. The likelihood among respondents in the lowest income quintile of reporting to not go to vote, to vote blank, or to cast a spoil vote is  percent, whereas it is around  percent for individuals in the three highest income quintiles (see Table A. in the appendix).
Preferences towards public childcare spending are the central independent variable. In the INVEDUC survey, respondents were asked: "Please tell me whether you would like to see more or less government spending in each of the following areas. Keep in mind that 'more' or 'much more' might require a tax increase: Pre-school and early childhood education"  Respondents could indicate their preference for "much more", "more", "same as now," "less," or "much less" spending. Higher values indicate a preference for more public spending. The advantage of this variable on spending preferences is that it is a more direct and probably more realistic measure of individual-level support for government expansion of childcare provision than other available measures, such as the widely used item "government responsibility to ensure sufficient childcare" from the European Social Survey (ESS, , ). As a comparison of factor analyses of preferences regarding various areas of social policy from the ESS and INVEDUC reveals, childcare preferences in the INVEDUC survey correlate much less with broader social policy preferences (see Table A. in the appendix). Using this measure of childcare spending preferences implies that I will be better able to capture the effect of preferences that is specific to childcare preferences, rather than measuring the effect of social policy preferences more generally. In addition, summary statistics show that support for public spending increases on childcare is more divided in contrast to the close to unanimous levels of support for childcare in the ESS (see Figure A. in the appendix). On average,  percent of respondents support additional public spending on childcare. Requiring the government to devote additional public resources to expand childcare might be more controversial in terms of political preferences, and it might be the more realistic measure of political preferences, compared with the more abstract notion that the government should "ensure sufficient childcare." In order to be able to identify the effects of childcare preferences on voting behavior, it is necessary to account as much as possible for potential endogeneity of childcare preferences to partisan preferences. If childcare preferences were endogenous to partisan preferences, this would imply that childcare preferences do not drive voting behavior, but rather that childcare preferences are a subset of broader ideological orientations. To control for this potential endogeneity, I add latent attitudes on social spending and social values as control variables. This allows me to capture the "net" effect of childcare preferences on voting behavior, independent of the two-dimensional left-right cleavage structure that has been found to matter in contemporary post-industrial societies (Beramendi et al., ). More specifically, I control for general preferences towards social spending ("Should the government spend more or less on social benefits and social services?") and for social value orientations, which are the       predicted values of factor scores of two items on multiculturalism and lawand-order policies ("People who break the law should be given much harsher sentences than they are these days" and "[COUNTRY]'s cultural life is generally enriched by people coming to live here from other countries"). The detailed operationalization of all variables included in the analysis is described in Table A. in the appendix.
I use household income (measured in quintiles) as an indicator of individuals' socioeconomic position, which I expect to moderate the effect of childcare preferences on voting behavior. Household income is a suitable indicator in this context because it corresponds to the measures commonly used to capture inequality in the input and the output sides of the political process discussed above. In addition, I include a range of control variables, which are standard in analyses of the determinants of childcare preferences (for example, Busemeyer and Neimanns, ; Chung and Meuleman, ). I control for age, gender, having kids above or below the age of ten, being a single parent, household size, employment status (interacted with gender), public sector employment, and whether a respondent lives in an urban or a rural area.
The analysis is run for the pooled sample of eight countries. I include country fixed effects in all models to control for unobserved heterogeneity at the country level. The main reason for pooling is to avoid the size of the different income-specific partisan voter groups becoming too small. Pooling is not unproblematic because childcare provision varies considerably across these eight European countries in terms of funding or coverage (Morgan, ). This variation could also be reflected in parties´programmatic positioning regarding childcare policy. As a consistency check of whether it is appropriate to expect support for childcare expansion to be associated with support for left parties across countries, I examine party positions on the broader issue of traditional morality based on data from the Comparative Manifesto Project (Volkens et al., ). Because the issue of childcare policy has often been embedded in a discourse on morality and debates about modern or traditional family and gender roles (Morgan, ; Emmenegger and Manow, ; Blome, ), and given the lack of more adequate data with suitable spatial and temporal coverage, this data provides a useful consistency check. Table A. in the online appendix shows that around the time of the survey left parties in basically all countries had less traditional positions than their center-right counterparts on the issue of traditional morality.  Although partisan differences have been declining since the s, positioning on traditional morality continued to clearly divide parties of the left and right ( Figure A. in the appendix). Thus, it appears reasonable to expect preferences for childcare expansion to translate into electoral support for left parties across countries.

Results
Before assessing the impact of preferences regarding childcare expansion on voting intentions, it is necessary to clarify the extent to which it is plausible to expect childcare politics to structure political competition along socioeconomic lines. As discussed above, studies diverge in their findings on the relevance of class conflict in childcare preferences. Figure  plots predicted probabilities of supporting public spending increases on childcare by income group, controlling for socio-demographic and latent attitudinal covariates. The full regression results are reported in Table A. in the appendix.  The main message of Figure  is twofold. On the one hand, on average, individuals from the lowest income quintile are the strongest supporters of increases in public childcare spending. The predicted probability of supporting additional public childcare spending is . percentage points higher among people in the lowest income quintile compared to people in the highest income quintile. With the exception of individuals from the second income quintile, the association between income and childcare preferences is linear. The remarkably low level of spending support of individuals from the second income quintiles requires further examination. Additional analyses ( Figure A. in the appendix) show that parents with small children are underrepresented and students and individuals with more traditional value orientations are overrepresented in the second income quintile. These characteristics matter in shaping preferences towards childcare spending and help to explain why individuals from the second  Table A. in the appendix.
      income quintile deviate from the negative linear effect of income on preferences towards public childcare spending.
On the other hand, the income effect identified in Figure  is fairly weak in substantive terms and does not achieve statistical significance. Thus, while, on average, there does not appear to be a large potential for social class conflict surrounding childcare expansion, it is also inappropriate to consider support for expanding public childcare spending to originate primarily from middleincome groups.  Given that the socioeconomic structure of childcare preferences resembles the one frequently found for preferences regarding social policy and redistribution more generally (for example, Finseraas, ), based on Figure , it appears plausible to expect preferences for expanding public childcare spending to translate into stronger electoral support for left vis-à-vis centerright parties. Figure  shows how the predicted probabilities of voting for left or centerright parties depend on childcare spending preferences (see Table A. in the appendix for the full results of the multinomial logistic regressions). Figure  reveals that preferences for increasing public childcare spending are indeed tightly connected to electoral support for the left. Comparing individuals with a preference for unchanged spending levels to those who want to see much more public childcare spending accounts for an increased likelihood of voting for the left by . percentage points. This finding is likely to reflect partisan  Table A. in the appendix. realignment, with left-wing parties being particularly responsive to value change in the electorate and with voters rewarding this (cf. Nelson and Giger, ). Figure  also indicates partisan dealignment for center-right parties. The likelihood of voting for center-right parties decreases by . percentage points among individuals with preferences for much more public childcare spending (compared to those wanting spending to remain unchanged). Thus, while the voting behavior of supporters of childcare expansion seems to matter for the electoral fortunes of both left and right parties, the stakes appear to be higher for left-wing parties. The electoral prospects of left-wing parties hinge to a particular extent on their capacity to attract the votes of those individuals supporting childcare expansion.
The findings presented in Figures  and  raise an important empirical puzzle: If childcare policies are demanded by individuals who would benefit from their redistributive potential, and if left-wing parties have incentives to expand such redistributive childcare policies, why does childcare enrolment continue to be highly socially stratified in most countries? Why do policymakers frequently fail to enable lower-income people to benefit more from childcare expansion, even though, on average, they are the policy's strongest supporters? As I have argued above, the impact of childcare preferences on voting behavior is unlikely to be the same across socioeconomic groups. To evaluate this claim empirically, I add an interaction between preferences regarding public childcare spending and household income and in Figure  report the marginal effects of childcare preferences by income on support for left or center-right parties.  The full regression results are presented in Table A. in the appendix. Figure  strikingly shows that the association between childcare spending support and voting intentions is conditioned by income. The positive effect of childcare spending support on voting for the left rises with income. For respondents in the lowest income quintile, spending preferences do not matter for voting choice. In contrast, for respondents in the highest income quintile, support for increasing childcare spending increases the propensity to vote for the left by . percentage points. The flip side of the positive effect of income on the preference-voting nexus for the left is a negative income effect for center-right parties. Childcare preferences do not matter for voting intentions for center-right parties among individuals in the lowest income quintile for whom the average marginal effect estimate is basically zero. Childcare spending support is negatively signed and statistically insignificant for three of the other income quintiles, but for individuals from the fourth income quintile there is a significant effect of spending support on a reduced likelihood to vote for center-right parties (. percentage points). Taken together, Figure  suggests that higher-income voters, and in particular those potentially voting for the left, are more likely to base their vote choice on their preferences regarding childcare expansion.  One implication of this finding is that parties might have an incentive to target reforms to higher-income voters because it might turn out to be electorally most rewarding.

     
Is the social bias in the preference-voting nexus specific to the case of childcare? While some of the arguments that led me to Hypothesis  should apply across policy issues, the long-term investment character (Beramendi et al., ) and the nearly omnipresent high level of social inequality of enrolment (Van Lancker, ) should render this social bias particularly pronounced for the case of childcare. To examine this, I replicate Figure Figure . One important difference, however, is that in these additional figures preferences are a significant predictor of vote choice across income groups, including individuals in the lowest income quintile.  Thus, the irrelevance of childcare preferences to the voting behavior of low-income individuals, as identified in Figure , seems to be particularly linked to the characteristics of childcare.

Conclusion
Recent research suggests that high levels of public support for social investment policies should provide incentives for political parties to expand social investment. To date, however, no study has directly examined the extent to which  Table A. in the appendix.
preferences regarding social investment in fact translate into politics. This paper has addressed this research gap for one step of this causal chain for the crucial policy field of childcare. It has analyzed how individual preferences for expanding public childcare spending translate into politics: namely, by affecting individual propensities to vote for left or center-right political parties. I find that preferences indeed seem to translate into voting behavior, with individuals supportive of childcare expansion being more likely to vote for left and less likely to vote for center-right parties. The transmission from preferences to votes is, however, highly socially biased. For individuals in the lowest income groups, preferences are not significantly related to voting behavior. With rising income, preferences become more decisive in affecting voting intentions.
My findings help us to understand the often socially stratified nature of social investment reforms. Despite being demanded in particular by individuals in a lower-income position, left-wing parties could have incentives to target childcare reforms to individuals located higher up the income distribution because the latter's voting behavior is most tightly connected to their preferences regarding childcare expansion. The same can be expected to hold true for rightwing parties, which might try to prevent (middle-and upper-income) voters interested in childcare expansion defecting to the political left. These incentives to discount the preferences of lower-income individuals are likely to be even exacerbated because lower-income individuals are more likely to abstain from voting (Schäfer, ). The results appear pessimistic from the perspective of social equality because they suggest that political opportunity structures make it difficult for political parties to maximize the egalitarian potential of social investment policies. More broadly, my results connect to the growing literature on unequal political representation (Bartels, ; Gilens and Page, ; Elsässer et al., ). The socially upward-biased link between preferences and voting presents one of several possible mechanisms that explain why parties might have stronger incentives to represent the preferences of more affluent citizens.
This paper raises several further questions, which should be addressed in future research. I examine only a small step in the causal chain of the politics of social investment, and I do this only for the single, but nevertheless crucial policy field of childcare. How voting shapes parties´programmatic offers and how it might in fact depend on those offers is beyond the scope of this paper. The same applies to cognitive decision-making processes at the level of parties: Are professional politicians aware of the varying preferences towards childcare provision among their different electorates and how could such information impact the positioning of parties regarding childcare expansion? Furthermore, a range of confounding and contextual factors can be expected to intervene in the relationship between childcare preferences and vote choice (cf. Emmenegger and Manow, ; Schwander, ). Future research should       explore in more detail how party competition and further contextual factors might either strengthen or attenuate the strength of and social bias in the preference-voting nexus. This would make it possible to identify political conditions under which childcare reforms are more likely to be designed in a way that in fact benefits the most disadvantaged individuals. other countries, the liberal parties are smaller than the main conservative or Christian democratic parties.  Clarifying notes in the questionnaire indicated that "pre-school early childhood education" covers all programs that include components of early childhood educational development before primary school.  The United Kingdom constitutes an exception with the Labour Party scoring somewhat higher than the Conservatives on the issue of traditional morality. This observation is in line with Emmenegger and Manow´s () notion of the absence of a strong religious cleavage in the UK. Nevertheless, given the extensive budgetary cuts in the area of childcare applied by the Conservative-led government in the period preceding the survey (Lewis and West, ), it appears plausible to expect support for childcare expansion to translate into support for the political left including in the UK.  To ease presentation, Figure  relies on a logistic regression model using a binary operationalization of childcare preferences. Preferences for "more" or "much more" spending (coded as ) are distinguished from preferences for "same as now," "less," or "much less" spending (coded as ). The results hold, if the original categorical measure is used with an ordered logistic regression model (Table A. in the appendix).  It should be noted that the country-pooled models presented here brush over cross-country differences in the potential for social class conflict across income groups (Neimanns and Busemeyer, ). Because I am interested in the average association between income and childcare preferences as a plausibility check for the hypothesized link between childcare preferences and electoral support for the left, I disregard cross-country differences in this analysis.  Because it is computationally too demanding to run the multinomial logistic regression model underlying Figure  with a categorical operationalization of childcare preferences (for which the number of responses in some response categories is low; see Figure