Depressed during the depression: has the economic crisis affected mental health inequalities in Europe? Findings from the European Social Survey (2014) special module on the determinants of health

Background: Economic crises constitute a shock to societies with potentially harmful effects to the mental health status of the population, including depressive symptoms, and existing health inequalities. Methods: With recent data from the European Social Survey (2006–14), this study investigates how the economic recession in Europe starting in 2007 has affected health inequalities in 21 European nations. Depressive feelings were measured with the CES-D eight-item depression scale. We tested for measurement invariance across different socio-economic groups. Results: Overall, depressive feelings have decreased between 2006 and 2014 except for Cyprus and Spain. Inequalities between persons whose household income depends mainly on public benefits and those who do not have decreased, while the development of depressive feelings was less favorable among the precariously employed and the inactive than among the persons employed with an unlimited work contract. There are no robust effects of the crisis measure on health inequalities. Conclusion: Negative implications for mental health (in terms of depressive feelings) have been limited to some of the most strongly affected countries, while in the majority of Europe persons have felt less depressed over the course of the recession. Health inequalities have persisted in most countries during this time with little influence of the recession. Particular attention should be paid to the mental health of the inactive and the precariously employed.

Introduction

Social scientists have had a longstanding interest in the implications of economic crises for health,¹ because such crises represent a shock to the affected societies with profound changes in the general social structure and thus, the social determinants of health. The theoretical expectations of these crisis effects are not straightforward. On the one hand, economic crises lead to increases in unemployment and income loss as well as reduced government funds² which could harm population health due to the elevated stress levels and the reduction in individual and societal coping resources (e.g. reductions in healthcare spending).³ On the other hand, economic downturns might reduce traffic, pollution and exposure to harmful work conditions (including stress).^4,5 Moreover, in times of crisis, individuals have less money for harmful health behaviors (e.g. smoking) and more time for health-promoting behaviors (e.g. exercise).⁶ as well as their social networks, an important resource for coping with stress.⁷ Finally, the cultural meaning of a time of crisis might reduce the health impacts of negative life events such as unemployment, because they can be interpreted as a collective experience rather than personal failure.⁸ Of course, these negative and positive effects can also cancel each other out, resulting in no clear ‘net effect’ of economic crisis.

Empirical evidence from various economic crisis provide mixed results of the relationship between macroeconomic changes and health.^2,4,9,10 The global financial crisis from 2007 has spurred new interest and empirical material for the investigation of crisis effects. Although some studies reported more mental health disorders, alcohol abuse,^11–13 a rise in poor self-reported health,^13,14 more infectious diseases, and increased suicide rates,^13–15 other studies found that health indicators and several health-promoting behaviors remained stable or improved in the period of the crisis.^3,6,16,17

Even though the economic crisis is a macro-level shock to societies as a whole, it may impact some groups more than others. Thus, we should be equally concerned with the potential influence of economic crisis on not only health, but also health inequality.¹⁸ Evidence from earlier economic downturns suggest that in the UK and Japan recessions have exacerbated health inequalities, while in Scandinavia inequalities have remained stable or even decreased during this time.¹⁹ This demonstrates the importance of cross-national comparative work on the effect of economic crisis on health inequalities. To our knowledge two studies have investigated the effect of the crisis on health inequalities in Europe:¹ Abebe et al.³ using panel data from the EU-SILC from 2005 and 2011 find no signs of a stronger effect of the crisis for self-rated health of the low educated and unemployed,² Buffel et al.¹⁷ used 2006 and 2012 data from the European Social Survey (ESS) to investigate the effect of the crisis on mental health for groups with different employment status. They find that persons in marginal part-time employment showed more depressive feelings in countries with a higher increase in unemployment as well as several groups of inactive men and homemaking women.

This study investigates the impact of the crisis on mental health inequalities in Europe—focusing on depressive feelings as one important indicator for low mental health. Our study advances the literature in several ways. First, by using three waves of the ESS including the first health module from 2014, we provide the most long-ranging perspective on the crisis so far. Having information from 2014 is crucial because the peak of the unemployment within the EU was in 2013 and thus was not included in earlier analysis. Second, previous analysis have focused on inequalities between educational groups^3,20 and groups with different employment status in the working-age population.¹⁷ Although the unemployed and persons with a precarious employment status are important and likely candidates for stronger impacts of the crisis, the crisis in Europe has also been strongly associated with (the threat of) austerity measures. Therefore, we investigate if persons from households that strongly depend on public benefits showed a stronger increase in depressive feelings. Third, we examined how the crisis affects persons that are economically vulnerable by looking at inequalities in depressive symptoms across income groups.

Methods

Data

The data used for this study are three waves of the ESS^21–23 including 21 European countries. All countries contain pre-crisis measurement from wave 3 (2006) and post-crisis-onset measurements from wave 6 (2012) and/or wave 7 (2014). The design of the ESS is based on strict random probability sampling and samples are representative for all persons aged 15 and over in the individual countries. All interviews were conducted face-to-face. Our analysis sample contains 106 158 respondents after we have deleted all respondents with missing values (2% of the ESS sample). As the number of missing cases is low, we expect bias resulting from listwise deletion to be negligible.²⁴

Variables

Depressive feelings were measured using the eight-item version of the Center of Epidemiological Studies-Depression (CES-D) scale.²⁵ Respondents are asked how often within the past week they (i) felt depressed, (ii) felt everything they did as effort, (iii) had restless sleep, (iv) were happy, (v) felt lonely, (vi) enjoyed life, (vii) felt sad and (viii) could not get going. There are four response categories: none or almost none of the time, some of the time, most of the time, all or almost all of the time.

We estimated measurement models to test different hypotheses about the dimensionality of this scale.^26,27 The one-dimensional (1D) model assumes that all eight items load together on one common factor, which we would label ‘generalized depression’. The two-dimensional model (2D) differentiates two factors ‘depressed affect’ consisting of items 1, 4, 5, 6, 7 (see above) and ‘somatic complaints’ including items 2, 3 and 8. Two of the eight items of the scale are reverse-worded which can lead to low scale-reliability and multi-factor solutions.²⁸ One of the reasons for these effects of reverse-worded items is careless respondent behaviour when all questions are answered with the same response without noticing the reverse-worded items. Such careless behaviour exhibits response bias which can be identified through zero variance around the mean score for individual respondents. Due to the potential negative implications for scale consistency,²⁸ we excluded the 1551 observations (1.5% of the analytical sample) with no variance on the eight depression items.

In addition, we used different modeling strategies for the two reverse-worded items. First, we test a three-factor solution (3D) that differentiates within the factor affect between ‘depressive affect’ with the three negative items 1, 5 and 7 and ‘positive affect’ including items 4 and 6. An alternative to account for the different polarity of these items is the specification of a covariance between the error terms of items 4 and 6 in the 1D or 2D solution. We compare these five different models in a multi-group confirmatory factor analysis framework in order to determine the best-fitting solution for the ESS-data.

Three independent variables are used to identify social inequalities in health in depressive feelings. First, information on ‘employment status’ is used to identify vulnerable groups in the labour market such as the unemployed and persons in precarious employment. Our categorical measure distinguishes between five groups: (i) Employed persons with unlimited contract, (ii) self-employed persons or persons working for a family business, (iii) persons in precarious employment (with no or a limited work contract), (iv) unemployed persons and (v) inactive persons covering those who are in education, retired, in community or military service, the (permanently) sick or disabled, and homemakers. Second, for the independent variable ‘subjective income’ we relied on the question ‘Which of the descriptions on this card comes closest to how you feel about your household’s income nowadays?’(We use subjective rather than objective income, because it has been measured consistently across countries and time periods. The measurement of objective income has changed over the waves in the ESS and there are many cross-national and—temporal income definitions even in the most recent waves. Therefore, while technically feasible a reliable and valid harmonization of objective income is extremely challenging. Moreover, a higher number of missing cases (20 compared with <1%) would require the use of multiple imputation for the harmonized variable. Since we include a larger number of waves and specifically analyse trends, we are convinced that subjective income is the most adequate operationalization of income groups in the context of the ESS data.). The four response categories are: (i) living comfortably on present income, (ii) coping on present income, (iii) finding it difficult on present income and (iv) finding it very difficult on present income. Due to low number of cases in some countries, we combined the latter two categories into one. Subjective income is used because the objective income measure changed between the waves and a valid harmonization is not available. Third, particularly within Southern Europe, the crisis has led to cuts of public benefits. Therefore, we look at persons whose primary income sources are based on public benefits compared with those with other primary income sources. All models also include sex, age, and education as control variables. Age is categorized in seven groups. Education, as an important predictor of both mental health²⁹ and employment status/income, is measured in years.

In order to directly assess the impact of the crisis we use the change in unemployment rate^4,17,30 and GDP per capita between the later waves and the pre-crisis year 2006. Since the impact of the crisis may depend on the previous situation in the economy, we also control for the pre-crisis GDP per capita and unemployment level as time-invariant macro-indicators. The macro indicators were extracted from the Eurostat database.

Appendix Table A1 reports descriptive statistics.

Estimation methods

Previous studies have tested measurement invariance for the CES-D 8 scale with ESS data across men and women with the 2006 wave.²⁶ The set-up of this study required an assessment of measurement invariance across socio-economic groups as well as waves. Multi-group Confirmatory factor analysis was estimated for each socioeconomic group and the country indicator to assess the cross-cultural invariance of the scale, and for each socioeconomic group and the wave indicator to assess inter-temporal invariance of the scale. The analyses were conducted with maximum-likelihood estimation using STATA Version 13.1. Goodness-of-fit statistics were used to assess and compare model specification. Chi-square tests, the traditional way of assessing goodness-of-fit, are heavily influenced by sample size and have proven not robust in large samples like ours. Therefore, while we report chi² values, we based or assessment on more robust absolute fit indices (Root-Mean-Square-Error of Approximation—RMSEA, Comparative Fit Index—CFI, Tucker-Lewis-index (TLI).³¹

In a second step, we use the predicted latent depression scale as a dependent variable in a multilevel growth curve analysis. Due to the repeated cross-sectional design we specified our multi-level model by considering the individuals (level 1) nested in 59 country years (level 2) nested in 21 countries (level 3). Since the time trends are of specific interest for the assessment of the crisis effects, year dummies are added to the model. Interactions between year dummies and socio-economic groups are used to assess trends in health inequalities.

Results

Model fit and measurement invariance

Table 1 reports the model comparison of the different configurations of the eight CES-D items. Due to the large number of observations, all models show a significant chi², but the three models that take the positive phrasing of the two items ‘enjoyed life’ and ‘were happy’ into account (models 2, 4, and 5) have a good model fit with the CFI and the TLI above .95 and the RMSEA below 0.06. These indices also indicate that models 4 and 5 fit the data better than model 2. However, in line with an earlier analysis of the ESS data,²⁴ the different model dimensions are highly correlated. Thus, the squared correlations (SCs) are higher than the average extracted variance (AVE) which indicates low discriminant validity. As the 1D specification with correlated errors (model 2) still shows a good fit to the data and is more parsimonious for the specification and the presentation of results, we proceed with this model.

Table 1

Goodness-of-fit statistics and discriminant validity tests for five measurement models of the CES-D in the ESS

	Model Fit: Goodness-of-fit statistics					Discriminant validity
Model	Chi2	df	RMSEA	CFI	TLI	AVE	SC (Depaffect, Somcomp)	SC (Depaffect, Posaffect)	SC (Posaffect, Somcomp)
1: 1D	28357	20	0.115	0.891	0.847	0.387	–	–	–
2: 1D, correlated errors	6202	19	0.055	0.976	0.965	0.379	–	–	–
3: 2D	24839	19	0.111	0.904	0.859	0.430 0.390	0.763	–	–
4: 2D, correlated errors	3534	18	0.043	0.986	0.979	0.410 0.390	0.790	–	–
5: 3D	3529	17	0.044	0.986	0.978	0.500 0.390 0.603	0.790	0.457	0.354

Source: ESS (2006, 2012, 2014), pooled sample, own calculations, RMSEA, root-mean-square-error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis-index; AVE, average variance extracted; SC, squared correlation, 1D = one-dimensional, 2D = two-dimensional, 3D = three-dimensional.

Next, we assessed measurement invariance across socioeconomic groups by wave or country for the pooled sample (see Table 2). The top of the left panel shows that both for the entire sample and for the different socio-economic groups configural, metric and scalar invariance can be assumed across waves, because all indices show a good fit (RMSEA < 0.06, CFI > 0.95, TLI > 0.95). In many configurations, the fit indices even indicate a better model fit when factor loadings are assumed identical (metric invariance) or even when item intercepts are constrained to be equal (scalar invariance). Below are the invariance tests between countries and socioeconomic groups. The indices indicate that there is both configural and metric invariance across countries (and socioeconomic groups) with either good or acceptable fit (RMSEA < 0.08, CFI > 0.9, TLI > 0.9) to the data. However, none of the models shows an acceptable fit when item intercepts are constrained to be equal (scalar invariance). Since the aim of this study is the analysis of time trends and associations, metric invariance is sufficient. A further indication that time trends can be compared across countries is that scalar invariance across years can be found also when tested for each individual country (In Hungary, the overall model fit cannot consistently be seen as acceptable. However, both RMSE and CFI suggest a better model fit if factor loadings and item intercepts are constrained.)

Table 2

Invariance tests

	invariance test, pooled sample
		Chi2	df	RMSEA	CFI	TLI			Chi2	df	RMSEA	CFI	TLI
Year	C	6395	57	0.056	0.976	0.964	AT	C	330	38	0.062	0.949	0.965
	M	6554	71	0.051	0.975	0.970		M	364	45	0.060	0.953	0.962
	S	6930	85	0.048	0.974	0.974		S	418	52	0.060	0.953	0.957
Year X Subjective Income	C	6786	171	0.057	0.970	0.955	BE	C	567	57	0.071	0.939	0.958
	M	7448	227	0.052	0.967	0.963		M	586	71	0.064	0.950	0.958
	S	8530	283	0.050	0.962	0.966		S	662	85	0.062	0.954	0.953
Year X State Income	C	6469	114	0.056	0.974	0.962	BG	C	497	38	0.084	0.947	0.964
	M	7048	149	0.051	0.972	0.968		M	555	45	0.081	0.950	0.960
	S	7888	184	0.048	0.968	0.971		S	596	52	0.078	0.954	0.957
Year X Employment status	C	6677	285	0.056	0.974	0.962	CH	C	346	57	0.057	0.950	0.966
	M	7331	383	0.050	0.972	0.969		M	362	71	0.051	0.960	0.966
	S	8870	481	0.049	0.966	0.970		S	412	85	0.049	0.962	0.962
Country	C	9161	399	0.066	0.966	0.949	CY	C	179	38	0.061	0.961	0.974
	M	12486	539	0.066	0.953	0.949		M	198	45	0.058	0.964	0.971
	S	28778	679	0.090	0.890	0.905		S	279	52	0.066	0.954	0.958
Country X Subjective Income	C	10291	1197	0.067	0.960	0.941	DE	C	667	57	0.061	0.943	0.962
	M	14283	1631	0.068	0.944	0.940		M	744	71	0.057	0.950	0.958
	S	32000	2065	0.092	0.869	0.888		S	910	85	0.058	0.949	0.948
Country X State Income	C	9642	798	0.066	0.964	0.946	DK	C	302	57	0.055	0.951	0.967
	M	13678	1085	0.068	0.948	0.944		M	335	71	0.051	0.958	0.964
	S	31380	1372	0.093	0.876	0.894		S	391	85	0.05	0.959	0.959
Country X Employment status	C	11222	1995	0.067	0.962	0.944	EE	C	492	57	0.063	0.953	0.968
	M	15855	2723	0.069	0.946	0.942		M	521	71	0.058	0.961	0.967
	S	34966	3451	0.095	0.871	0.890		S	576	85	0.055	0.965	0.964
ES	C	492	57	0.063	0.967	0.977	FI	C	340	57	0.049	0.960	0.973
	M	489	71	0.057	0.971	0.976		M	378	71	0.046	0.965	0.971
	S	553	85	0.055	0.973	0.973		S	457	85	0.046	0.965	0.964
FR	C	734	57	0.078	0.928	0.951	GB	C	663	57	0.069	0.945	0.963
	M	763	71	0.071	0.941	0.950		M	691	71	0.062	0.955	0.962
	S	805	85	0.066	0.949	0.948		S	714	85	0.057	0.962	0.962
HU	C	1121	57	0.106	0.909	0.938	IE	C	558	57	0.065	0.954	0.969
	M	1172	71	0.097	0.925	0.936		M	588	71	0.059	0.962	0.968
	S	1272	85	0.092	0.932	0.931		S	643	85	0.056	0.966	0.965
NL	C	735	57	0.080	0.921	0.946	NO	C	413	57	0.063	0.935	0.956
	M	758	71	0.073	0.936	0.946		M	445	71	0.058	0.945	0.953
	S	833	85	0.069	0.941	0.941		S	498	85	0.055	0.949	0.948
PL	C	575	57	0.073	0.953	0.968	PT	C	536	57	0.069	0.959	0.972
	M	590	71	0.065	0.962	0.968		M	582	71	0.064	0.965	0.970
	S	712	85	0.066	0.962	0.961		S	783	85	0.068	0.960	0.959
SE	C	366	57	0.055	0.959	0.972	SI	C	382	57	0.067	0.949	0.966
	M	375	71	0.049	0.968	0.973		M	430	71	0.063	0.955	0.962
	S	450	85	0.049	0.968	0.967		S	469	85	0.060	0.960	0.959
SK	C	219	38	0.054	0.957	0.971
	M	243	45	0.052	0.960	0.968
	S	293	52	0.053	0.958	0.961

Source: ESS (2006, 2012, 2014), own calculations, RMSEA, Root-mean-square-error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis-index; C, configural invariance; M, metric invariance; S, scalar invariance; numbers in italics indicate less than good model fit.

Trends in depression during the economic crisis

Figure 1 shows trends in depressive feelings for the individual countries as a whole and three vulnerable groups. In the vast majority of the countries, the general population reported fewer depressive feelings in 2012 than in 2006. Only the trends in Cyprus and Spain indicate a rise in depressive feelings. Most countries also show a further decline in depressive feelings between 2012 and 2014. Three Nordic countries (Denmark, Norway, and Sweden) as well as Portugal, and Slovenia have a higher average of depressive feelings in 2014 than in 2012.

Figure 1

Trends in depressive feelings (0-24) between 2006 and 2015, by country (sorted by increase in unemployment rate). Source: ESS (2006, 2012, 2014), own calculations, weighted means, AT, Austria; BE, Belgium; CH, Switzerland; DE, Germany; FR, France; NL, Netherlands; DK, Denmark; FI, Finland; NO, Norway; SE, Sweden; CY, Cyprus; ES, Spain; PT, Portugal; GB, Great Britain; IE, Ireland; BG, Bulgaria; EE, Estonia; HU, Hungary; SI, Slovenia; SK, Slovakia.

If we compare the trend of the vulnerable population groups with the overall trend, we get a first impression of how inequalities developed during the economic crisis. Inequality trends are rather complex. In the four countries with the strongest increase in unemployment rates (Portugal, Ireland, Cyprus and Spain) inequalities seemed to have decreased between 2006 and 2012 and either remained stable or increased again, thereafter. In a number of countries particularly the unemployed seem to have caught up to the general population average during the crisis (e.g. in Austria, Sweden, Finland). However, in Hungary depressive feelings of the unemployed have substantially risen between 2006 and 2012 compared with a declining trend of the overall population. In sum, over the course of the recession, inequalities generally have neither increased nor decreased, but there are indications that in the ‘crisis countries’ inequalities initially might have become smaller.

In the multivariate three-level growth curve analysis (see Table 3), we find that for the pooled sample depressive feelings decreased in 2012 (β = −0.046⁺) and even stronger in 2014 (β = −0.077^***). Countries with a higher GDP per capita before the crisis exhibit a lower level of depressive feelings (β = −0.008⁺). The impact of the crisis operationalized by the change in unemployment rates shows a positive association with depressive feelings (β = 0.008^***) (Model 1). This effect seems to be mainly due to changes in the composition of income and employment groups, because the effect is no longer significant in Model 2. In addition, sensitivity analysis excluding one country at a time shows that the effect of the crisis is sensitive to the inclusion of individual countries.

Table 3

Three-level linear growth curve regression of depressive feelings

	Model 1	Model 2	Model 3	Model 4
Year (Ref. 2006)
2012	−0.046 + (0.024)	−0.044 + (0.023)	−0.035 (0.025)	−0.036 (0.026)
2014	−0.077*** (0.017)	−0.067*** (0.017)	−0.073*** (0.020)	−0.074*** (0.021)
Macro variables
GDP per capita (pre)	−0.008*** (0.002)	−0.005*** (0.002)	−0.005*** (0.002)	−0.005*** (0.002)
Unemployment (pre)	0.007 (0.009)	0.005 (0.008)	0.005 (0.008)	0.005 (0.008)
Change in GDP	0.005 (0.004)	0.008 + (0.004)	0.007 + (0.004)	0.007 + (0.004)
Change in unemployment	0.008*** (0.002)	0.003 (0.002)	0.003 (0.002)	0.003 (0.003)
Individual variable
Subjective income (Ref. living comfortably)
Coping on present income		0.122*** (0.013)	0.123*** (0.015)	0.123*** (0.015)
Difficult on present income		0.441*** (0.014)	0.444*** (0.017)	0.445*** (0.016)
Employment status (Ref. employed with unlimited contract
Inactive		0.099*** (0.008)	0.091*** (0.011)	0.091*** (0.011)
Unemployed		0.139*** (0.011)	0.141*** (0.018)	0.141*** (0.018)
Precariously Employed		0.077*** (0.010)	0.068*** (0.014)	0.067*** (0.014)
Self-employed/family business		0.008 (0.010)	0.004 (0.014)	0.003 (0.014)
Household income source (Ref. not primarily public benefits)
Primarily from public benefits		0.098*** (0.025)	0.112*** (0.026)	0.111*** (0.026)
Interaction effects
Public income × 2014			−0.048** (0.016)	−0.055*** (0.017)
Inactive × 2014			0.036** (0.014)	0.037** (0.014)
Precarious employment × 2014			0.038* (0.019)	0.041* (0.020)
Public income × change in unemployment				0.004 + (0.003)
Low subjective income × change in unemployment				−0.004 + (0.002)
Constant	0.312** (0.189)	−0.127 (0.097)	−0.124 (0.097)
Variance
Country	0.006	0.002	0.002	0.002
Country-year	0.002	0.001	0.001	0.001
Individual	0.414	0.379	0.379	0.379
N	106,158	106,158	106,158	106,158

Source: ESS (2006, 2012, 2014), pooled sample, own calculations, SE, standard errors in parentheses, significance levels: *P < 0.05, ^**P < 0.01, ^***P < 0.001, models are controlling for sex, age in categories, and years of education; random effects for income, employment status and income source are specified both on the country and the country-year level and are all significant.

In Model 2, the individual-level variables were added. Significant inequalities in depressive feelings are found for all three indicators. Individuals with a lower subjective income show significantly more depressive feelings (β = 0.122^***; β = 0.441^***). Compared with employed persons with an unlimited work contract, inactive (β = 0.099^***), unemployed (β = 0.139^***), and precariously employed persons (β = 0.077^***) felt more depressed. Finally, even when controlling for the income level, persons whose household income stems primarily from public benefits report significantly more depressive feelings (β = 0.098^***).

In order to investigate potential crisis effects on inequalities, we estimate interactions both with the time trend to see if trends developed differently across the social groups, and with the change in unemployment rate (model 3). Only significant interactions are reported. We find that differences in depressive feelings between those who primarily rely on public benefits for their income and those who do not have become smaller between 2006 and 2014, because of a stronger downward trend of those with primarily public income (β = −0.048^***). In contrast inequalities in employment status (reference: employed with unlimited work contract) have increased for the inactive (β = 0.036^**) and the precariously employed (β = 0.038*).

A direct test of the effect of the crisis is performed in model 4 which includes interactions between the change in unemployment rate and the socio-economic variables. A larger increase in unemployment is associated with larger inequalities between persons relying mainly on public benefits and those who do not (β = 0.004⁺). In contrast, a larger increase in unemployment is associated with less inequality in depressive feelings between the employed and the unemployed (β = −0.004⁺). However, both interactions are only marginally significant.

Discussion

The 2007 banking crisis had a strong impact on the European economy in the following years. As a result, both researchers and policymakers are concerned about the implications of the crisis for overall population health and health inequalities. Several studies have documented negative health implications in the Southern European countries that were both strongly hit by the crisis, and implemented austerity measures during the recession.^11,14,20,32 The aim of this article was to investigate if an impact of the crisis on health can be found across Europe with most up-to-date data (2006–14) that cover the whole development of the recession including the peak in unemployment at 2013.

First, has the economic crisis affected the level of depressive feelings in Europe? No. Our pan-European analysis indicates that the economic crisis was not associated with a general increase in depressive feelings, since both the overall trend showed a decline in such feelings and the direct effect of crisis (change in unemployment) was not robust. Second, has the economic crisis resulted in larger health inequalities? Not generally. We found no systematic effect of the recession on health inequalities across Europe, except that in the crisis countries, some inequalities decreased rather than increased.

Thus, our study adds a longer-term perspective and confirms Buffel et al.’s¹⁷conclusion that the crisis had no overall negative effect on mental health in Europe. However, in contrast, we do not find effects of the unemployment rate on inequalities between groups with different employment status nor for the additional vulnerable groups that we analysed (based on low subjective income or public-benefit based income). Thus, our conclusion with respect to inequalities are in line with the panel-data results based on the EU-SILC(3) which also concluded that inequalities are stable and have not been affected by the crisis.

There are some limitations to the data and analysis presented here. First, the study used a research design with repeated cross-sectional data which is limited in the identification of causal effects. However, a comparison of individual health trajectories is not possible with the existing cross-national panel data (e.g., EU-SILC).³ Second, the ESS sample unfortunately excludes several countries that were strongly affected by the economic crisis, viz. Greece, Iceland, and Romania. Third, bias may arise through selective non-response, particularly as depressive feelings might lower the probability of survey participation. Moreover, persons with a lower socio-economic status are less likely to participate. Since selective non-response might have been even stronger after the crisis, we might underestimate health inequalities after the recession, particularly in the countries that have been most strongly affected by the crisis. Fourth, the subjective income measure we use may not concur with objective income measures and the crisis could lead to respondents overestimating financial difficulties. If this was the case, we may not be able to measure the full extent of health inequalities between income groups during the peak of the crisis. Finally, the ESS includes only one measure of depressive feelings before the crisis so that we do not know how they developed up until then. Thus, we might miss a deceleration of mental health improvements during the crisis.

Limitations acknowledged, we believe our study carries several implications. For Europe as a whole, health inequalities present a consistent public health issue independent of the economic recession. Our study suggests that the health disadvantage of the inactive and precariously employed has increased in the last decade and thus interventions should focus particularly on these two groups.

Acknowledgments

Terje A. Eikemo, Clare Bambra and Tim Huijts led the design of the ESS special module on the social determinants of health in coordination with Rory Fitzgerald of the ESS.

Funding

This article is part of the HiNEWS project—Health Inequalities in European Welfare States—funded by NORFACE (New Opportunities for Research Funding Agency Cooperation in Europe) Welfare State Futures programme (grant reference:462-14-110). For more details on NORFACE, see http://www.norface.net/11.

Conflicts of interest: None declared.

Appendix

Table A1

Descriptive statistics by country

	AT	BE	BG	CH	CY	DE	DK	EE	ES	FI	FR	GB	HU	IE	NL	NO	PL	PT	SE	SI	SK
Depression scale	5.1	5.2	7.0	4.5	5.2	5.6	4.5	6.5	5.9	4.6	5.2	5.3	7.7	4.4	4.7	4.0	5.9	6.9	4.7	5.2	6.8
Female	0.5	0.5	0.6	0.5	0.5	0.5	0.5	0.6	0.5	0.5	0.5	0.5	0.6	0.5	0.5	0.5	0.5	0.6	0.5	0.5	0.5
18–24	0.2	0.2	0.1	0.1	0.2	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.2	0.2	0.1	0.1	0.1	0.1
25–34	0.1	0.1	0.1	0.1	0.2	0.1	0.1	0.1	0.2	0.1	0.1	0.1	0.1	0.2	0.1	0.1	0.2	0.1	0.1	0.1	0.2
35–44	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.1	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2
45–54	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2
55–64	0.1	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.1	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2	0.2
65–74	0.1	0.1	0.2	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1
75–84	0.1	0.1	0.1	0.1	0.0	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.0	0.1	0.1	0.1	0.1	0.1	0.1	0.0
85 and older	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
Income
Living comfortably	0.3	0.4	0.0	0.5	0.2	0.3	0.7	0.1	0.3	0.2	0.3	0.4	0.1	0.3	0.5	0.6	0.1	0.1	0.6	0.4	0.1
Coping	0.5	0.4	0.2	0.3	0.5	0.5	0.3	0.6	0.5	0.6	0.5	0.4	0.5	0.5	0.4	0.3	0.6	0.5	0.3	0.4	0.5
(Very) difficult	0.1	0.2	0.7	0.1	0.4	0.1	0.1	0.3	0.2	0.1	0.2	0.2	0.5	0.2	0.1	0.1	0.3	0.4	0.1	0.2	0.4
State income	0.2	0.3	0.3	0.2	0.2	0.3	0.3	0.3	0.3	0.4	0.3	0.3	0.4	0.3	0.3	0.2	0.3	0.3	0.3	0.3	0.2
Employment status
Inactive	0.4	0.5	0.5	0.4	0.4	0.5	0.4	0.4	0.4	0.4	0.4	0.4	0.5	0.5	0.5	0.4	0.5	0.5	0.4	0.5	0.4
Unemployed	0.0	0.1	0.1	0.0	0.1	0.0	0.0	0.0	0.1	0.1	0.1	0.0	0.1	0.1	0.0	0.0	0.1	0.1	0.0	0.1	0.1
Employed with unlimited contract	0.4	0.4	0.3	0.4	0.2	0.4	0.4	0.4	0.3	0.4	0.4	0.3	0.4	0.2	0.4	0.5	0.3	0.3	0.4	0.3	0.4
Precarious employment	0.1	0.1	0.1	0.0	0.2	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1
Self-employed, family business	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1
Unemployment rate	5.3	8.3	9.0	3.8	4.6	10.1	3.9	5.9	8.5	7.7	8.8	5.4	7.5	4.5	5.0	3.4	13.9	8.9	7.1	6.0	13.5
hange in unemployment rate	0.1	-0.2	2.1	0.2	3.9	-3.3	2.1	2.2	10.8	0.3	0.8	1.0	1.4	6.5	1.1	0.0	-2.9	3.9	0.6	2.1	0.3
GDP per capita	32.2	31.0	3.6	45.5	21.5	29.5	41.5	10.0	22.7	32.8	29.2	33.9	9.1	43.3	35.4	59.1	7.2	15.8	36.9	15.7	8.4
GDP growth	2.1	1.3	2.6	2.4	0.8	1.9	1.7	5.7	1.0	1.5	1.1	2.3	1.6	3.5	1.1	2.4	3.6	-0.7	2.3	2.2	4.9

Source: ESS (2006, 2012, 2014), pooled sample, own calculations, weighted means and proportions.