Literature review

This article broadly connects to the extensive literature on public opinion about the welfare state, which has grown significantly in recent years in line with the increasing availability of high-quality international comparative survey data; see Svallfors (Reference Svallfors2012) for a recent overview. As a consequence, scholarly knowledge about the individual and institutional determinants of welfare state attitudes has increased enormously. First, scholars have repeatedly found that material self-interest influences individual-level policy preferences to a significant extent (see e.g. Meltzer and Richard Reference Meltzer and Richard1981; Iversen and Soskice Reference Iversen and Soskice2001; Cusack et al. Reference Cusack, Iversen and Rehm2006; Busemeyer et al. Reference Busemeyer, Goerres and Weschle2009; Rehm Reference Rehm2009; Hacker et al. Reference Hacker, Rehm and Schlesinger2013; Margalit Reference Margalit2013; Kitschelt and Rehm Reference Kitschelt and Rehm2014; Schwander et al. Reference Schwander, Häusermann and Kurer2015). For instance, wealthy citizens were more likely to oppose welfare spending (Meltzer and Richard Reference Meltzer and Richard1981; Rehm et al. Reference Rehm, Hacker and Schlesinger2012); those with “specific skills” were more likely to support it (Iversen and Soskice Reference Iversen and Soskice2001; Cusack et al. Reference Cusack, Iversen and Rehm2006), as were those facing higher levels of labour-market risk (Rehm Reference Rehm2009; Rehm et al. Reference Rehm, Hacker and Schlesinger2012). The elderly were found to be more opposed to spending on education (Busemeyer et al. Reference Busemeyer, Goerres and Weschle2009), whereas the young, educated middle-classes were more supportive of social investment policies (Schwander et al. Reference Schwander, Häusermann and Kurer2015). Second, scholars have repeatedly pointed out that, besides self-interest, normative orientations and values also shape policy attitudes and preferences, e.g. altruism, political ideology and religiosity, as well as different perceptions of fairness, deservingness and social justice (Kangas Reference Kangas1997; Alesina and Angeletos Reference Alesina and Angeletos2005; Scheve and Stasavage Reference Scheve and Stasavage2006; Reference De La O and RoddenDe La O and Rodden 2008; Margalit Reference Margalit2013). Third, the literature has paid particular attention to the question of how institutional contexts shape preferences and attitudes at the individual level. A first wave of scholarship concentrated on how welfare state regimes, broadly conceived as configurations of macro-level institutions, are reflected in individual attitudes towards the welfare state (Svallfors Reference Svallfors1997; Andreß and Heien Reference Andreß and Heien2001; Blekesaune and Quadagno Reference Blekesaune and Quadagno2003; Svallfors Reference Svallfors2004; Jaeger Reference Jaeger2006; Jaeger Reference Jaeger2009; Svallfors Reference Svallfors2010; Svallfors Reference Svallfors2012). The initial expectation that preferences would be largely congruent with macro-level institutions was not entirely borne by the data (Jaeger Reference Jaeger2006; Jaeger Reference Jaeger2009). Hence, more recent contributions to the literature focus on particular institutional dimensions of the welfare state rather than on broad regimes, and/or they are more concerned with understanding how institutions mediate associations between variables at the microlevel in the form of cross-level interaction effects (Gingrich and Ansell Reference Gingrich and Ansell2012; Busemeyer Reference Busemeyer2013; Jordan Reference Jordan2013; Zhu and Lipsmeyer Reference Zhu and Lipsmeyer2015).

The literature on institutional determinants of welfare state preferences draws on the notion of policy feedback (Pierson Reference Pierson1993; Mettler and Soss Reference Mettler and Soss2004; Soroka and Wlezien Reference Soroka and Wlezien2010; Svallfors Reference Svallfors2010; Campbell Reference Campbell2012; Jordan Reference Jordan2013; Kumlin and Stadelmann-Steffen Reference Kumlin and Stadelmann-Steffen2014). In an early conceptual contribution to this literature, Pierson (Reference Pierson1993, 624) famously distinguished between “resource/incentive” and “interpretive” feedback effects. In the former case, welfare institutions create material incentives for citizens to support the continued existence (or even expansion) of certain social policy programs, e.g. pensioners support more generous pensions, the unemployed support a more generous unemployment insurance scheme. Thus, this argument is closely linked to accounts that emphasise the role of material self-interest in shaping attitudes and preferences. In contrast, “interpretive” effects concern the “impact of policies on the cognitive processes of social actors” (Pierson Reference Pierson1993, 610). Hence, institutions also shape cognitive and, one could add, normative expectations of citizens (Svallfors Reference Svallfors2010) regarding the role of the welfare state both in society and in their individual lives. In the empirical analysis below, we distinguish between these two feedback mechanisms by assessing whether fairness perceptions are influenced by the actual socio-economic characteristics of individual respondents or only by the traits of hypothetical individuals in the vignettes. The former hints at the presence of resource-based feedback, whereas the latter points to the importance of interpretive feedback.

More recent work on policy feedback has provided additional insights into which features of the institutional context are more important than others. For instance, Soss and Schram (Reference Soss and Schram2007) argued that policy feedback is more likely when policies combine high visibility with high proximity, i.e. a small regional or social or other distance between citizens and the policy in question. Gingrich and Ansell (Reference Gingrich and Ansell2012) added that the strength of policy feedback also depends on the linkages between policies and labor-market risks, whereas Kumlin (Reference Kumlin2014) showed that benefit generosity (in his case, the generosity of unemployment insurance) matters as well. Gingrich (Reference Gingrich2014) confirmed that the visibility of welfare state policies also affected the degree to which economic and social policy preferences matter with regard to vote choices. For the present article, the visibility and proximity of social policies are most important. Clearly, in the case of social services such as childcare which are delivered and – in our case – financed locally, we would expect strong policy feedback effects.

Besides improving scholarly knowledge about feedback mechanisms, recent work has also challenged Pierson’s (Reference Pierson1993) original claim that feedback effects are predominantly self-reinforcing, i.e. congruent with existing institutions. Increasingly, scholars have pointed out that institutions can also create feedback effects, which in the long run undermine rather than reinforce existing institutions (Streeck and Thelen Reference Streeck and Thelen2005; Soroka and Wlezien Reference Soroka and Wlezien2010; Weaver Reference Weaver2010; Jacobs and Weaver Reference Jacobs and Weaver2015). Thus, if citizens critically evaluate and assess a given set of policies, they might demand policies which support a change in the prevailing status quo rather than simply reinforcing it. Work on citizens’ perceptions of the performance of the welfare state, for instance, showed that citizens were well able to distinguish between abstract demands regarding what the welfare state should do and the concrete implementation of those abstract demands (Roosma et al. Reference Roosma, Van Oorschot and Gelissen2014). Popular attitudes might well be congruent with some aspects of prevailing policies and institutions, while at the same time, citizens might demand change regarding other aspects.

Moving from the broader literature on welfare state attitudes and policy feedback to the specific subject of childcare, we note a relative dearth of studies on public opinion on childcare policies despite the wealth of scholarship on social policy attitudes in general. This largely reflects the lack of survey data on childcare policies in existing international surveys, which are mostly concerned with the traditional parts of the welfare state such as pensions, labour-market policy or health care. There are, however, a few exceptions: a number of studies analysed public opinion on childcare policies across a larger number of OECD (Organisation for Economic Co-operation and Development) countries (Goerres and Tepe 2010, 2012a, 2012b, Reference Goerres and Tepe2013; Meuleman and Chung Reference Meuleman and Chung2012; Guo and Gilbert Reference Guo and Gilbert2014; Chung and Meuleman Reference Chung and Meuleman2017): there was overall widespread support for these policies, and ideology, in terms of egalitarian and gender attitudes, matters more than material self-interest as a determinant of individual-level support. These findings can be broadly replicated in studies that focus on particular countries; e.g. Henderson et al. (Reference Henderson, Monroe, Garand and Burts1995) and Bolzendahl and Olafsdottir (Reference Bolzendahl and Olafsdottir2008) for the United States; Goerres and Tepe (Reference Goerres and Tepe2012a) for Germany and Ellingsaeter et al. (Reference Ellingsaeter, Kitterod and Lyngstad2017) for Norway. A second common finding is that once established, childcare policies create “self-reinforcing policy feedback effects” in the sense that public support for further government involvement in the provision of childcare increases (Reference Hedegaard and LarsenHedegaard and Larsen 2014; Ellingsaeter et al. Reference Ellingsaeter, Kitterod and Lyngstad2017).

A significant shortcoming of this existing research is that it is mostly focused on exploring policy preferences at a rather abstract level. Typical questions on the welfare state in the European Social Survey or in the various modules of the International Social Survey Programme concern preferences in various areas of government spending and government responsibility or involvement without clarifying what kind of involvement is meant and how exactly public monies should be spent. To a great extent, this is of course related to the fundamental problem that a higher degree of specificity in question wording would endanger the comparability of responses across countries since the individual country contexts are so different. More importantly for the purpose of this article, there are no straightforward questions about the citizens’ attitudes on whether they regard paying for a particular service such as childcare as fair. In addition, commonly used survey designs would have difficulty in assessing this issue because citizens’ responses could be influenced too much by their respective personal situations. By using vignettes to measure citizens’ fairness perceptions of different fee levels, we provide a new perspective on the study of public opinion on childcare policies specifically, and on welfare state policies more generally.

Theoretical expectations and hypotheses

Our analysis essentially centres on the question of whether popular attitudes towards a particular policy are congruent with the prevailing institutional context or not. This in turn provides insights into whether policy feedback (in that particular case) exhibits a self-reinforcing or self-undermining dynamic. Furthermore, by being able to determine the association between the characteristics of hypothetical individuals in vignettes and the characteristics of the actual respondents, our study contributes to a better understanding of the relative importance of interpretive and resource-based feedback mechanisms. We focus on a particular policy in a local context. This policy is highly visible as childcare fees directly affect parents and are regularly debated in the city council and in the local newspapers. Therefore, we expect strong feedback effects in general, allowing us to tease out the previously mentioned mechanisms more easily than in a study of abstract preferences at the national level.

We focus on two issues: first, the level of fees that on average would be considered as fair by respondents, and second, the fee structure, in particular whether fees should be dependent on parents’ income and family situations or not. Compared to traditional social policies, these institutional factors correspond to welfare state generosity (lower childcare fees imply a more “generous” provision of social services) and degree of regressivity (if fees are strongly dependent on parental income, the scheme is more progressive, i.e. poorer citizens are relatively better off). Empirically, we capture attitudes towards the fee structure by assessing which characteristics of the (fictitious) parental background would influence fairness perceptions of the level of fees. This approach is preferable to asking citizens directly for their preferences on fee structures because it requires less knowledge about the technical details of fee schemes. It also allows us to assess the impact of a broader set of characteristics on fairness perceptions of fees, for instance religious or migration background and employment status.

Data and methods

The vignette study

We ran a factorial survey experiment in which we presented a representative sample of citizens from Konstanz with different “vignettes” to obtain a better measure of their fairness perceptions about the fee level and the structure of the childcare fee system. The difference from conventional survey designs was that respondents were asked to evaluate a hypothetical situation, which usually is different from their own personal situation. Although individual respondent characteristics might still matter to some extent, this research design was particularly well-suited for identifying abstract policy preferences independent of the respondents’ personal situation. Factorial survey experiments and conjoint analyses are not yet widely applied in the field of comparative welfare state research, but they have several advantages compared to traditional surveys, in particular because they allow for the manipulation of multiple treatment conditions (see Hainmueller et al. Reference Hainmueller, Hopkins and Yamamoto2014; Auspurg and Hinz Reference Auspurg and Hinz2015; Hainmueller et al. Reference Hainmueller, Hangartner and Yamamoto2015 for accessible introductions). There are some recent examples in political economy and welfare state research that indicate a growing interest in this methodological approach and that demonstrate its applicability and usefulness. For instance, Gallego and Marx (Reference Gallego and Marx2017) analyse popular preferences for different kinds of labor-market policies, paying particular attention to the potential influence of trade-offs. Bansak et al. (Reference Bansak, Hainmueller and Hangartner2016) and Hainmueller and Hopkins (Reference Hainmueller and Hopkins2015) study public attitudes towards migrants and asylum seekers.

The field period for the vignette study was between 17 October and 26 December 2014. The sample was randomly drawn from the official resident register in which all residents in Germany are obliged to register. The respondents were part of a long-term panel sample with regular updates as part of a collaborative project between the university and Konstanz (Konstanzer Bürgerbefragung). The survey was conducted online. Ideally, the sample represented the resident population of Konstanz aged 18 and above, but it was somewhat distorted in comparison to the general population in the sense that young, better-educated and politically more interested people were overrepresented. In contrast, individuals with foreign citizenship, and those living in one-person households, were underrepresented. Further details can be found in Appendix A. The respondents lived in 14 areas of Konstanz that they identified themselves in the survey. Since the sample was based on “persons-living-at-addresses” in the register, the areas provided a context for the sample because the respondents actually resided in those areas.

The vignette describes a family with a 2-year-old first child who is attending a public childcare facility in Konstanz. The respondents were asked to rate the fairness of an indicated monthly fee for a fulltime place at the facility (excluding the cost of meals). We used the common bipolar 11-point scale, where the value of “0” is located in the middle and is equivalent to an assessment of the fee as “Fair”. Values ranging from “−1” to “−5” indicate various degrees of “Unfairly too low”, and the values of “+1” to “+5” various degrees of “Unfairly too high”. If respondents did not choose the mid-point, they were given the opportunity in a second question to indicate the fee that they considered fair. Hence, for each respondent, we either have the accepted fair fee or the fair fee amount indicated by him or herself. Each respondent was given a deck of five vignettes and each had different characteristics. Details about the construction of the vignette universe and the decks can be found in Appendix B. A screenshot of the vignette is in Appendix C.

We created a new variable “fair fee”, which we employed as the dependent variable in the analysis. The value of this variable was either equal to the originally displayed fee, if it was considered fair, or equal to the new fee level entered by the respondent herself. For 26.1% of all ratings, we took the numerical value of the fee level that was given in the respective vignette if it was rated “fair” (between 0 € and 500 €, see black columns in Figure 1). For the remaining 73.9%, we took the fee level that the respondent indicated as a fair fee level for that situation. The resulting variable fair fee estimate varies between 0 € and 5,000 €. Seven estimates are clear outliers at 1,400–5,000 € and have been deleted from the analysis. We included a technical control variable that flags whether the respondent rated the original fee level as fair or whether she entered a new value. Figure 1 shows the resulting dependent variable and highlights the number of ratings at 0 €, 100 €, 200 €, 300 €, 400 € and 500 € that were accepted as fair.

Figure 1 Distribution of the dependent variable fair fee. Note: The grey columns are frequencies of the dependent variable “fair fee”. The black columns show how many given fees were considered fair.

The vignette has eight dimensions: (1) the fee, as discussed above (0 €, 100 €, 200 €, 300 €, 400 € and 500 €);Footnote ³ (2) the household situation (child lives with single mother, child lives with single father, child lives with both parents); (3) the mother’s involvement with the formal labor market (looking for a job, housewife, working part time, working full time); (4) the father’s involvement with the formal labour market (looking for a job, househusband, working part time, working full time); (5) familial support (grandparents can help out in everyday life, grandparents cannot help out in everyday life); (6) net household income [for a single-parent household: 750 €, 1,000 € (roughly the median in Germany), 1,250 €, 1,500 €, 2,000 € (75th percentile), 5,000 € (95th percentile): for a two-parent household multiply by 1.4]; (7) local roots (grew up in Konstanz, moved to Konstanz from abroad, moved to Konstanz from elsewhere in Germany); and finally, (8) the religious affiliation of the family (Christian, Muslim, no religious affiliation). The vignettes do not vary the rank of the child within the sequence of siblings (i.e. we only focus on a first-born child), the age of the child (we only concentrate on 2-year olds) and the type of the organisational structure of the kindergarten (city-run, church-run, run by a social organisation or fully private). We did not capture any characteristics of the child. This means that if respondents wanted to adjust the fee level according to the characteristics of the child, they could not do so and had to rate all vignettes in the same way.

In order to control for the respondents’ personal characteristics, we included a number of respondent-level variables: gender, age in years, education (an ordinal variable of highest qualification obtained), being a German citizen by birth, the recorded number of individuals living in the respondent’s household, whether there is a kindergarten-aged child in the household, and religious affiliation (Christian, other, no religious affiliation). The descriptive statistics can be found with the regression estimates in Table 1.

Table 1 Random-intercept models of the dependent variable fair fee

Note: Standard errors in brackets.

Additional models show that the effect of income of the vignette family is linear, as modelled here.

p-Value of a two-sided test of the null that β is equal to 0.

*p<0.05, **p<0.01, ***p<0.001.

Empirical analysis

The average level of fees which respondents considered to be fair is 191 €; about 2,300 € per year. This is, surprisingly, slightly below the actual fee level at the time of the survey. This evidence strongly suggests that the public in Konstanz accepts a positive, substantial fee for childcare services. It clearly rejects Hypothesis 1a and supports Hypothesis 1b, namely that residents accept a fee-based system at the expense of the actual service-takers. The accepted mean level is not higher than the actual level, dismissing notions of acceptance of higher fees that were in place in Switzerland, just across the border.

This is the first piece of evidence for self-reinforcing policy feedback, since citizens’ perceptions are close to the actual status quo. However, this feedback effect is self-reinforcing only in the sense that German citizens have come to accept paying fees for childcare, even though paying fees for social services is otherwise rather uncommon in the German welfare state regime. The acceptance of a significant level of fees in the case of childcare might be related to the fact that in the public’s perception, attending kindergarten is more than just education; it also allows parents to pursue work, in addition to providing children with educational possibilities. For that reason, the public may be more willing to accept childcare fees since they are balanced out by enhanced employment opportunities. Furthermore, the expansion of opportunities in early childcare is a recent phenomenon; citizens might be more willing to support the expansion of new services with additional discretionary spending on their part, implicitly acknowledging the limitations on general welfare spending.

The standard deviation in the “perceived fair fee level” is relatively high at 159 €, another piece of evidence against the acceptance of a flat-fee system. This heterogeneity in acceptable fee levels is related to varying vignette traits, although not all of the vignette dimensions produce significant results. In the following statistical analysis, we correct for multiple ratings from the same respondent by using a multi-level random-intercept model. We start with the empty null model (Model 1), which just contains the estimate of an intercept for every respondent as well as a (technical) dummy variable indicating whether the fee level is provided by the respondent herself or is perceived as fair (see discussion above). Then, we estimate the model with all vignette traits (Model 2), followed by Model 3 with vignette traits, respondent characteristics and additional contextual information from the city areas where respondents live.

Model 1 reveals an intraclass correlation coefficient of 22.0% in a theoretically empty model. This means that the ratings per individual are moderately similar to one another when we do not control for any particular characteristics of the respondents’ or the vignette traits. The positive and significant coefficient 21.35 of the technical dummy variable indicates that when the fee estimate is provided by the respondent herself, the perceived fair fee level is slightly higher compared to the situation when it was first given in the vignette and then perceived as fair.

In Model 2, the vignette variables are added. Including vignette traits significantly increased the goodness of fit: The R ² of within-variation jumps to 34.3% of explained variance; the intraclass correlation coefficient increases to 34.2%, meaning that respondent answers are more similar when we take into consideration the large variance in ratings between vignettes. The core finding of the analysis is the strong support for a fee structure, which takes into account monetary or other resources in assessing the level of fair fees. We can thus reject Hypothesis 2a (vignette traits do not matter because all children/families are perceived as equal in terms of fees) and give more credence to Hypothesis 2b. Residents causally react in their ratings to the traits of the families described.

By far the strongest associations within the vignettes are related to parental income. Model 2 shows that an increase in (hypothetical) household income by 1,000 € is associated with an increase of 53 € in perceived fair monthly fees for childcare. This is strong support for Hypothesis 3.

The results also show that respondents take into account other resources besides income in their assessment of a fair fee level. Being a single parent is associated with a “discount” of 33 € in the perceived level of fair fees, compared to two-parent households, supporting Hypothesis 4. Alternatively, having “grandparents available for help” has an effect as well: In this case, respondents would consider an additional fee of 9 € per month as fair, supporting Hypothesis 5. This suggests a broader consideration of resources among Konstanz residents which goes beyond just monetary resources. Based on fairness perceptions, single parenthood qualifies for lower childcare fees, whereas families with familial support from grandparents are expected to pay more. Comparatively speaking, the latter top-up on fees is significantly smaller than the “discount” for single parents, however.

Surprisingly, the employment status of the mother does not matter. In fact, an F-test shows that the three variables for the employment status of the mother (capturing working part-time, working full-time looking for a job and homemaker) can be excluded from the model without a significant drop in the statistical fit of the model.

The finding is slightly different for the employment status of the father. These variables (father homemaker, father job-seeker, father working part-time, father working full-time) cannot be dismissed based of the result of an F-test. In the eyes of the respondents, fathers who are unemployed and looking for a job warrant a fee for their child that is 13 € lower than fathers who work full time. Thus, there is some evidence that respondents are willing to support a “discount” of fees, when the (presumably) male breadwinner’s employment opportunities are compromised. In sum, however, the composition of earners in the vignette family does not impact the ratings, thus rejecting Hypothesis 6.

The last set of variables can be dismissed as lacking any value for explaining statistical results. Whether the family in the vignette comes from Germany, from Konstanz or from abroad, does not matter. The religious affiliation of the child makes no difference either. The F-test suggests that we could drop these four variables without comprising the fit of the model.Footnote ⁴ From a political perspective, these nonfindings are positive news, since they indicate that the residents of Konstanz do not discriminate against foreigners, new residents or people with a non-Christian religious affiliation by demanding higher levels of childcare fees. Cultural predispositions do not have any causal impact, thus rejecting Hypothesis 7.

Model 3 introduces respondent-level and city area characteristics into the analysis. The increase in the overall R ² between models 2 and 3 is a mere 1.2 percentage points, due to respondent-level characteristics (recall that they are uncorrelated with the vignette traits). The increase from models 1 to 2 is 21.3 percentage points – clearly much higher. Thus, we can reject Hypothesis 8a and accept Hypothesis 8b: vignette traits matter much more than respondent variables, putting an emphasis on normative rather than resource-based feedback effects.

Going into further detail, we do not find many significant associations. What is remarkable is that the respondents’ income does not affect the level of fair fees, even though income is the most important vignette dimension related to fairness perceptions. This indicates a broad, cross-class consensus on the fairness of the income-dependent fee model: rich individuals are not opposed to this model, even though they would have to pay higher fees and would be disadvantaged compared to the current status quo.

Before we turn to an exploratory interaction analysis, Table 2 summarises all of our hypotheses and their evidence.

Table 2 Overview of hypotheses and the empirical results

The exploratory analysis focuses first on the interaction between income as a respondent characteristic and (the fictitious) income as a vignette trait. A series of F-tests (see Appendix D) confirms that the association between income in the vignettes is moderated by the respondents’ household income. Surprisingly, the direction of the interaction is positive: rich respondents consider even higher fees to be fair compared to residents in low-income families.Footnote ⁵ The magnitude of this effect is not huge, but it still indicates that richer Konstanz residents perceive higher levels of fees as fair compared to residents with lower incomes. At first sight, this seems to run counter to material self-interest, and it could indeed indicate altruistic behaviour. However, it may also be the case that richer citizens support and accept higher fees in order to promote quality in childcare facilities, even if this implies a higher degree of selectivity when accessing those facilities. Potentially, richer citizens might also work with different frames of reference, i.e. the rich and poor might differ in their subjective perceptions of the same fee level. Unfortunately, our data cannot tell us which of these explanations holds.

With regard to other respondent characteristics, we find that respondents who are older, live in larger households and whose religious affiliation is “other than Christian” or “nonbeliever” perceive lower childcare fees as fair. The presence of a small child in the household of the respondent does not make a difference to the perception of fairness, neither does gender. For age, we surprisingly estimate a negative coefficient that implies that older individuals are more likely to perceive lower fees as fair. This could indicate that the elderly are indeed more supportive of ensuring affordable childcare for young families. Alternatively, it could be the case that the youngest age group in the study contains a lot of individuals who do not have children of their own who do not think of Konstanz as the place to raise a family as they are students or recent graduates likely to move away soon.

The residential area variables do not yield any significant coefficients, indicating that the perceptions of fair childcare fees are not contingent on where respondents live in the city: neither purchasing power (between 14,200 € and 27,600 €), the percentage of foreigners (between 4.9 and 26.7%), the relative proportion of children in public childcare between birth and three years (between 0.0 and 79.2%) nor population density (between 4.5 and 100.1 residents per acre of habitable area) are related to perceptions of fair childcare fees.