Hiring, employment, and health in Scandinavia: the Danish ‘flexicurity’ model in comparative perspective

Europe is currently struggling with a deep and long-lasting economic crisis,¹ resulting in high unemployment rates in several countries. In the 28 European Union member countries, the unemployment rate increased from 7% in the start of 2008 to 11% in 2013 (Eurostat 2015a). Correspondingly, there has been a renewed research interest in unemployment (e.g. Stuckler et al. 2009; Karanikolos et al. 2013; Norström and Grönqvist 2015). However, there is considerably less energy devoted to hiring and employment, which is the topic of the current study. During a crisis, employers have more applicants to choose among for each available job opening, and individuals with some kind of negative signal attached to their résumé are therefore less likely to be hired. To have bad health status is one type of negative signal, and this paper asks: Are people with ill health less likely than people who report good health to gain employment in a period with low labour demand?

Ill health is measured in two ways in the current study: (i) those reporting bad/fair health and (ii) those stating to have a limiting long-standing illness (LLSI). The research context is set to Denmark, Norway, and Sweden. Although the Scandinavian countries share a whole range of characteristics, there is especially one critical difference. The Danish ‘flexicurity’ labour market model has lenient hiring and firing regulations as one of its core elements (Heyes 2011; van Kersbergen and Hemerijck 2012), and it is therefore quite easy for a Danish employer to fire an employee. In contrast, employment protection legislation (EPL) is strong in both Norway and Sweden. Because the costs of firing someone are low in Denmark, this could imply that Danish employers tend to ‘take the risk’ associated with hiring someone with bad health. Norwegian and Swedish employers, on the other hand, could be more reluctant to hire a person with ill health. The candidate could, for instance, tend to be less productive because of the impaired health status, and with strong EPL it will be more difficult (and hence more costly) to fire him/her. The second research question is thus whether the ‘flexicurity’ labour market model – with weak employment protection – is an advantage for people with ill health's hiring probabilities.

To investigate individual hiring probabilities is obviously important, because it will demonstrate the mobility flow (or lack thereof) for people with health problems. However, it is also interesting to explore how people with ill health perform ‘as a whole’ on the labour market. Permanent employment is unquestionably the best way to ensure a firm attachment to the labour market. Accordingly, this paper will look into overall employment rates and the use of temporary work contracts for people with ill health, and compare the rates to people reporting good health. The third and final research question is hence how firmly attached to the labour market are people with ill health in Scandinavia?

In order to investigate these questions, the longitudinal part of the European Union Statistics on Income and Living Conditions (EU-SILC) data material is utilised (time window: 2008–2011). Here, we can follow the same individuals for a maximum of four years, and see whether people with ill health have lower hiring probabilities than those with good health status. The models are estimated with generalised least squares regression (GLS). Since the EU-SILC data material is harmonised for comparative purposes, we can compare labour market outcomes for people with health problems between Denmark, Norway, and Sweden.

The current study adds to the existing literature on health and employment status on two areas. Firstly, by investigating hiring and employment patterns during an economic downturn. The low labour demand implies that it is possible for employers to ‘skim the cream’ to a higher extent, because of a larger amount of applicants for each available job opening. ‘Cream-skimming’ means that employers get to handpick employees, and people with health problems might therefore be in a particularly vulnerable position on the labour market during a crisis. Secondly, through an explicit comparative focus on the health–employment relationship. Potential differences in results between Denmark on the one hand and Norway and Sweden on the other could indicate whether the ‘flexicurity’ model is favourable for individuals who traditionally struggle on the labour market, here exemplified by people with ill health.

There are four possible mechanisms able to explain why people with ill health are ‘picked last’ in the recruitment process (Hedström and Swedberg 1996; Hedström 2005). First, employers are profit maximising, and therefore wish to hire the most productive employees. In an effort to do so, the employer might look for signals of physical/mental strength (e.g. few sick days). Second, employers are risk-averse (Aigner and Cain 1977). To hire someone with bad health represents a risk factor because he/she could deteriorate further in health, possibly implying high sick leave (followed by an expensive firing decision, and a time-consuming recruitment process). Third, employers might discriminate against people with ill health (Becker 1971 [1957]; Arrow 1973), either because they dislike people who are unfit, or because they believe health to be correlated with undesirable personality characteristics (e.g. low levels of conscientiousness). Fourth, because of the scarring effects of unemployment (Oberholzer-Gee 2008; Eriksson and Rooth 2014). Other employers might have emphasised one or several of the abovementioned factors, and the job applicant with ill health could have struggled to gain employment in previous application processes, implying more unemployment on the résumé. Employers could therefore be sceptical about him/her not because of the health status, but rather because of the accumulated amount of unemployment (a signal of low productivity).

Unfortunately, the current data material is not well suited for distinguishing between the abovementioned mechanisms. What the EU-SILC data are suited for, however, is cross-national comparisons of results, and potential differences between the Scandinavian countries could be of interest from a policy point of view. If the employment prospects for people with ill health are better in Denmark, this could indicate that weak employment protection is one way to improve the situation for individuals with a rather ‘loose’ labour market attachment.

Previous research has established a robust relationship between health and employment status. Analysis of 11 European countries showed that healthier people were more likely to become – or remain – employed than less healthy people (Schuring et al. 2007). Similarly, impaired health status was associated with longer unemployment spells in both Canada (Stewart 2001) and Australia (Butterworth et al. 2012). Furthermore, workers with ill health were less likely to return to work after unemployment in the Netherlands (Schuring et al. 2013) and in Britain (García-Gómez et al. 2010). It is not obvious, however, that these findings are generalisable to the Scandinavian context, where employment rates are comparatively high. A Swedish study found that the association between subjective mental distress (GHQ-12) and re-employment rate was insignificant once a number of covariates was included in the regression (Skärlund et al. 2012).

Having ill health could be particularly disadvantageous when the demand for labour is low, because employers are more able to ‘skim the cream’ in the recruitment process. In line with this argument, a study from Britain (observational period: 1973–1993)² found that people with ill health struggled to re-enter the labour market in the aftermath of economic downturns (Bartley and Owen 1996). Similar results were observed for Norway in the years 1980–2005 as well (van der Wel et al. 2010). The current study will investigate if people with ill health struggle to enter the labour market during the current economic crisis, and asks:

I. Do people with ill health have a lower probability of gaining employment than people with good health during the economic downturn in Scandinavia?

Denmark, Norway, and Sweden all have high tax levels, free or heavily subsidised education, and a universal health care system. Furthermore, the countries share an emphasis on work and employment being the single most important mean for integration and social participation (Lødemel and Trickey 2001; Bengtsson 2014). Hence, the respondents in the study samples live in countries that are organised in a comparable manner, ensuring that the cultural dissimilarities are few. Nonetheless, we have to consider (potential) cross-national differences relevant for the relationship between ill health and employment status.

A first important difference concerns overall demand for labour in Scandinavia 2008–2011 (see Figure A1 in the appendix). The unemployment rate has been roughly 3% in Norway and between 6% and 8% in Sweden. There was a rapid increase from 3.5% to 7.8% in Denmark 2008–2011. Denmark (7.6) and Sweden (7.8) are experiencing similar unemployment levels in 2011, while labour demand is considerably higher in Norway. Thus, it is particularly interesting to compare results for Denmark and Sweden in 2011.

In the years 2010–2014, the employment rate for 20–64-year-olds has been approximately 76% in Denmark, and 79–80% in Norway and Sweden (Eurostat 2015b). The share of temporary work contracts in the same age- and time span is low for Denmark and Norway (7–8%), while Sweden is on a higher level at roughly 14% (Eurostat 2015c) due to legislative amendments in 2003 and 2007. The share of employees in the public sector is comparable in the three countries: 32.6 in Sweden, 33.6 in Denmark and 35.4% in Norway (Dølvik et al. 2015). Overall, the industries of the three Scandinavian labour markets are very similar (Nordic Statistical Yearbook2014, see Table 8.2). Manufacturing and other industry make up a similar share of the labour market in Denmark (13.9), Norway (13.0) and Sweden (12.6), but the service sector is somewhat larger in Sweden (15.2) than in Denmark (11.2) and Norway (11.4). There are some slight differences in average retirement age: 62.3, 63.5 and 64.4 in Denmark, Norway, and Sweden, respectively, in 2010 (Halvorsen and Tägtström 2013), and it is therefore important to include age in the regressions.

Apart from employment rates being somewhat lower in Denmark, and the more widespread use of temporary work contracts in Sweden, there are few differences between the countries in how the labour markets are organised. Perhaps more relevant, however, is the generousness of the unemployment benefits. If the benefit is very generous, people with health impairments might be less inclined to search for a (new) job. A short-term unemployed single person without children on average wage would in 2012 receive 65% of previous income in Norway, 57% in Denmark, and 45% in Sweden (OECD 2015a). Hence, there might exist a larger ‘incentive’ to stay unemployed in Norway, and this is worth remembering while interpreting the results.

It is more challenging to compare the countries regarding disability benefits, because the benefit is means-tested in Denmark, while previous income level is the basis in Norway and Sweden. Luckily, EU-SILC includes disability data, and we can investigate whether there are cross-national differences in the extent to which people with ill health report ‘disabled’ as economic status (see Table A1 in the appendix). In general, both men and women with ill health in Sweden report being disabled less often than their neighbouring counterparts do in 2011. This is probably a reflection of the stricter eligibility criteria for disability benefits introduced in recent years (Hägglund 2013; Lidwall 2013). The differences between Norway and Denmark are minor, although Norwegians with ill health are disabled to a slightly higher extent.

To summarise, Denmark, Norway, and Sweden are similar on many domains, and to compare results across these countries is therefore possible. However, there is one important difference between the Scandinavian countries, which is the topic of the subsequent section.

The Danish ‘flexicurity’ labour market model consists of three main parts: (i) minimal job protection, (ii) generous unemployment benefits, and (iii) extensive use of active labour market policies (Heyes 2011; van Kersbergen and Hemerijck 2012). Accordingly, there is a high level of worker- and job turnover rate, made possible by the lenient hiring and firing legislation (Madsen 2004; Andersen and Svarer 2007). It is especially among skilled and unskilled workers that the employment protection is weak, while employers have less flexibility in dismissing traditional ‘white collar’ employees (Jensen 2011). Hence, there is an important skill component: hiring and firing regulations are more lenient in low-skill occupations, while higher skilled employees are protected to a larger extent.³

The impact of EPL on labour market attachment will probably be especially important for vulnerable groups, such as people with health problems. Results from a study of 26 European countries indicate that stricter EPL is able to lower the firing likelihood for people with ill health, but only in countries facing less severe or no economic crisis at all (Reeves et al. 2014). Similarly, we might expect EPL to have an impact on employers’ hiring decisions. The flexible legislation in Denmark could imply that employers are more inclined to take the ‘risk’ associated with hiring a person with bad health status. If the newly hired employee turns out to be unproductive – e.g. has too many sick days, or is not fit enough to do the job – the employer can simply fire him/her, without worrying about any major costs involved. This is different for Norwegian and Swedish employers, who have to take more rigorous EPL into account while considering whom to hire. The strong EPL in these two countries could mean that people with ill health struggle to gain employment, because (risk-averse) employers worry about potential difficulties with how to fire people (with ill health) that turn out to be unproductive. Correspondingly, we ask:

II. Is the ‘flexicurity’ labour market model – with weak employment protection – an advantage for people with ill health's hiring probabilities?

Note that high sickness absence does not constitute proper grounds for dismissals under normal circumstances. Nonetheless, recent evidence has shown that respondents who deteriorate in health tend to lose their jobs in Denmark (Heggebø 2015). Moreover, sickness absence was considered important while deciding whom to fire in a Danish manufacturing company in 2010 (Svalund et al. 2013: 194). Thus, employers are apparently sensitive to (developing) health problems among employees, and it is reasonable to assume that health status is of importance in hiring decisions as well.

Although hiring probabilities are important, it is perhaps equally interesting to investigate how people with ill health perform ‘as a whole’ on labour market outcomes. Here it will be particularly rewarding to compare Denmark and Sweden, who experienced quite similar demand for labour in 2010–2011. Employment rates and the use of temporary work contracts will therefore be explored, in order to answer the third and final research question:

III. How firmly are people with ill health attached to the labour market in Scandinavia?

The longitudinal part⁴ of the EU-SILC was utilised in the present study. These panel data are structured in a rotary format, where individuals are followed for a maximum of four years (2008–2011). The panel is unbalanced, which means that not everyone is followed for four consecutive years. Often, there are only two or three observations for each individual. Due to this shortcoming, it is not sensible to follow people from one year to the next and use a change variable (e.g. unemployed 2008 to employed 2009) as the outcome measure. This implies that a (potentially) large number of employment status transitions would get lost, yielding problems with statistical power. All transitions happening before the respondent was included in the sample would also go unrecorded. This is especially important for the current study, since the outcome is an infrequent event (159, 343 and 331 hirings for Denmark, Norway, and Sweden, respectively) due to the economic downturn.

Luckily, EU-SILC includes a question that enables us to deal with these difficulties. The respondents are asked about the most recent change (during the last year) in employment status, which means that individuals gaining employment during the year before he/she was included in the sample is recorded. Similarly, respondents who gained employment in 2009, but only contribute with information in 2008 and 2010, are also registered.

Because the EU-SILC data are harmonised for comparative purposes, we are able to examine whether there are cross-national differences in labour market attachment among people with ill health. Information is gathered on an individual level throughout Scandinavia. People not selected for answering health questions are dropped from the sample, along with those with missing information on health variables (81 observations). Moreover, people below the age of 16 are not included, yielding a total sample of 38 922 observations. We place no further restrictions on the sample, for two reasons. First, due to the large time span between survey rounds (minimum one year), a large number of initially employed people are likely to experience unemployment and re-employment⁵ before follow-up. Second, even people once stating to be retired, disabled or students gain employment in the present data material, although it is quite uncommon.

Dependent variable in the following analysis is hiring, a dummy variable based on a question regarding most recent change (i.e. during the past 12 months) in employment status. People who state that they have went from unemployment or being economically inactive to employment are coded 1 (else = 0). The most important independent variable is bad/fair health, a dichotomised measure computed from a question on self-assessed health status. People reporting very bad, bad or fair health are coded 1 (very good or good = 0). Those with fair health are included for two reasons. Firstly, because the number of respondents stating to have very bad or bad health is low (5.7%, 6.9%, and 5.0% in Denmark, Norway, and Sweden, respectively), yielding problems with statistical power. Secondly, even people with less serious health impairments could face difficulties in gaining employment. To check the robustness of the results, the health measure is changed to limiting long-standing illness (LLSI) in all model specifications. Two questions are used: whether the respondent suffers from a chronic long-standing illness, and whether the respondent is limited in activities people usually do because of health problems. People answering yes on both are coded 1 (else = 0).

LLSI should capture respondents with quite serious (and for employers: more visible) health challenges, whereas bad/fair health is comprised of a more heterogeneous health population. It is therefore likely that the more visible health measure will yield stronger negative effects (i.e. people reporting LLSI should have lower hiring probabilities than those with bad/fair health). The correlation between the two health measures is 0.514, 0.521, and 0.526 in Denmark, Norway, and Sweden, respectively, implying a moderate to strong association (Cohen 1988). The fact that the correlation is not even higher indicates that these measures are capturing somewhat differing aspects of health. More objective health measures would obviously have been preferable, but the reliability of self-reported indicators seems to be acceptable (Martikainen et al. 1999).

A number of covariates are included in the analyses. The dummy variable woman takes the value 1 for women, 0 for men. Educational qualifications, based on the highest International Standard Classification of Education level attained, consist of three dummy variables. Pre-primary, primary, and lower secondary is collapsed to primary education. (Upper) secondary and post-secondary non-tertiary is collapsed to secondary education (higher education = reference category). Age is derived from questions on year of birth and year of survey, and is thereafter recoded into three dummy variables: Young age (≤30 years), old age (≥60 years), and prime age (30–59 years, the reference category). The dichotomous measure married takes the value 1 if the respondent is married (else = 0). In order to see how ‘firmly’ attached people with ill health are to the labour market, the dummy variables employment (currently employed coded 1, else = 0) and temporary work contract (temporary = 1, permanent = 0) will be investigated.

Table 1 presents descriptive statistics⁶ stratified by country and gender. The hiring rate during the investigated time window is very low⁷ in all three countries, reflecting the economic crisis. The rate is lowest among Danish men (1.63) and highest among Swedish women (2.78). Women report significantly⁸ more ill health – both bad/fair health and LLSI – than men throughout Scandinavia. The amount of reported ill health is similar between the three countries (e.g. LLSI for men: 12.76, 10.75, and 12.80 in Denmark, Norway, and Sweden, respectively).

Educational level is distributed quite similarly in the three Scandinavian countries. Women have higher education to a somewhat larger extent in all three countries, and the ‘gender gap’ is largest in Sweden. There are comparatively few respondents below the age of 30 in Denmark and somewhat fewer older respondents in Norway. Respondents are married to a higher extent in Denmark, while the differences between Norway and Sweden are negligible. These minor cross-national differences in covariates are unlikely to cause large problems for the following analysis, and the main pattern is that of similarity.

Although the samples in the three countries are very similar overall, there might still be noticeable cross-national differences in observable characteristics among people reporting bad/fair health. Inspection of descriptive statistics for this subsample, however, does not indicate that this is the case (see Table A2 in the appendix). The three countries are still very similar, the main exception being the somewhat ‘negatively selected’ Swedish bad/fair health-sample. Compared to Denmark and Norway, Swedes reporting bad/fair health hold higher education to a lesser extent, report more often to have a limiting long-standing illness, and are more often above 60 years old. We need to remember this while interpreting the results.

Linear probability models are performed throughout. Logistic regression analysis is not preferred because of difficulties in comparing results across different models, groups, and samples (Allison 1999; Mood 2010). Nevertheless, logistic regression is run as a robustness check because a linear model could be an incorrect specification. GLS are preferred over ordinary least squares (OLS) because the former corrects for the fact that we follow people over time (Allison 1994). Hence, robust standard errors are reported. OLS models with standard errors clustered on individuals have also been estimated, and the results are almost identical as those derived from GLS (see Table A3 in the appendix for an example). Calendar year dummy variables are included in the regressions in order to account for the differential demand for labour (and other time trends).

The analysis section is structured in the following fashion. First, we see whether people with bad/fair health are less likely to gain employment in Scandinavia during the economic downturn. Afterwards we run the same models, but switch focus to people reporting more serious health impairments (LLSI). We then proceed to another sensitivity test, namely logistic regression analysis. Lastly, descriptive evidence on employment rates and the use of temporary work contracts is examined, in order to see how ‘firmly’ people with ill health are attached to the labour market.

Results from GLS regression of hiring by bad/fair health are presented in Table 2. Model 1 does not include any additional covariates, while model 2 adjusts for gender, age, marital status, and educational level. In the ‘naïve’ model, there is a significantly lower hiring probability for people reporting bad/fair health, but only in Sweden. The coefficient is negative but insignificant for Norway, and positive and significant for Denmark. The latter result even holds in model 2, where the reference group consists of 30–59 years old unmarried men with higher education. The coefficient is now positive and significant for Norway as well, but the effect size is smaller than in the Danish sample (0.005 vs. 0.012). People reporting bad/fair health do not differ in hiring probability for the Swedish sample in the adjusted model (b = 0.003, SE = 0.003).

Old age is associated with a lower hiring probability in all three countries, and young age is positively associated with likelihood of hiring in Norway and Sweden. Women have a higher probability of hiring than men in Sweden. Educational qualifications apparently matter more for labour market attachment in Sweden, where both the primary and secondary educated are worse off than respondents with higher educational qualifications. This is probably a reflection of the continuingly low labour demand in Sweden.

The analysis of hiring probabilities has also been stratified by education, age, gender, and marital status, in order to investigate possible interaction effects (see Table 3). It is among people with higher education and bad/fair health that the hiring probability is relatively high in Denmark (panel a). People with primary education and ill health, on the other hand, have a significantly lower hiring probability in Denmark. The latter result is found for Norway as well, but not for Sweden. It is somewhat surprising that people with bad/fair health and higher education quite often gain employment in Denmark, considering that the flexible employment protection regulation mostly applies to ‘low-skill’ workers (Jensen 2011). We return to this finding in the discussion.

The age stratified analysis (panel b) show that, in Denmark, people of prime age (30–59) with bad/fair health are significantly more likely to gain employment than people with good health. People above the age of 60 with ill health are less likely to gain employment in both Denmark and Sweden, but not in Norway. Lastly, the results indicate that it is particularly among men (panel c) and the unmarried (panel d) where ill health is negatively related to hiring likelihood in Sweden.

The results thus far lead to the following preliminary conclusion: People with bad/fair health and higher education are hired to a relatively high extent in Denmark during the economic downturn. Those with bad/fair health and low education, on the other hand, are significantly less likely to gain employment in Denmark. People with health problems are hired to a comparatively low degree in Sweden in 2008–2011, perhaps a reflection of the continuingly low labour demand. People reporting bad/fair health have quite similar hiring probabilities as those with good health status in Norway, the only exception being among those with low education (b = −0.018, SE = 0.008). Next, we turn to a number of robustness checks, in order to see whether these results hold.

All of the preceding regressions have also been calculated with a different health measure – limiting long-standing illness (LLSI) (Table A4 in the appendix). In several model specifications, the LLSI coefficient is negative and significant for Norway and Sweden, but this is never the case for Denmark. In these models – when the health measure is a more ‘serious’ one – the results indicate that Norway is the country where those with ill health fare the worst. The empirical pattern is therefore slightly different when the health measure is changed, but the main conclusion is the same: people with LLSI fare somewhat better in Denmark than in Norway and Sweden.

The differences in hiring probabilities between Denmark and Sweden (the reference group) are confirmed in a regression where all observations are pooled, and country dummies are interacted with bad/fair health or LLSI (see Table A5). The differences between Norway and Sweden are not statistically significant (for either health measure).

Lastly, the preferred model (Table 2, model 2) is run with logistic regression using both bad/fair health and LLSI, and the same pattern as before is evident (see Table A6). In summary, neither choice of health measure nor a linear model is responsible for the presented findings, and it seems as though people with ill health are more likely to be hired in Denmark than in Norway and Sweden. In the following – and last – analysis section, we investigate how firmly people with health problems are attached to the labour market in Scandinavia.

Table 4 presents the percentage who report ‘employment’ as their economic status in 2011, stratified by bad/fair health (panel a) or LLSI (panel b). Here we can investigate potential cross-national differences in the overall employment rates, in order to see whether the ‘flexicurity’ model is able to integrate more people with ill health into the labour market. The year 2011 is chosen because Denmark and Sweden experienced similar demand for labour.

Unsurprisingly, the employment rates are highest in Norway, and this holds for both people with good and ill health. It is, however, more rewarding to compare Sweden and Denmark. The employment rates are somewhat higher for those with bad/fair health in Denmark than in Sweden (37.62% vs. 34.58%), whereas the difference is almost non-existent for LLSI. The differences between Denmark and Sweden are not statistically significant (t-tests available on request). Although Sweden has experienced worse economic conditions than Denmark in the years preceding 2011 (see Figure A1), people with ill health report to be employed to a similar extent in these two countries.

Another important aspect is the use of temporary work contracts (Table 5). 2011 is chosen because it is the only year for which temporary work contract information is available for Denmark. The cross-national differences are striking. People reporting ill health (both health measures) in Norway have temporary work contracts to the same extent as those with good health. People with ill health in Sweden, on the other hand, are roughly 50% more likely to have temporary work, compared to people with good health. This holds for both bad/fair health and LLSI. The ‘health penalty’ is even more evident in the Danish sample. Approximately 4.5% of those with good health in Denmark have a temporary work contract. The corresponding share for those who report ill health are noticeably larger: 7% and 10% for bad/fair health and LLSI, respectively. The differences for Denmark and Sweden are statistically significant,⁹ for both health measures.

In summary, labour market attachment remains rather ‘loose’ among Danish respondents with health problems, as indicated by the larger share holding a temporary work contract. There is a ‘health component’ in the use of temporary work in Sweden as well, although to a lesser extent. In Norway, however, people with ill health hold temporary work to the same degree as those reporting good health. Lastly, the employment rates in 2011 for those with ill health are very similar in Denmark and Sweden, despite Sweden having experienced considerably lower demand for labour in the preceding years.

This paper has investigated three research questions, namely (i) whether people with ill health struggle to gain employment during the economic downturn in Scandinavia, (ii) whether the Danish ‘flexicurity’ labour market model is an advantage for people with ill health, and (iii) how firmly people with ill health are attached to the labour market. In general, people who report bad/fair health are not very disadvantaged in their hiring probabilities in either of the Scandinavian countries, but there are some signs of health being a negative feature in the Swedish context. When people with LLSI are considered, the likelihood of gaining employment seems somewhat worse than among people with good health in both Norway and Sweden. Overall, the results are most positive for the Danish sample, perhaps indicating that weak EPL makes employers more prone to ‘take the risk’ associated with hiring someone with bad health.

Nevertheless, one should be reluctant to conclude that the ‘flexicurity’ model is favourable for people with ill health, for four reasons. Firstly, it is only among those with bad/fair health and higher educational qualifications that the hiring probability is comparatively high. This is surprising given the fact that it is mostly among ‘low-skill’ workers that EPL is weak in Denmark (Jensen 2011). Hence, if weak EPL acts as an incentive in favour of hiring people with ill health, one should observe it among people with lower educational qualifications. Yet, the opposite empirical pattern is found: People with bad/fair health and primary education are significantly less likely to gain employment in Denmark. It is therefore doubtful that weak EPL is beneficial for people with a negative signal on the CV (e.g. ill health or unemployment ‘scar’). Note that weak physical health status is more of an obstacle to perform manual labour, and selection effects are thus stronger in ‘low-skill’ labour market segments. Still, this does not explain the differences between Sweden (−0.003) and Denmark (−0.017) in hiring probabilities for individuals with ill health and low educational qualifications.

Secondly, the overall employment rates for people with ill health are similar in Denmark and Sweden, even though Sweden has experienced worse economic conditions. Theoretically, we would expect those with ill health to be more disadvantaged in Sweden. The combination of low labour demand and strong EPL means that ‘cream-skimming’ should be more widespread in Sweden, at least compared to Denmark. Yet, the results do not correspond to this expectation, perhaps indicating that strong EPL is not very harmful after all.

Thirdly, the use of temporary work contracts is much more prevalent among people with ill health in Denmark. Here, people with LLSI are twice as likely to hold temporary work, compared with people reporting good health. The same pattern is present in Sweden as well, although to a lesser extent. Recall that temporary work contracts are more widespread in Sweden (Dølvik et al. 2015). This means that although the relative differences between people with good and ill health are greater in Denmark, the absolute number of people with ill health on temporary work contracts will be larger in Sweden. In Norway, the use of temporary work contract seems unrelated to health status. However, the current Norwegian government (elected in 2013) has decided to follow Sweden in making it easier for firms to hire temporarily (Dølvik et al. 2015), and the prevalence of temporary work contracts is thus expected to rise in the future. If so, people with health problems will likely be overrepresented in Norway as well, as their health status represents a risk from the employer's point of view.

Fourthly, because of cross-national differences in mobility rates. The comparatively high hiring probabilities for people with ill health and high education in Denmark are most likely a result of two important processes: First, a higher worker turnover rate overall in the Danish labour market (Madsen 2004; Andersen and Svarer 2007), and second, a high unemployment risk for people with ill health in Denmark during the recent economic downturn (Heggebø 2015). This means that people with health problems are in the ‘pool’ of potential candidates for a job opening to a large extent, and this could account for some of the cross-national differences in hiring likelihood.

How does the presented findings correspond to previous research on this topic? Reeves et al. (2014) found that stricter EPL was able to lower the firing likelihood for people with ill health in European countries experiencing less severe (or no) economic crisis. The current study shows that people with ill health are not particularly disadvantaged regarding hiring likelihood in Norway and Sweden, perhaps indicating that strong EPL is not necessarily harmful for ‘vulnerable groups’ on the labour market. Moreover, the positive results for people with ill health in Denmark are only visible among the higher educated. This is in line with McAllister et al. (2015), who show that the employment rates of people with low education and limiting long-standing illness was on a very low level in Denmark in 2000–2005, especially compared to Sweden. Apparently, the ‘flexicurity’ model is unable to integrate people with ill health and low education into the labour market.

As a whole, people with ill health seem to fare rather well on the labour market in Scandinavia, but there are some challenges too. For Norway, people with LLSI are somewhat disadvantaged regarding hiring likelihood, but the use of temporary work contracts is apparently unrelated to health status. For Sweden, individuals reporting health problems (LLSI and bad/fair health) have a lower hiring probability, and there is a noticeable ‘health component’ in the use of temporary work contracts as well. Thus, the results overall are quite negative in Sweden, perhaps reflecting the continuingly low labour demand during the investigated time window. The hiring outcomes are more favourable for people with bad/fair health in Denmark, but only among the higher educated. Furthermore, labour market attachment remains rather loose for people with ill health, as indicated by (i) more temporary work contracts and (ii) a high unemployment risk during the current economic downturn.

In conclusion, the ‘flexicurity’ model might lead to more people getting ‘a foot in the door’, but the final test is whether people with a negative signal (e.g. ill health) get a safe and permanent job. Denmark does not seem to pass the test.

This paper adds to the existing literature by investigating how institutional differences (EPL) are associated with employment prospects for people with health problems. An obvious strength of the current design is related to the similarity of the three Scandinavian countries, ensuring that cross-national heterogeneity is kept to a minimum. The EU-SILC data material is well suited for the present study because it includes health information, and because the data material is harmonised, which enables cross-country comparison of results.

Although the data material is harmonised, there are some potential pitfalls. It might be more legitimate to stay economically inactive for those with bad health in one of the countries, for instance. This seems rather implausible, however, since the Scandinavian countries share an emphasis on employment being an important mean for integration and social participation. In addition, public expenditure on active labour market policies is high in 2011 (1.93%, 0.56%, and 1.16% of GDP for Denmark, Norway, and Sweden, respectively), further strengthening the ‘work first’ approach. In fact, out of 30 OECD countries it was only the Scandinavian countries along with Poland and Switzerland who spent more on active than on passive labour market measures (OECD 2015b).

The differential demand for labour in Scandinavia (and other time trends) is dealt with through calendar year dummy variables. Still, two other differences could have an impact on the results. First, people with ill health report ‘disabled’ as economic status to a lesser extent in Sweden (see Table A1), perhaps indicating that their health problems are less serious on average. If so, the results could be biased (i.e. labour market attachment seems worse than it actually is). This is, however, unlikely to be the case, considering that the Swedish bad/fair health sample is somewhat ‘negatively selected’ on observable characteristics (see Table A2). The reason why Swedes less often report being disabled is probably related to the stricter eligibility criteria¹⁰ for disability benefits introduced in recent years (Hägglund 2013; Lidwall 2013). Second, unemployment benefits are more generous in Norway, perhaps giving people with ill health an incentive to stay unemployed. Yet, this seems unlikely according to the unemployment rate among people with LLSI in 2011: only 1.15% (women) and 3.21% (men) report unemployment as economic status (results available on request).

As mentioned above, people with ill health fare reasonably well overall on the Scandinavian labour market during the economic downturn. This means that other ‘vulnerable groups’ probably have experienced the main bulk of the disadvantages, such as immigrants and younger individuals. Remember that ill health correlates with old age, and older individuals are frequently protected by last-in-first-out seniority rules in Scandinavia (Lindbeck 1994; Von Below and Thoursie 2010). Thus, older individuals with health problems are rather unlikely to be dismissed, and will not have to apply for a new job. Recall, however, that people developing health problems tend to lose their jobs in Denmark (Heggebø 2015), indicating that seniority rules perhaps are insufficient when employment protection is weak.

Another challenge concerns the imprecise health measures used in this study. Unemployed and inactive people reporting ill health could be more/less healthy in one of the Scandinavian countries. Moreover, there might be cross-national differences in the degree of mismatch between educational qualifications and the job accepted by people with ill health (i.e. he/she might be forced to lower the ‘reservation wage’). It is difficult to conclude on these issues, since we do not directly observe the hiring decision in EU-SILC. Future research – preferably using experimental data – should try to dissect the relationship between ill health and labour market attachment even further.

Since the data material is longitudinal, it would have been possible to specify individual level fixed effects models in order to come closer to identifying a causal relationship. Unfortunately, there is not enough statistical power to run these models (i.e. very few individuals both change health status and gain employment). More importantly, the main aim of this paper is not to establish a causal link, but rather to investigate labour market attachment for people with ill health in Scandinavia. Furthermore, the EU-SILC data material is not well suited for the testing of which explanatory mechanisms (see Section 2.1) that are important for the health–employment status relationship. Lastly, it is important to stress that the investigated time window was quite ‘extreme’, in the sense that overall demand for labour was quite low throughout Scandinavia. Hence, the results of this study cannot necessarily be generalised to more booming economic conditions.

Earlier versions of this manuscript were presented at the Norwegian Sociological Associations’ annual meeting on 23 January 2015, at a workshop on economic crisis and population health in Oslo on 18 May 2015, and at an ESPAnet workshop on health impacts of social policy in Stockholm on 26 September 2015. I wish to thank Jon Ivar Elstad, Gunn Elisabeth Birkelund, Espen Dahl, Elisabeth Ugreninov, and the participants at these seminars for valuable suggestions. I would also like to thank the reviewers for excellent comments.

No potential conflict of interest was reported by the authors.

Kristian Heggebø has a master's degree in sociology (2012) from the University of Oslo. He is currently a Ph.D. fellow at Oslo and Akershus University College, on a project named ‘Health Inequalities, Economic Crisis and the Welfare State’. Research interests include labour market analysis, health sociology, educational attainment, and causal inference. His recent work has appeared in Social Science & Medicine, International Journal for Equity in Health, and European Sociological Review.