Introduction
The transformations associated with the shift to a post-industrial society gave rise to a set of new social risks (NSRs) more heterogeneous than those generated in industrial economies, and hence more difficult to compensate for via social insurance (Bonoli, Reference Bonoli2005; Esping-Andersen, Reference Esping-Andersen2002; Armingeon and Bonoli, Reference Armingeon and Bonoli2007). Instead, NSRs are tackled by a measure of social investment, such as active labour market policies and childcare services (Morel and Palier, Reference Morel and Palier2011). Social investment policies are thus considered important instruments to confront the risks generated by post-industrial societies, and the slow development of these services is puzzling (Bouget et al., Reference Bouget, Frazer, Marlier, Sabato and Vanhercke2015).
This research adopts a policy diffusion perspective and proposes a political learning mechanism through which the consequences of the politics of social investment can both facilitate and hamper the development of these policies. In this framework, policymakers are assumed to be Bayesian learners whose beliefs about the consequences of a policy adoption are updated when new information is gathered (Braun and Gilardi, Reference Braun and Gilardi2006; Tversky and Kahneman, Reference Tversky and Kahneman1974). It is argued that information about the political success or failure of a policy decision in other jurisdictions is expected to affect policymakers’ judgments about the political feasibility of a given policy proposal altering their policy choice (May, Reference May1992).
The central argument is developed around the electorally uncertain character of the politics of social investment expansion. Social investment policies are not supported equally by all individuals in a polity. NSR groups are more likely to prefer an emphasis on expenditures that allow them to participate in the labour market, while already integrated individuals, such as production workers with specific skills, are more likely to prefer compensation for ‘old risks’ (Häusermann, Reference Häusermann2010; Häusermann et al., Reference Häusermann, Kurer and Schwander2014; Häusermann et al., Reference Häusermann, Pinggera, Ares and Enggist2019). Overall, people are more likely to oppose the expansion of benefits from which they do not directly benefit (Busemeyer et al., Reference Busemeyer, Garritzmann and Neimanns2015; Busemeyer and Neimanns, Reference Busemeyer and Neimanns2017; Busemeyer and Garritzmann, Reference Busemeyer and Garritzmann2017; Chung and Meuleman, Reference Chung and Meuleman2017).
Even though individuals belonging to different social groups have, on average, different preferences for social policy, their party preferences are not always as different (Häusermann and Kriesi, Reference Häusermann and Kriesi2015). In this study, it is argued that incumbents whose electorate is divided on which policy to prioritise face a political dilemma. Indeed, in times of austerity, part of the electorate of governing parties may perceive the expansion of social investment as a shift of resources from the protection for old risks and translate this into a loss of support. Thus, uncertain of the political outcome of their policy choice, the decisions of parties in government may be informed by the past experience of their foreign counterparts in an attempt to learn from their electoral consequences and reduce political risks.
To test this argument, I estimate the association between the domestic and foreign levels of childcare expenditure, where the past foreign policy decisions are weighted by their closest electoral outcomes. The policy change under study is the level of expenditure in childcare services, as family policy is a key element of the social investment approach, which emphasises investments in children’s social and cognitive skills (Esping-Andersen, Reference Esping-Andersen2002) and removes a major obstacle to mothers’ labour market integration (Lewis et al., Reference Lewis, Knijn, Martin and Ostner2008; Hook, Reference Hook2015; Esping-Andersen, Reference Esping-Andersen2009).
The electoral relevance of childcare services and other work-friendly family policies has been growing together with the declining importance of the core constituencies of mainstream parties, generally more in favour of male-breadwinner social insurance policies, such as industrial workers for the parties of the Left and religious voters for centre parties (Schwander et al., 2018). Promoting social investment, these parties attempt to attract new voters, such as women and young highly educated professionals (Morgan, Reference Morgan2013; Karreth et al., Reference Karreth, Polk and Allen2013; Häusermann, Reference Häusermann2018; Schwander and Häusermann, Reference Schwander and Häusermann2013). Under different circumstances, these electoral strategies have proven able to produce both electoral rewards and sanctions (Abou-Chadi and Wagner, Reference Abou-Chadi and Wagner2019; Nelson and Giger, Reference Nelson and Giger2019; Neimanns, Reference Neimanns2021).
The organisation of the paper will start with a section outlining the theoretical mechanism through which political learning is expected to influence an incumbent policy choice. This is followed by a section on the politics of social investment that contains an illustration of the electoral dilemma faced by governments with a divided electorate and formalises testable hypotheses on the effect of political learning. The subsequent sections describe the research design and empirical strategy, present and discuss the findings, and the final section concludes.
Learning from Foreign Electoral Consequences: A Mechanism of Policy Diffusion
Policy diffusion occurs when government policy decisions in one country are systematically conditioned by prior policy choices made in other countries (Simmons et al., Reference Simmons, Dobbin and Garrett2006). There is a widespread consensus on three broad classes of diffusion mechanisms: learning, emulation, and competition (Braun and Gilardi, Reference Braun and Gilardi2006; Simmons et al., Reference Simmons, Dobbin and Garrett2006; Shipan and Volden, Reference Shipan and Volden2008; Graham et al., Reference Graham, Shipan and Volden2013; Gilardi, Reference Gilardi2012). In this study, learning from the consequences of other units is proposed as a mechanism to explain the slow development of social investment.
Learning can be defined as the process of acquisition of new relevant information that permits the updating of beliefs about the effects of a new policy (Meseguer, Reference Meseguer2004; Meseguer, Reference Meseguer2005; Meseguer, Reference Meseguer2003; Braun and Gilardi, Reference Braun and Gilardi2006). The learning process is best described in terms of Bayesian updating. Individuals update their beliefs by looking at the experiences of others, either rationally or relying on ‘cognitive shortcuts’ such as representativeness, availability and anchoring (Tversky and Kahneman, Reference Tversky and Kahneman1974; McDermott, Reference McDermott2001; Weyland, Reference Weyland2012; Weyland, Reference Weyland2009).
The acquisition of new relevant information can lead to two forms of learning. Policy learning occurs when the information acquired concerns the effectiveness of a policy, i.e. when it achieves what it is designed to do (Meseguer, Reference Meseguer2003). Political learning occurs when the information acquired concerns the payoffs associated with policies, i.e. their electoral rewards and their closeness to the policymaker ideal point (Braun and Gilardi, Reference Braun and Gilardi2006, p. 301).
The focus of this paper is on political learning, i.e. ‘judgments about the political feasibility of policy proposals and understandings of the policy process within a given policy domain’ (May, Reference May1992, p. 339). Examples of political learning are political actors adapting their political strategies to advance their political agendas without suffering severe electoral costs (Pierson, Reference Pierson1994), e.g. proposing fewer radical reforms and negotiating with interest groups to achieve reforms (Natali, Reference Natali2002).
The scholarship on policy diffusion is regarded as the field of study that has done the most to link theoretical models of learning to empirical analyses (Dunlop and Radaelli, Reference Dunlop and Radaelli2013, p. 600), developing models that included both mechanisms of policy and political learning (Braun and Gilardi, Reference Braun and Gilardi2006; Gilardi and Wasserfallen, Reference Gilardi and Wasserfallen2019). Empirically, this research showed the importance of political learning as a mechanism influencing many policy developments. Policy diffusion is often conditional on ideological ground, as political/ideological proximity between units facilitates diffusion (Volden, Reference Volden2006; Butler et al., Reference Butler, Volden, Dynes and Shor2017), and public opinion support (Pacheco, Reference Pacheco2012; Pacheco and Maltby, Reference Pacheco and Maltby2017; Abel, Reference Abel2019). Governments also react to the electoral consequences of the prior policy choices of other countries (Gilardi, Reference Gilardi2010) and tend to reframe diffusing laws in more electorally acceptable forms (Mallinson and Hannah, Reference Mallinson and Hannah2020).
Research in party politics has shown that political learning also occurs between parties, which strategically adjust their ideological orientations according to the shifts of their counterparts that have recently governed (Williams, Reference Williams2015; Williams and Whitten, Reference Williams and Whitten2015). Parties also adjust their positions according to shifts in public opinion when the direction of change is clearly disadvantageous for the party (Adams et al., Reference Adams, Clark, Ezrow and Glasgow2004). A growing amount of research in the intersection between party politics and policy diffusion is collecting evidence that this phenomenon takes place also cross-nationally (Böhmelt et al., Reference Böhmelt, Ezrow, Lehrer, Schleiter and Ward2017; Böhmelt et al., Reference Böhmelt, Ezrow, Lehrer and Ward2016; Schleiter et al., Reference Schleiter, Böhmelt, Ezrow and Lehrer2021; Adam and Ftergioti, Reference Adam and Ftergioti2019). Ideological shifts, however, do not come without a cost. Voters update their party support accordingly and hold governing parties accountable for their policy outputs (Adams et al., Reference Adams, Bernardi and Wlezien2020; Bernardi and Adams, Reference Bernardi and Adams2019; Adams et al., Reference Adams, Clark, Ezrow and Glasgow2006).
Figure 1 contains a stylised model of political learning via foreign electoral consequences. Let i be a country in which the government must decide how to allocate the yearly budget for social expenditure, let j be a country that recently had a national election, and let the government of the two countries have similar constituencies. The way government in country i will allocate social expenditure will depend on a number of domestic factors, such as its ideological orientation, the existence of a specific problem pressure, and the availability of sufficient resources. There will also be, however, non-domestic factors affecting the government decisions via policy or political learning. One form of political learning occurs when the government in country i observes the outcome of the election in country j (dashed line) and updates its perception of adequacy of a given policy choice accordingly. In this way, the expenditure decision in country i is affected by the electoral outcome of country j (dashed arrow). More specifically, if the incumbent in country j wins the election, the government in country i will be more likely to make similar policy decisions; if the government in country j loses the election the government in country i will be less likely to make similar policy decisions.
FIGURE 1. Model of political learning via electoral consequences
The model of political learning in figure 1 relies on the assumption that the learning government is uncertain of the optimal course of action to undertake. This is not, however, always the case. The next section elaborates on the reasons why social investment expansion may be considered a source of uncertainty and the conditions under which such uncertainty may occur.
The Uncertain Politics of Social Investment
Unconstrained public opinion support for social investment is broad (Garritzmann et al., Reference Garritzmann, Busemeyer and Neimanns2018) but is significantly reduced when its expansion has to be financed through cuts in existing entitlements (Busemeyer et al., Reference Busemeyer, Garritzmann and Neimanns2015; Busemeyer and Garritzmann, Reference Busemeyer and Garritzmann2017). In times of austerity, the expansion of some programs tends to occur at the expense of others (Häusermann, Reference Häusermann2010), and social insurance programs compete with needs-based benefits (Palier, Reference Palier2010; Palier and Thelen, Reference Palier and Thelen2010; Emmenegger et al., Reference Emmenegger, Häusermann, Palier and Seeleib-Kaiser2012; Kim and Choi, Reference Kim and Choi2020).
As a consequence, the political conflict in advanced political economies is characterised by friction between individual preferences for policies that distribute benefits on the basis of contributions and those that do so on the basis of need (Beramendi et al., Reference Beramendi, Häusermann, Kitschelt and Kriesi2015), and preferences between expenditure that generates immediate consumption for the insured and expenditure whose returns are diffused and delayed, such as active labour market policies, investments in human capital and child and elderly care services (Häusermann and Kriesi, Reference Häusermann and Kriesi2015; Rueda, Reference Rueda2005; Schwander and Häusermann, Reference Schwander and Häusermann2013; Van Lancker, Reference Van Lancker2013).
Individuals with different preferences for social policy, however, do not necessarily vote for different parties (Häusermann and Kriesi, Reference Häusermann and Kriesi2015). On the one hand, part of the electorate of the Left opposes increasing spending on social investment if it perceives it as a threat to existing entitlements (Abou-Chadi and Wagner, Reference Abou-Chadi and Wagner2019). On the other hand, the same parties are rewarded for the expansion of certain social investment policies, i.e. childcare (Nelson and Giger, Reference Nelson and Giger2019), but only from individuals further up the income distribution, i.e. those more likely to make use of them (Neimanns, Reference Neimanns2021).
I argue that a divided electorate and the unpredictability of electoral returns of a social investment expansion generate uncertainty in the decision-making process of policymakers lacking complete information about what is the optimal course of action to undertake. To overcome information deficits, governments may look at the experiences of others and learn from their consequences (Braun and Gilardi, Reference Braun and Gilardi2006; Meseguer, Reference Meseguer2004).
Therefore, when cabinets allocate more resources to childcare expenditure and are rewarded in subsequent elections, the other countries’ expenditure levels should increase (H1). Similarly, when cabinets allocate more resources to childcare expenditure and are sanctioned in subsequent elections, the other countries’ expenditure levels should decrease (H2).
Political learning, however, should occur only in situations of uncertainty. To be valid, the hypotheses above should fulfil two conditions: first, there must be a divided electorate, i.e. groups of individuals prioritising, on average, different social expenditures, and voting, on average, for the same party; second, the divided electorate should be the electorate of the governing parties.
Thus, as a first scope condition (SC1), the effect should be significant only in countries in which the size of groups more in favour of social insurance is declining in favour of groups with a preference for social investment (Schwander et al., 2018). Low-skilled workers with specific skills are especially likely to oppose investment-oriented policies (Häusermann, Reference Häusermann2010; Häusermann et al., Reference Häusermann, Kurer and Schwander2014) while high-skilled and female labour market outsiders strongly favour social investment policies (Häusermann et al., Reference Häusermann, Kurer and Schwander2014). The welfare priority of these groups is even stronger when the relative importance of social insurance or social investment is included in the picture: Production workers prioritise social insurance and social consumption more than every other category, while high skilled socio-cultural professionals are the most positive towards social investment (Häusermann et al., Reference Häusermann, Pinggera, Ares and Enggist2019). Therefore, the effect of the foreign political consequences should be stronger in countries where the ratio between socio-cultural professionals and production workers is skewed in favour of socio-cultural professionals and weaker where the ratio is skewed in favour of production workers (SC1).
A second scope condition is that the governing party or coalition should have a divided electorate. The majority of social democratic parties balance their programmatic offer between social investment and protection for old risks (Häusermann et al., Reference Häusermann, Pinggera, Ares and Enggist2019), arguably reflecting the evidence that socio-cultural professionals and production workers do not significantly differ in their probability of voting parties on the Left (Häusermann and Kriesi, Reference Häusermann and Kriesi2015). Indeed, an expansion of childcare services increases the electoral support for the Left only in the part of its electorate with higher income, generally socio-cultural professionals (Neimanns, Reference Neimanns2021). This implies that the electorate of the parties of the Left is the most divided on these policy preferences. Conversely, the electorate of Christian democratic and Centre-right parties is also changing as the religious voters are losing importance, and these parties may compete for centrist voters embracing more progressive views on childcare, especially when a radical-right competitor is missing (Schwander et al., 2018; Fleckenstein, Reference Fleckenstein2011). In contrast, far-right parties are not expected to shift their positions on childcare to follow public opinion shifts, as they are generally sanctioned by a more resilient electorate (Adams et al., Reference Adams, Clark, Ezrow and Glasgow2006). Therefore, while all parties on the ideological spectrum but the far-right may be to an extent susceptible to the foreign electoral consequences of childcare expansion, the effect should be stronger for the parties on the Left (SC2).
Data and Methods
To test the hypotheses stated in the previous section, I model the diffusion effects with the inclusion of spatial lags in the model specification, relying on a longitudinal dataset with observations for 23 OECD countriesFootnote 1 from 1999 to 2013. The dependent variable is the total public and mandatory private social expenditure for childcare and early education services as a percentage of GDP (OECD, 2018b). The restriction of the sample size to 23 countries is due to data availability of the same countries in the European Social Survey Cumulative File (ESS, 2018), and the time frame is given by data availability for all 23 countries at the time of analysis.
Data on election dates, election outcomes and cabinet composition for the definition of the weighting matrices are taken from Döring and Manow (Reference Döring and Manow2019). Government positions on the ideological dimension are measured on a left-right scale from 1 to 10. It is calculated by averaging the position of the cabinet parties, weighted by the number of seats occupied, using expert survey data (Döring and Manow, Reference Döring and Manow2019). I use ESS (2018) waves 1-8 to measure the size of production workers (PW) and socio-cultural professionals (SCP) according to Oesch’s (2006) class schema. Each individual is assigned to one category according to their ISCO-code and then categories are aggregated by country and wave. Since the waves are published every two years and the survey is conducted in the years in between, the values of a wave done at year t are imputed to the years t-1, t-2… t-n if values are missing. For example, the values of the first wave that took place in 2002 are assumed as valid also for 1999, 2000 and 2001. The ratio between socio-cultural professionals and production workers (SCP-to-PW ratio) is calculated dividing the share of socio-cultural professionals by the share of production workers.
The model includes several variables that are expected to influence the outcome of interest (see Table A1 for an overview of the variables included in the model).
Theoretically, the level of childcare expenditure in a year is likely to depend on the level of the previous year, thus I include the lagged value of the dependent variable. Methodologically, whether the lagged dependent variable (LDV) should be included or not in the model specification, even when part of the data generating process, is an open debate. While LDV models are known to produce biases in the estimation of coefficients (Achen, Reference Achen2000; Plümper et al., Reference Plümper, Troeger and Manow2005), it has been argued that there is nothing pernicious in using it when the time dimension is large enough
$\left( {T \ge 15} \right)$
(Beck and Katz, Reference Beck and Katz2011). Furthermore, the omission of LDV would itself produce severe omitted variable bias (Wilkins, Reference Wilkins2018), especially in the spatial specification where the LDV controls for important common trends between cross-sections that may lead to spatial patterns (Plümper and Neumayer, Reference Plümper and Neumayer2010).
Childcare expenditure levels are expected to depend on the amount of resources available in a country; to control for economic performances I include the log of the GDP per capita (World Bank, 2019b) and real GDP growth (OECD, 2018a). To control for the sociodemographic demand of childcare, the models include the total fertility rate (World Bank, 2019c) and the female labour force participation rate (World Bank, 2019a). A larger share of women in the legislative chamber (Armingeon et al., Reference Armingeon, Wenger, Wiedemeier, Isler, Knöpfel, Weisstanner and Engler2019) may also be associated with higher levels of expenditure, as female MPs are assumed to be more responsive to working women interests (Bonoli and Reber, Reference Bonoli and Reber2010). I include the level of government expenditure in social security transfers as a percentage of GDP (OECD, 2018b) as it may negatively affect childcare expenditure due to the crowding out effect (Bonoli and Reber, Reference Bonoli and Reber2010; Kim and Choi, Reference Kim and Choi2020). I control for the government’s ideology as parties on the Left are expected to be associated with higher levels of expenditure that promotes the participation of women in the labour market (Korpi, Reference Korpi2000). I include a spatial lag of geographic proximity as neighbouring countries represented in national media are known to affect family policy development (Linos, Reference Linos2013) and to control for other exogenous shocks and common trends (Plümper and Neumayer, Reference Plümper and Neumayer2010). Finally, I include country dummies to control for all the country-specific time-invariant unobserved heterogeneity that may be associated with changing expenditure.
Model specification
I estimate a series of dynamic spatial autoregressive models (Franzese and Hays, Reference Franzese and Hays2007, Reference Franzese and Hays2008), which allow for dependence between observations due to strategic interaction as a consequence of political learning. In this analysis, the level of expenditure of one country at time t is modelled as a function of the average level of expenditure on other countries at
$t - 1$
, weighed by different attributes discussed in the following section.
In equation 1, a weighting matrix specifies the set of countries and the relevant linkages between the countries. Accordingly, the spatial lag model is defined as:
$${y_t} = \varphi {y_{t - 1}} + \;\beta {X_{t - 1}} + \rho W{y_{t - 1}} + \varepsilon $$
Where
${y_t}$
is the dependent variable (childcare expenditure at time t),
${y_{t - 1}}$
signifies the temporally lagged dependent variable (childcare expenditure at time
$t - 1$
),
${X_{t - 1}}$
is a matrix of temporally lagged explanatory variables.
$W{y_{t - 1}}$
stands for the product of the connectivity matrix W and the lagged value of the dependent variable
${y_{t - 1}}$
, i.e.
$W{y_{t - 1}}$
is the spatial lag and
$\rho $
is the corresponding spatial autoregressive coefficient.
In panel analysis, the connectivity matrix W, is given by a
$NT \times NT$
matrix with
$T\;\left( {N \times N} \right)$
sub-matrices along the block-diagonal, with an element
${w_{ij}}$
capturing the relative connectivity of unit (country) j to unit (country) i and with the diagonal elements
${w_{ii}} = 0$
. A common way to define the spatial lag is to use the temporally lagged values of the dependent variable (Gleditsch and Ward, Reference Gleditsch and Ward2008).
A common estimator is the ordinary least square (OLS) regression (Böhmelt et al., Reference Böhmelt, Ezrow, Lehrer and Ward2016; Williams, Reference Williams2015; Williams and Whitten, Reference Williams and Whitten2015). However, the spatial lag in OLS introduces endogeneity in the specification, and maximum likelihood (ML) and two-stage least square (2SLS) have been shown to provide superior estimates and more accurate standard errors (Franzese and Hays, Reference Franzese and Hays2007; Plümper and Neumayer, Reference Plümper and Neumayer2010). As suggested by Anselin et al. (Reference Anselin, Le Gallo and Jayet2008), ML may also be a way to deal with the endogeneity induced by the LDV (Elhorst, Reference Elhorst2014).
Operationalisation of country linkages
For the operationalisation of country interdependencies, I rely on four distinct weighting matrices. Matrix
${W_{won}}$
captures the interdependencies between the government in country i and the incumbent in country j, and the strength of the relationship is given by the share of parties in the incumbent coalition who were also part of the newly appointed cabinet after the elections. Thus, the government in country i gives more weight to the decisions of its successful counterparts.
${W_{lost}}$
captures the interdependency between the government in country i and the incumbent in country j and the strength of the relationship is given by the share of parties of the incumbent coalition that were not part of the newly-appointed cabinet after the elections. Thus, the government in country i gives more weight to the decisions of its unsuccessful counterparts.
${W_{ideo}}\;$
captures the ideological distance between the government in country i and the government in country j – that is to say, the government in country i gives more weight to decisions made by more ideologically similar counterparts. Finally,
${W_{geo}}\;$
captures the geographical distance between country i and j, so the government in country i weights more decisions made by governments in proximate countries.
Each element
${w_{ij}}$
of the underlying connectivity matrix
${W_{won}}$
receives a value
$0 \le {w_{ij}} \le 1$
, equal to the share of parties of an incumbent cabinet coalition in country j that were part of the cabinet coalition that followed the elections. Each element
${w_{ij}}$
of
${W_{lost}}$
receive a value
$0 \le {w_{ij}} \le 1$
, equal to the share of parties of the incumbent cabinet coalitions that were not part of the cabinet coalition that followed the elections. Each element
${w_{ij}}$
of
${W_{ideo}}$
receives a value
$1 \le {w_{ij}} \le 10$
, equal to the absolute difference between the values of the left-right positions of the governments in country i and j. Finally, each element
${w_{ij}}$
of
${W_{geo}}$
takes a value equal to the inverse distance between the capitals of country i and j.
It is common practice to row standardize the proximity matrices to facilitate the interpretation of results, removing dependence on the scale factor and avoiding singularity, so that each row sums up to 1. Row-standardization generates spatial lags that are a weighted average of the values of the dependent variable with weights dependent on the existence and strength of a postulated network tie between a pair of cases. The spatial lags
$W{y_{won}},\;W{y_{lost}}\;,\;W{y_{ideo}}$
and
$W{y_{geo}}\;$
are calculated by multiplying the relative weighting matrices with a vector containing the time-lagged value of childcare expenditure, the resulting vector represents the average value of childcare expenditure of the countries in the sample, corrected by the weights specified in the weighting matrices.
Findings
Table 1 contains the estimated parameters. Unsurprisingly, a large part of childcare expenditure depends on the level of expenditure in previous years across all models. Besides that, the strongest predictor in model 1 is the total fertility rate: a one unit increase in the TFR is associated with a rise in expenditure by 0.09 percent of the GDP. The log of GDP per capita has a very small association with the level of childcare expenditure, such that it becomes indistinguishable from zero when further controls are included in the model specification. The level of expenditure for social security transfer as a share of GDP has a consistent significant negative effect on the level of childcare expenditure, in line with the crowding out hypothesis, indicating that resources tend to be either allocated to social insurance or social investment. In most models, the effect of government ideology is not associated significantly with the outcome levels, in line with previous research that found an inconsistent effect of ideology in the development of post-industrial social policy (Bonoli, Reference Bonoli2013; Bonoli and Reber, Reference Bonoli and Reber2010). The coefficient for ideology, however, becomes a significant predictor of childcare expenditure levels when the effect of geographical diffusion of childcare expenditure is controlled for: For a unit increase in the Left-Right dimension, the expenditure for childcare is reduced by 0.004 percent of GDP.
TABLE 1. Spatiotemporal autoregressive models (Maximum likelihood estimates).
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
For what concerns the effect of the size of occupational groups: the larger the share of socio-cultural professionals, the higher the demand for childcare services. A significant portion of socio-cultural professionals are women and younger and highly educated individuals, who generally prefer social investment over social consumption. Each unit percentage increase in the share of socio-cultural professionals is associated with a rise in spending on childcare services by 0.007 percent of GDP. In contrast, the share of production workers does not significantly predict the level of childcare expenditure. Model 2 includes the ratio between SCP and PW instead of their overall shares, and the effect is positive and strong. Each unit increase in the SCP-to-PW ratio increases the expenditure for childcare services by 0.08 percent of GDP.
Model 3 includes the spatial lag for government ideological proximity, i.e. the average level of spending of every other country in the sample, weighted by the ideological distance between governments. The larger the ideological distance between the two governments, the larger the importance given to its value. The coefficient is negative, suggesting that policy decision in one country is followed by a similar decision in a country with an ideologically proximate government. The coefficient in model 3 is not significant, but it becomes significant once the control for geographic proximity is introduced in the specification, suggesting that policy diffusion takes place between countries with ideologically similar cabinets only as far as these countries are also geographically close. An increase of one percentage point in the average expenditure of ideological proximate governments is associated with a rise in expenditure by about 0.12 percent of GDP, indicating a strong diffusion between like-minded governments.
Models 4 and 5 include the indicators for political learning, i.e. the average expenditure levels in countries that recently experienced elections, weighted for the success and failure of incumbent coalitions. The larger the share of parties of the previous coalition that (do not) belong to the newly appointed cabinet, the larger the importance given to their expenditure choices. The coefficient of
$W{y_{won}}$
is positive but not significant while the coefficient for
$W{y_{lost}}$
is negative and significant. This suggests that success stories are not valued as much as failure stories. Such an effect is in line with research on cognitive biases that has consistently shown that negative and positive information is weighted differently, with higher value given to negative information (Kahneman and Tversky, Reference Kahneman and Tversky1979; Ito and Cacioppo, Reference Ito and Cacioppo2005; Rozin and Royzman, Reference Rozin and Royzman2001), as well as with the findings in party politics research showing that political parties adjust their policy orientations according to shifts in public opinion only as long as such shifts are clearly disadvantageous for the party (Adams et al., Reference Adams, Clark, Ezrow and Glasgow2004). Thus, H2 is supported but and H1 is rejected. The main effect on an increase of one percentage point in average spending level of governments that subsequently lost an election is associated with a reduction of spending by 0.04 percent of GDP. The estimated effect does not change once geographical proximity is controlled for.
Model 6 tests the scope conditions under which political learning should occur. Since it is not possible to infer meaningful conditional effects simply by looking at the significance of the interaction term (Brambor et al., Reference Brambor, Clark and Golder2006), in Figure 2 the conditional effects of political learning are plotted.
FIGURE 2. Marginal effects of Wy Lost for different SCP-to-PW ratios and for differentgovernment ideologies, with 95% CINote: Left = 2.5 on the left-right scale; Centre = 5.5 on the left-right scale; Right = 8.5 on the left right scale
Figure 2 contains the average marginal effect of political learning for countries with different ratios between socio-cultural professionals and production workers, and different government orientations. For governments of right-wing and centrist ideologies, the slope of the marginal effects of
$W{y_{lost}}$
is not significantly different from zero. Instead, for governments made by parties of the Left, the effect of political learning on the level of spending is strong and statistically significant: The effect of
$W{y_{lost}}$
is close to zero when the size of production workers is much larger than socio-cultural professionals (values close to zero on the x-axis) and becomes increasingly negative for larger shares of socio-cultural professionals, signalling an increasingly divided electorate.
To sum up, childcare expenditure levels were found to be positively associated with higher fertility rates, larger shares of socio-cultural professionals, the ratio between socio-cultural professionals and production workers, and negatively to the size of the expenditure for social security transfers. Left-wing governments were found to be associated with a higher childcare development when the influences of geographically proximate governments are controlled for. One of these influences is the behaviour of governments that is also ideologically proximate, as it appears that governments of similar ideology tend to make similar decisions with regards to childcare expenditure decisions. The analysis showed that these governments also learn from the negative political consequences of their foreign counterparts that have recently governed: In particular, governments of the Left with a divided electorate tend to reduce their expenditure levels whenever a foreign incumbent that previously did so loses the elections.
Conclusions
Political explanations of social policy development are central in the comparative politics literature on the welfare state, especially with regards to the expansion and retrenchment of benefits. Early theories posited that a large working-class with homogenous preferences for state intervention represented by left-wing parties in government was central to welfare state development. More recently, evidence has shown that the role of politics in welfare state development has changed. Post-industrial societies brought about new social risks and new strategies to cope with them, i.e. social investment policies. These new strategies imply an expansion of instruments that compete with existing entitlements, creating new conflicts, and fragmenting the historical constituencies of political parties.
I have argued in this paper that this situation generates a political dilemma for the parties whose historical core constituencies are losing importance and must attract new voters. In doing so, these parties may have an interest in the expansion of social investment but lack complete information on the political consequences of such a course of action. I hypothesized that as a consequence of such lack of information, governments learn from their foreign counterparts’ decisions and their electoral consequences.
To test the political learning mechanism, I modelled childcare expenditure levels in 23 welfare states using a set of dynamic spatial autoregressive models that allow for interdependencies between units to be included in the regression. I hypothesised that incumbent cabinet coalitions which increased levels of expenditure and were rewarded or sanctioned in a subsequent election should be respectively associated with higher or lower levels of spending in other units. The results of the analysis do not support the hypothesis that higher levels of expenditure of successful cabinets are positively associated with expenditure levels in other units but support the hypothesis of governments learning from the electoral failures of foreign counterparts.
As predicted, political learning significantly affects childcare expenditure only when a relevant political dilemma has the conditions to emerge, i.e. when the importance of groups preferring social consumption – production workers – is declining and that of groups with a preference for social investment – socio-cultural professionals – is growing. The effect is significant only for the parties on the left of the political spectrum, whose electorate is particularly divided between these two groups.
Adopting a policy diffusion framework, this study contributed to the literature on the determinants of the development of social investment investigating it from a novel angle and proposing a link between institutional and political explanations of policy development. Despite being widely advocated at the supranational level and in academia as a productive instrument to protect individuals and families in post-industrial societies from new social risks, social investment policies are developing slowly in Europe. The findings of this study show that such a weak diffusion may be due partly to the uncertainty that parties in the government coalitions have regarding the political consequences implied by a trade-off between new policies and existing entitlements. More broadly, the implications of these findings indicate that political actors may moderate the feedback effect of existing institutions via the perception that they have of possible political consequences, suggesting a channel through which institutional feedback influences the development of new policies. This general conclusion is, however, a plausible claim made on the findings of this study but it was not empirically tested and contains an element of speculation. Future research should explore more carefully the role of political moderation and use more accurate measures of the perception that political actors have of the environment in which they operate.
Supplementary material information: the dataset and scripts for the replication of the analysis will be made available upon request of the editor, the reviewers, or for publication.
Acknowledgements
Earlier versions of this paper were presented at the ESPAnet 2019 in Stockholm and at the 11th Nordwel Summer School in Bremen. I am grateful for the participants’ constructive input. Moreover, I wish to thank Tobias Böger, Julian Garritzmann, Eloisa Harris, Johannes Huinink, Carina Schmitt and two anonymous reviewers for their valuable feedback, and Hanne Gaukel for her helpful assistance. Funded by Deutsche Forschungsgemeinschaft (DFG) Projektnummer 374666841 – SFB 1342.
Competing interests
The author declares none.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0047279421000908
Open access
