In order to analyse the micro-foundations of the link between objective risk of unemployment, the perception of this risk and finally demand for more protection in the specific domain of unemployment benefit, we rely on SHP data from the year 2011 which includes all variables of interest and for which unemployment figures across occupations were available. We focus on objective risk as defined by the level of unemployment in one’s occupation - a common measure of objective unemployment risk using data from the Federal Statistical Office. Given our focus on unemployment risk, we restrict the sample to the active population who are the only ones who might be directly impacted.
Our analysis relies on two distinct steps. In a first step, we assess whether the Occupational Unemployment Rate is indeed associated with the perception of unemployment risk. In a second step we analyse the link between occupational unemployment rate and policy preferences regarding unemployment policies. More precisely, we are interested in assessing to what extent this link is mediated by one’s perception of risk and to what extent mediated by other factors – in particular education and income – that might also affect the relationship.
Our empirical strategy implies having two dependent variables corresponding to the two stages of the analyses. The first one is perceived unemployment risk which has been measured using the same question as the one presented aboveFootnote 3.
The dependent variable in our second set of models is respondent’s preferences with regard to unemployment benefits. Here, we use a specific question on whether spending for unemployment benefits should be increased, decreased or kept at its current levelFootnote 4. As is often done in studies on redistributive policies we recode this 3-category variable into a dummy variable distinguishing between those who favour an increase in unemployment benefit spending (coded 1) and those who either support a cut or favour a status quo (both coded 0), the latter de facto corresponding to an opposition of an increase in unemployment benefits.
The focus independent variable in both analyses is the Occupational Unemployment Rate (OUR). This has been created using data from the Swiss Labour Force Survey conducted in 2011 which includes about 30′000 individuals and allows computing average unemployment rates per occupation. We have computed these measures for each of the ISCO-2 digit occupations documented in the survey and matched these OUR measures with individuals based on their occupation. This measure corresponds very closely to what unemployment occupational risk has been operationalized in the literature so far (see Rehm 2009).
Our models further include a series of control variables that can be classified into two main categories: sociodemographic variables and individual characteristics that are associated with the economic self-interest of respondents. Among variables that belong to this second subset, we include the yearly net income at the individual level. We also include a variable measuring relative skill specificity. For this variable, we used data that is made available by Torben Iversen on his website and provides skill specificity measures that have been computed based on the ISSP surveys. This variable measures how specific one’s skills are and should therefore capture how difficult it is for a person to find a job that makes use of these specific skills in case of job loss (see Cusack et al. 2006; Iversen and Soskice 2001). Whether one has a temporary or long term working contract, thus whether a person is a labour market outsider or insider (see Rueda 2007), is measured with a dummy variable (0 = unlimited contract; 1 = temporary contract). We further include a control variable that measures previous unemployment experience (coded 1 for all those who reported having been unemployed in the past and zero otherwise). Furthermore, we also control for the number of earners in a household by distinguishing between individuals who are the only wage earners in their household (coded 0) and those who live with at least one other wage earner (coded 1). The underlying idea is that the consequences of unemployment on one’s financial situation are tougher if that person is the only wage earner in the household.
The sociodemographic control variables includes age (measured as a metric variable), gender (0 = man; 1 = woman), education (a three category variable in which the reference category is no or only compulsory education; the other two categories being “above compulsory” and “tertiary” education). We also add a dummy variable for region which controls for all the observed and unobserved regional characteristics of the 7 Swiss regions as defined by the Swiss Federal Statistical Office (corresponding to the NUTS-2 level). In some of the models we further include a variable measuring social class. For this we use Oesch’s class scheme which includes 8 different categories: Self-employed professionals and large employers, small business owners, (associate) managers and administrators, office clerks, technical professionals and technicians, production workers, socio-cultural (semi-)professionals and service workers. According to its author, this class scheme captures social stratification in modern societies, taking also account of the deindustrialization, the welfare state expansion, and the increased participation of women in the labor market. The focus is not only on hierarchical divisions but also on horizontal cleavages to reflect better the heterogeneity of today’s middle class (Oesch 2006).
As the two dependent variables in our models have different measurement levels, we use an OLS regression to model perceived unemployment risk (which is measured with a11 point scale) and a logistic regression for the model in which preferences for unemployment benefits represent the dependent variable coded as a dummy. In addition to these analyses, we perform a product of coefficients test for the second set of models. This enables us to estimate how much of the relationship is mediated by perceived unemployment risk and how much by the socioeconomic factors we focus on: education and income. This test is similar to the Sobel test (Baron and Kenny 1986; Sobel 1982), used to determine mediation effects.Footnote 5
In Table 15.1, we find in the first model with only the four objective variables related to the theories on risk: OUR, skill specificity, contract type (insider/outsider status), and income that all except the last are highly significant predictors of perceived unemployment risk with the relationships working in the expected direction. The relationships remain, and become significant for income, when controlling for basic sociodemographics in model 2, social class in model 3, and past unemployment and number of earners in model 4. This seems to show that these objective measures, and namely OUR, do capture something that people are aware of and can be used as a way of measuring the insecurity one feels at her or his workplace. Depending on the model, the difference in perception of risk between workers in an occupation with the lowest and highest rates of unemployment is of between 1.4 and 1.8 points on an 11-point scale, a similar effect of that of having a temporary versus fixed contract. It should be pointed out that the R-squared, especially in the model with only the objective measures, is very low, indicating that there is much else determining one’s perceived risk.
Table 15.1 OLS regression models predicting perception of unemployment risk
As expected, having been unemployed in the past increases worries among individuals, possibly making unemployment possibilities seem more real for. Not being the only earner in the household does also diminish somewhat the perceived risk, as expected.
There are relatively strong regional effects, which seem to reflect at least in part the fact that unemployment tends to be higher in French and Italian speaking regions, as well as in the canton of Zurich in the German speaking part of Switzerland. Older individuals also show a tendency towards a higher perceived risk, but no gender differences arise. Surprisingly, class, at least as operationalized here with the eight classes of Oesch (2006) has practically no effect, which goes against our suspicion that the OUR might confound somewhat with class differences.
In the second step, we ran four models predicting individual preferences for increasing unemployment spending (see Table 15.2). We find that in the first model with OUR and basic sociodemographic controls, the effect of the first on preferences is highly significant, as predicted by the theory. In the second model, we add risk perception. If OUR was completely mediated by the perceived risk, which seemed already unlikely given our first set of models, we would expect the effect of OUR to become non-significant. This does not happen, which tells us that both measure in part different things and influence individual preferences by different mechanisms.
Table 15.2 Logistic regression models predicting being in favour of increased unemployment spending
In Model 3, we test our hypothesis of the effect of OUR on preferences being mediated by income and education, and find strong support. OUR becomes completely non-significant (p = 0.122) when these two variables are added, thus hinting at a strong mediation effect. Both education and income are highly significant at predicting one’s preferences regarding unemployment spending. Risk perception in turn remains significant with these variables added. Interesting again is the strong effect of regions, showing that expectations towards the welfare state are heterogeneous within Switzerland.
So as to better understand the mediation effects, we present in addition the results of a product of coefficients tests (Baron and Kenny 1986; MacKinnon and Dwyer 1993). In this test, five regression models are run, the resulting coefficients are standardized and, based on the latter, the effects of the different paths are calculated. The regressions comprise the five variables of interest: preferences regarding unemployment spending, OUR, perceived unemployment risk, education, and income. The variables are coded as in the models above. In all five models, we control for age, sex, and region. In Fig. 15.2, we show the resulting unstandardized effects of the three paths of interest. We use bootstrapping to calculate standard errors and the significance of the paths. Table 15.3 shows the rest of the results, namely the proportion of the total effect that is mediated by each of the three variables and the total direct and indirect effects.
Table 15.3 Product of coefficients test: proportion of total effect mediated and total effects
As expected by the theory, we find that perceived risk significantly mediates the relationship between OUR and preferences (p < .01). However, this only represents about 14% of the total effect between OUR and preferences, which is relatively low compared to what we should find were this the main explanation for this effect. The size of the mediation effect by income is the same as that of risk perception but is not significant. Education in turn mediates the effect of OUR on preferences by 18% and is highly significant (p < .01). Income is highly significant and mediates as much as the two others combined, 32%. The combined effect of both sociodemographic variables is 0.042, more than three times that of risk perception. In total, they explain 50% of the total effect between OUR and preferences. A proportion of 64% of the total effect is mediated, the remaining direct effect being non-significant. This means that the two explanations proposed are likely to account for most of the relationship and no crucial component has been omitted from this analysis.
Our results indicate that individuals’ perception of unemployment risk is to some extent linked to the unemployment rate in their occupation and that this link in part explains the relationship between the latter and policy preferences. However, we find that a more important explanation is related to the socioeconomic characteristics of individuals in a given occupation. We argue that this is because unemployment tends to be higher among individuals with low income and education levels. Thus occupations with higher unemployment rates tend to also to group together individuals with similar sociodemographic background. This means that belonging to a group more affected by unemployment will make the individual more inclined to support policies that help the unemployed, not so much out of self-interested calculations, but rather through a higher awareness of the issue, identification with those affected, and shared norms or solidarity (Svallfors 2006).