The Effect of Constitutional Commitment to Social Security on Social Expenditure Schemes

1 Introduction

In recent decades, politicians and academics have emphasized the role of social rights for providing social and economic development (ILO 2014; Townsend 2007). The main argument for a rights-based approach to development is that it gives an entitlement that can be enforced in court. Without such a right, people are fully dependent on the ‘good-will’ of the government of that time for proper education, health care and social security. In theory, the constitution can play an important role for social rights, as constitutions provide universal rights for everyone and protect minorities against the majority. However, the number of empirical studies on the effect of social rights in the constitution is still very limited. Ben-Bassat and Dahan (2008) were the first to investigate the effects of the rights to social security, education, health, housing and workers’ rights. They find no relation between these rights and expenditure on these different categories, except for the positive relation between the degree of constitutional commitment to social security and transfer payments and between constitutional commitment to health and health policy performance. In a more recent paper, Ben-Bassat and Dahan (2016) find a positive relationship between CCSS and the extent and coverage of actual measures of social security laws. The studies of Ben-Bassat and Dahan (2008, 2016) are the only two studies on the relation between CCSS and spending on social security, which makes more research on this topic a valuable contribution. In addition, Ben-Bassat and Dahan (2008, 2016) only report simple OLS estimates, so that a contribution can be made to address endogeneity problems more thoroughly.

In this paper, we study the effect of constitutional commitment to social security (CCSS) on different kinds of social expenditure.^[1] We define CCSS as a dichotomous variable, being 1 if there is at least a general statement in the constitution on a social right to income, unemployment, sickness, work injury, old age, survivor or disability and 0 if there is no statement on any of these categories. First, we are interested in the effect of CCSS on total social expenditure, which shows whether CCSS has an effect at all. Second, we study if the effect of CCSS is most sizable on social expenditure schemes for beneficiaries who are seen as less deserving by the public opinion. We expect this if the median voter cares less about these social expenditure schemes, leaving a larger role for the constitution.

We run OLS models, 2SLS regression models and the Heckman two step model with the rigidity of the constitution as an instrument to correct for possible endogeneity. In line with our expectations, we find a positive effect of the rigidity of the constitution on CCSS. First, this is in accordance with Landes and Posner (1975), who argue that the discounted value of the constitution is larger when the constitution is more durable. Second, uncertainty increases over time and thereby the risk that politicians themselves, but also the median voter, or their offspring become dependent on social security is larger for the distant future rather than the near. These two reasons show an added value of the constitution that supplements laws and policies.

We use a panel data set for 17 EU-countries from 1990 until 2012. The data on social expenditures as a % of GDP are taken from the Social Expenditure Database (SOCX) of the OECD. For CCSS, we use the indicator created by Ben-Bassat and Dahan (2008), which we defined as one or zero, depending on the presence of a legal provision on assistance to old age, survivors, disability, unemployment, sickness, work injury or the poor in the constitution.

Our main findings are as follows. First, we find a positive significant effect of CCSS on total social expenditure. This includes a positive effect on spending on old age and survivor, incapacity, unemployment and active labor market policies. Second, the most sizable effects, expressed as a percentage of average spending, are found for spending on unemployment and active labor market policies. These are the expenditure schemes targeted at people who are perceived as less deserving by the public opinion. Thirdly, no positive effect is found on expenditure on health and family, which are not covered in CCSS. This suggests that the positive effect we observe for social security types of social expenditure is due to CCSS and not caused by a positive attitude toward redistribution.

We make the following contributions to the literature. First, knowing if there is an effect of CCSS on total social expenditure contributes to research on the popular rights-based approach, as it tells us if social rights in the constitution contribute to social security. Second, studying the effects on different kinds of social expenditure allows us to explain if CCSS has the strongest effect on social expenditure targeted at people who are perceived as less deserving by the public opinion. Third, an important contribution is how we correct for possible endogeneity in our empirical methodology. We select similar EU-countries and correct for the endogeneity problem by using both 2SLS models and the Heckman two step model. We use the rigidity of the constitution as an instrument in order to derive the causal effect of CCSS on social expenditure. This contributes to the current literature, which does not go beyond correlations inferred from OLS models for a sample with a wide variety of countries, which we deem as insufficient to deal with the endogeneity issue as well as to draw conclusions for the more homogeneous group of EU-countries.

The outline of the paper is as follows. Section 4.2 gives a literature review, in which we start with a theoretical framework on the effects of CCSS in Section 4.2.1 and proceed with the related empirical literature in Section 4.2.2. We continue with describing the methodology with an elaborate discussion on the causes of endogeneity and the methodological solutions to deal with this in Section 4.3.1. We give the empirical specification in Section 4.3.2. We discuss the data in Section 4.4 and the results in Section 4.5. We conclude with a discussion on the implications of the results in Section 4.6.

2 Constitutional Rights and Social Security

3 Methodology

4 Data

We use a pooled cross sectional data set for 17 EU-countries covering 23 years from 1990 to 2012.^[7] The small number of countries is a limitation of this study, but we choose to focus on EU-countries that are represented in the OECD for the reason of data availability and because there is less heterogeneity between these developed countries, making cross-country comparison more reliable. The countries Estonia, Latvia, Luxembourg and Slovenia were removed from the database, because both the index for CCSS and the index for the rigidity of the constitution are not available for these countries (Ben-Bassat and Dahan 2008; Lorenz 2005). This makes our selected countries even more comparable with regard to GDP per capita, geographical location and being consolidated democracies, reducing the risk that third factors obscure our results. We focus on the period from 1990 onwards, making the data set highly balanced, as this enables us to take the post-Soviet countries into account; a substantial share of the data is missing for these countries for the period before 1990.

We choose to use the CCSS indicator created by Ben-Bassat and Dahan (2008, 2016) as we consider these data of higher quality than the data sets from the Toronto Initiative for Economic and Social Rights (TIESR) (Jung and Rosevear 2011) and the data set from the Comparative Constitutions Project (CCP) (Elkins et al. 2014). First, because Ben Bassat and Dahan studied the various constitutions thoroughly to code both different types of commitment to social security and different degrees of commitment. The other data sets only capture a dummy if there is a social right to social security in the constitutions.^[8] Looking more carefully in the constitutions reduces the risk of making mistakes. Second, we consider the indicator of Ben Bassat and Dahan most accurate as The Netherlands and Sweden do commit to social security in the constitution, which is not given for the Netherlands in the CCP data set and not for Sweden in both the TIESR and the CCP data sets.^[9] Third, a lot of data on commitment to social security is missing for different OECD countries and years in the CCP dataset. We could have exploited the panel nature of this dataset if this dataset was more complete regarding social rights to social security. However, the added value of using this panel nature is limited as constitutions hardly change over time. Looking at the CCP dataset, I only found four changes in CCSS for our sample of 28 OECD countries during the period 1990–2012.^[10] Besides, constitutional changes are expected to be highly correlated with other political or economic shocks, making it difficult to separate the effect of social rights from the effects of these other shocks.

Another reason why we accept that we lose some countries when we use these CCSS and rigidity indicators is that Ben-Bassat and Dahan (2008, 2016) already studied the association between CCSS and social expenditure for a large variety of countries from all over the world. Our contribution is in finding effects rather than associations, using an IV approach, and looking at different social expenditure schemes. For these purposes, it is important that countries are not too different in unobserved characteristics that might drive our results. Therefore, we value quality of the data over quantity.

We transform the CCSS indicator of Ben-Bassat and Dahan (2008, 2016) in a dichotomous variable, being 1 if there is at least a general statement in the constitution on a social right to income, unemployment, sickness, work injury, old age, survivor or disability and 0 if there is no statement on any of these categories. The choice of taking the sum of these five categories corresponds to Ben-Bassat and Dahan (2008, 2016). A high overlap and substitutability between the different types of commitment to social security, caused by the abstract formulation of the legal provisions, makes us believe that the sum has more explanatory power than the individual commitment to social security variables.^[11] We use a dichotomous variable in our baseline models, because we expect the existence of a legal provision in the constitution to be more important than the concreteness of this legal provision.

The outcome variables we consider are social expenditure variables for which we use the Social Expenditure Database (SOCX) of the OECD. Our main variable for social security is public and mandatory private gross total social expenditure as a % of GDP, which we define as total social expenditure.^[12] This total social expenditure consists of spending on old age and survivor, incapacity, unemployment, ALMPs, health and family, which are our next dependent variables.^[13] A description of the different social expenditure variables is given in the Appendix in Table A.1.

Table 1 gives the descriptive statistics on the different social expenditure schemes for countries with and without CCSS, for EU countries over the period 1990–2012. We find that total social expenditure is on average 23.5 percent of GDP in the countries with CCSS and 22.0 in the countries without CCSS. This difference is 7 percent relative to the mean of total social expenditure for countries without CCSS. The relative differences are the largest for spending on unemployment and ALMPs (respectively 76 percent and 88 percent relative to the means in countries without CCSS). Further, in countries with CCSS, we observe less spending on old age and survivor (−2 percent) and more spending on incapacity (+29 percent). Regarding social expenditure which is not taken into account in CCSS, we find slightly larger spending in countries with CCSS on family (+4 percent), whereas we find less spending on health (−3 percent).

Our instrumental variable, the rigidity of the constitution, is the average of the standardized indices for rigidity in Lorenz (2005). This index considers the factors: kinds of majority, success rate, times of voting, unicameral/bicameral legislature, initiative actors, special body or regulator legislature, need of elections between two votes, electoral system, approval by referendum, approval by states’ legislatures. The rigidity of the constitution, CCSS and total social expenditure are given for the different countries in the year 2008 in Table A.2. We consider the rigidity of the constitution of Lorenz (2005) a better instrument than the amendment rate because it contains more dimensions and has the highest correlation with most other variables for the rigidity of the constitution (Ginsburg and Melton 2015).

Further, we create an interaction variable between CCSS and political party in office to study how constitutional commitment and political party in office have a combined effect on social security benefits. For politics, we use left-wing/center/right-wing cabinet posts in percentage of total cabinet posts from the comparative Political Data Set (Armingeon et al. 2013).

The control variables we use are GDP per head of population (USD in thousands, constant prices, 2010 PPPs), the old age dependency ratio (percentage of 65+ relative to 15–64 years old), dummies for legal origin, unemployment rate (standardized unemployment rate, all persons) and year dummies, see the AppendixTable A.3 for the descriptive statistics. These control variables are chosen as they have the largest effects on the social expenditure schemes and are expected to influence the effect of CCSS on social expenditure. These control variables are in line with the literature (Ben-Bassat and Dahan 2008; Ben-Bassat and Dahan 2016; Mulligan et al. 2010; Kittel and Obinger 2003). For legal origin, we use dummies for French, English, German and Scandinavian legal origin, where we use socialist legal origin as the reference category.

In our robustness analysis, we also investigate the effect for the sample of EU-countries together with Iceland, Switzerland and Norway, as well as for a sample of OECD countries without Japan and Korea.^[14] Japan and Korea are outliers as they have a different Asian system with very low levels of social spending.^[15] Hence, Japan and Korea are outliers for reasons independent of CCSS and therefore we decided to drop these two countries from our database.

5 Results

Table 2 presents the regression results of CCSS on total social expenditure.^[16] Column (1) shows the results for the year 2008 where we only control for GDP per capita and the old age dependency ratio. This specification suggests a positive effect, significant at a 10 percent level, indicating that countries with CCSS spend on average 2.0 percentage points of GDP more on total social expenditure. This implies that the mean total social expenditure rate for countries with CCSS is 9% higher than the baseline expenditure percentage for countries without CCSS (22%). In Column (2), we use the years 1990–2012 and add year dummies to our empirical specification. The robust standard errors are replaced by panel corrected standard errors and we include a first order serial correlation component. The effect of CCSS on total social expenditures slightly increases and is highly significant now.^[17]

Adding control variables for the unemployment rate and legal origin, in column (3), does not change much. German legal origin and Scandinavian legal origin have the largest positive effect on total social expenditure, socialist and English legal origin the lowest. The unemployment rate increases total social expenditure, which we expected as it controls for business cycle differences.

The effect of CCSS on total social expenditure increases to 3.1 percentage points in our 2SLS model, in which we use the rigidity of the constitution as instrument. An increase in the coefficient suggests an underestimation of the effect size when we do not control for endogeneity. We expect that endogeneity comes from (a broad definition of) political decision making, because social expenditure is primarily the result of political decision making independent of CCSS. Finding a larger effect of CCSS when we control for endogeneity therefore implies that political decision making in favor of social expenditure is negatively correlated to CCSS. We can infer from this that the constitution acts as a substitute for political decision making. In the first stage regression, we find a large positive effect of the rigidity of the constitution on CCSS. The F-test of excluded instruments is easily rejected with an F-value of 73.6, see Table 3, suggesting sufficient relevance of our instrument.

In column (5), the effect size increases to 3.8 percentage points when we control for unobserved correlation between our selection model and second stage regressions by using the Heckman two step model. The effect size in our preferred specification, column 5, is a substantial 17% relative to the mean of total social expenditure for countries without CCSS. Also the extent to which the Heckman two step model is correcting for unobserved correlation, denoted by rho, is with a value of −0.4 within acceptable proportions. The negative rho means that there is a lower probability of CCSS when other factors already take care of social expenditure (e.g., politicians or labor unions).

Table 4 presents the effect of CCSS on the different social spending categories when we split up total social expenditure. The largest effect sizes, expressed as a percentage of average spending, are found for expenditure on unemployment and ALMPs. In our preferred specification, column 5, a positive effect of 2.1 percentage points is found for expenditure on unemployment, which is an increase of 248% relative to the mean of expenditure on unemployment in countries without CCSS. For expenditure on ALMPs we find a coefficient of 0.5, which is an increase of 99% relative to the mean. Although smaller in relative size, we still find large positive significant effects of CCSS on expenditure on old age and survivor and on incapacity. A positive coefficient of 2.46 is found for old age and survivor, which is about 26% relative to the mean in countries without CCSS and we find a positive effect of 0.7 percentage points for incapacity which is about 28% relative to the mean. However, for the effect on old age and survivor we find no significant effect in specifications (2) and (3), suggesting that we have to be more careful in drawing conclusions. We find a negative rho for all kinds of social expenditure except for expenditure on family benefits. Meaning that for all these other social expenditure schemes, there is a lower probability of CCSS when other factors (e.g., politicians or labor unions) already take care of social expenditure.

No significant positive effect is found on health and family spending, which are not taken into account in the CCSS variable. This provides extra evidence that the effects we find on social security expenditure are due to CCSS and not caused by a third factor, such as a large welfare state. We even observe a negative significant effect on family, suggesting that this social expenditure type is substituted by expenditure on social security. This may be explained by government budget constraints or because the government takes into account the total budget of the needy, which is already higher when they can rely on generous social security benefits.

The results remain the same in the robust analysis. Table A.4 in the Appendix shows that the results are robust when we add the European non-EU countries: Iceland, Norway and Switzerland. Table A.5 shows robust results in our sample of 28 OECD countries, when we exclude Japan and Korea as they have a different Asian system. Further, we find the same positive effects when we only consider the period before the Great Recession (1990–2009), in Table A.6. Finally, the results remain robust when we standardize the three values with the lowest and highest rigidity of the constitution and when we transform the variable for the rigidity of the constitution in a dichotomous variable, to correct for possible outliers, see Tables A.7 and A.8.

We study non-linear effects in Tables A.9 and interaction effects with politics in Table A.10, see the Appendix. In Table A.9, we observe significant negative effects of the square of CCSS on total social expenditure, suggesting that the concreteness of CCSS is less important than the statement itself. Regarding interaction effects with politics, no effect is found of left-wing cabinet seats on social expenditure, nor of left-wing cabinet seats interacted with CCSS, see Table A.10Appendix. We find some evidence that more right-wing cabinet seats translate in lower total social expenditure but that the interaction between right-wing cabinet seats and CCSS has a positive effect on total social expenditure. This suggests that right-wing politicians reduce total social spending less when there is CCSS. However, more research is required on this result as the effect is not significant in specifications (2) to (4).

6 Discussion and Conclusion

In this paper, we studied the effect of constitutional commitment to social security (CCSS) on different kinds of social expenditure. We used a pooled cross sectional database for 17 EU-countries from 1990 till 2012.

The main challenge of research on institutions-like CCSS is that they are related to many other things like culture, religion, legal origin, geography, political institutions, etc. We deal with this potential endogeneity problem extensively by limiting the sample to more similar EU-countries, control for legal origin and use 2SLS models and the Heckman two step model with the rigidity of the constitution as instrumental variable.

First, we find a positive significant effect of CCSS on total social security expenditure, which increases when we control for endogeneity. This includes positive effects on the categories of social expenditure on old age and survivor, incapacity, unemployment and active labor market policies. This is in accordance with the rights-based approach to development, which supplements the focus on market institutions and property rights with human rights and social policies (ILO 2014; Townsend 2007). This result corresponds with the findings of Ben-Bassat and Dahan (2008, 2016) who find a positive relation between CCSS and transfer payments and between CCSS and the extent and coverage of measures of social security laws.

Second, the results show that the added value of CCSS is mostly found for expenditure on unemployment and ALMPs. Blekesaune and Quadagno (2003) and Van Oorschot (2006) show that the general public perceives the unemployed as less deserving than the old and disabled, suggesting lower support for spending on the unemployed by the median voter. This could explain why CCSS, rather than the median voter theory alone, can explain the scope of expenditure on unemployment and ALMPs. Hence, the importance of CCSS is mainly to protect people who are perceived as less deserving, which makes CCSS a substitute for political decision making. This is in line with the theory of the interdependence cost calculus, in which Buchanan and Tullock (1962) argue that the role of the constitution is mainly to protect minorities. These are also supported by finding more sizable effects when we control for endogeneity and by a negative rho in the Heckman model. This suggests that there are third factors (e.g., political decision making) that have a positive effect on social expenditure and a negative effect on CCSS.

Thirdly, no positive significant effect is found for expenditure on families and health which are the two social expenditure categories that are not included in CCSS. This indicates that the positive relationship between CCSS and the social security types of social expenditure is really due to CCSS and not due to different social preferences that affect both CCSS and social expenditure.

The result that a constitutional right to social security has a positive effect on social protection may have policy implications. Governments that want to improve (long-term) social security for their citizens should consider to add CCSS when the constitution is amended. Other policies then CCSS can increase social security more directly, but CCSS may increase commitment to social security in the long-run, whereas policies and laws give less security for the long-run.