Income inequality and foregone medical care in Europe during The Great Recession: multilevel analyses of EU-SILC surveys 2008–2013

The study utilized survey data from the European Union Statistics on Income and Living Conditions program (EU-SILC). EU-SILC provides rich information from large population samples, obtained through harmonized surveys conducted yearly [33, 34]. Six survey rounds (2008–2013) with data from 30 countries were analyzed (all European Union (EU) member states except Croatia, plus the three non-member countries Iceland, Norway, and Switzerland). Individuals aged 30 to 59 years were selected for the analyses. The very few respondents with missing values on relevant variables were excluded from the analyzed samples which number more than one million respondents (N = 1,242,361).

In the EU-SILC surveys, respondents were asked whether they “really needed” medical examination or treatment, but did not get it on at least one occasion during the last 12 months. If “yes”, respondents were asked to state the reason by choosing from a list of eight alternatives [33]. The dichotomous outcome analyzed in this study, foregone medical care, indicates self-reported unmet need for medical care either because of costs, waiting lists, or travel difficulties. This variable, also used in previous research [e.g. 28, termed “enforced unmet need”], is likely to reflect difficulties in access to medical care due to circumstances associated with an economic crisis, such as austerity, insufficient supply of health care, higher co-payments, and lack of household economic resources. Among the three alternative reasons for foregone medical care, “too expensive” was the most common [35].

Other individual-level variables were gender, age, educational level (three categories low, medium, and high), and total disposable household income [33]. Real Gross Domestic Product (GDP) per capita [36] was used for characterizing the country’ overall economic level, and change in this variable indicates macro-economic growth or decline. Income inequality was indicated by Eurostat’s calculations of the S80/S20 ratio, i.e., the ratio of total income received by the 20 % of the population with the highest income (top quintile), to the total income of the fifth with lowest income (lowest income quintile) [37, 38].

As equity in health care is a central issue, foregone medical care among less privileged population categories was especially examined. Each country sample was therefore divided into two parts. The Disadvantaged were defined by two criteria: they were in the lowest income tertile (the lower third of the income distribution in the country sample, age 30–59, in the given survey), and they reported health difficulties in terms of either a long-standing (chronic) disease or self-rated overall health status as fair or bad. Reports and statistics (see, for instance, [27]) indicate that in almost all European countries, foregone medical care is seldom reported by middle-income and high-income individuals [31]. Moreover, although everybody will need medical assistance from time to time, delays in obtaining medical care may be quite unproblematic for those in good health. A crucial test, so to speak, for a healthcare system is therefore whether people with limited economic resources whose well-being depends on access to medical services, have few difficulties in obtaining medical care. The category Disadvantaged is intended to represent this vulnerable part of the population, in so far as possible with the available information in the EU-SILC surveys. However, also the remaining parts of the samples, termed Others, were analyzed in order to address more broadly how foregone medical care has developed in European populations during The Great Recession.

First, data were described, and percentages reporting foregone medical care in each country, both among Disadvantaged and Others, are shown. Since sample distributions could vary randomly from year to year, the surveys were pooled in pairs. Changes during The Great Recession were indicated by the 2008/2009 and 2012/2013 years, i.e. roughly the pre-crisis and “peak-of-crisis” situation. The economic downturn started earlier in some countries (e.g., Iceland) than in others (e.g., Greece) [36]. The EU-SILC surveys indicate that foregone medical care was at its lowest level in 2009 [27].

Second, three-level regression models were analyzed, which correspond to the hierarchical data structure: cross-sectional survey samples (individual level), conducted at different time-points (level of country-years), nested within countries (country level). The country level variables were coded as centred means for the period under study, while the country-years variables measured deviations from these country-specific means at different time points. This allows for simultaneous estimations of the effects of both time-invariant and time-varying country characteristics (see [39, 40] and [41‐44]). Linear probability regression was used. Logistic regression is often preferred for dichotomous outcomes, as it avoids some difficulties with linear probability models, but the two techniques give mostly very similar findings [45]. Linear probability models have, on the other hand, some advantages. Results are more easily interpreted than logits and odds ratios, as the linear probability coefficients represent the change in proportions (or percentages) reporting foregone medical care given a one-unit increase in the explanatory variable. Moreover, statistical packages have often difficulties in fitting logistic models in large samples, especially when three levels (as in this study) have to be handled.

At the country level, two variables with constant values during the study period were used: Average Real GDP per capita 2007–2012 [36], measured in fixed Euros and recalculated into natural logarithms, and average income inequality, i.e. the mean S80/S20 ratio 2007–2012 [38]. Both variables were centred around the mean value for the 30 countries.

At the next country-years level the variable GDP deviation indicated how much the country’s GDP deviated from the country average 2007—2012 at three time points: 2007/2008, 2009/2010, and 2011/2012. The unit of this variable was ten percent decline in GDP per capita, implying, for example, that a value of +0.8 for 2011/2012 means that Real GDP per capita was eight percent lower than the 2007—2012 average. Thus, the coefficient for this variable indicates how much the proportion experiencing foregone medical care could be expected to change if GDP per capita decreased by 10 %. The data file was arranged so that GDP deviation in 2007/2008 corresponded to the individual-level information in 2008/2009, and so on (i.e. GDP deviation 2011/2012 corresponded to the 2012/2013 surveys). The rationale for this was an assumption about a likely time lag of about 1 year between a macro-economic change and possible effects on access to medical care.

Similarly, the variable Income inequality deviation measured how much the S80/S20 ratio in 2007/2008, 2009/2010, and 2011/2012, respectively, deviated from the average S80/S20 ratio for the 2007—2012 period.

The outcome variable foregone medical care was measured at the individual level (for 2008/2009, 2010/2011, and 2012/2013), and at this level there are also individual controls: gender (women = 1), age (in years, centred around the midpoint 45 years = 0), age squared, dummy variables for low and high education (reference = medium education), and two dummy variables which indicated whether the respondent was unemployed at the time of interviewing, and whether he/she lived in a rural area.

Intercept variances and random slope variances at the country level for gender, age, and low education were estimated [44]. To include other random slope effects was not feasible since they resulted in non-converging models. Estimations were made by means of MLwiN, version 2.32, run from within Stata [46].

Table 1 presents the samples and reports some country-level indicators. On average, 14.7 % of the survey samples aged 30–59 were classified as Disadvantaged, varying from 9.3 % in Romania to 22.2 % in Latvia. This percentage varied little within each country during the study period. The level of income inequality was indicated by the average S80/S20 ratio during 2007–2012, which varied from lowest (3.30) in Slovenia to 7.27 in Latvia. Changes in the S80/S20 ratio during 2007–2012 were quite small (not shown in table). Average Real GDP per capita varied about ten-fold, from lowest in Bulgaria and Romania (less than 6,500 Euros) to highest in Switzerland, Norway, and Luxembourg. The impact of The Great Recession is indicated in the right-hand column which shows that Real GDP per capita declined from 2007/2008 to 2011/2012 in 21 of the 30 countries. In the “declining” countries, GDP was usually reduced from two to ten per cent (16 countries); Greece was exceptional with a GDP decline of 22 %. Only Poland had a substantial GDP increase from 2007/2008 to 2011/2012.

The correlation coefficients (Pearson’s r) between the country indicators (not shown in table) provide interesting information. “Richer” countries tended to have less income inequality: The correlation between average GDP per capita and average S80/S20 ratio 2007–2012 was −0.49 (p-value <0.01). Income inequality was on the other hand practically unrelated to GDP change from 2007/2008 to 2011/2012 (r = 0.09, p-value 0.65). The correlation between average GDP and GDP change was insignificantly negative (−0.19, p-value 0.31). Thus, neither egalitarian countries nor “rich” countries seemed to be more successful in avoiding GDP decline during the crisis years than other European countries.

Figure 1 and Table 2 display the reporting of foregone medical care in the Disadvantaged and Others subsamples. Among the Disadvantaged, the percentage reporting foregone medical care increased from 8.3 to 10.1 % during The Great Recession (unweighted country average), while it rose from 2.1 to 2.6 % among Others (Fig. 1). Reporting foregone medical care because it was perceived as “too costly” was especially prevalent among Disadvantaged. Table 2 gives country-specific information. Two well-known patterns [27, 28, 30, 31] can be noted – the huge between-country differences, and the ubiquitous social inequalities. In 2012/2013, for example, foregone medical care in the Disadvantaged subsamples varied from 0.1 % in Slovenia to 35.4 % in Romania. Moreover, in every country and in both periods, percentages reporting foregone medical care were higher (sometimes very much higher) among Disadvantaged than Others – only one exception: Slovenia in 2012/2013.

For this paper, the changes during the crisis years are the central interest, however. Figure 1 and Table 2 indicate both deteriorated access and increasing social inequalities during The Great Recession. Among Disadvantaged, foregone medical care increased in 21 countries (significantly, i.e., p-values < 0.05, in 17 countries), was stable in United Kingdom, and decreased in eight countries. Among Others, foregone medical care also increased in 21 countries (although often marginally), while three countries saw no changes, and foregone medical care decreased, usually just a little, in six countries. Table 2 reveals larger increases in foregone medical care (in percentage points) among Disadvantaged than among Others in the majority of the countries. The differences between Disadvantaged and Others increased by more than 0.5 percentage points in 18 of the 30 countries, decreased by at least 0.5 points in seven countries, and remained relatively stable in five countries.

Table 3 reports the results from the multilevel analyses of the Disadvantaged subsamples. As the individual-level coefficients are of less interest for this paper, comments will focus on the country and country-years variables.

In Model 1, the individual-level predictors (both as fixed and random effects), the country’s GDP level (natural logarithms of Real GDP per capita, average 2007–2012), and country level of income inequality (S80/S20 ratio, average 2007–2012), were included. This model indicated a significant effect (using the p < 0.05 criterion) of the country’s average economic level: For each log unit increase in Real GDP per capita, the proportion Disadvantaged reporting foregone medical care could be expected to fall by 0.033, i.e. 3.3 percentage points. The income inequality coefficient (0.024, p-value < 0.01) indicated moreover that, adjusted for respondents’ characteristics and average GDP, income inequality was markedly associated with the level of foregone medical care during these years. For each step “up” on the scale of income inequality, the percentage among Disadvantaged who experienced foregone medical care could be expected to become 2.4 points higher.

Model 2 added the country-years variables, i.e., GDP and income inequality deviations from the country means at the three time points. This model indicated whether changes in GDP and/or income inequality had any impact on the level of foregone medical care, net of the impact of the 2007–2012 mean of these variables. The effect of Income inequality deviation was insignificant, but the GDP deviation coefficient (0.015, p-value < 0.01) indicated that overall, a 10 % decline in GDP corresponded to an increase in foregone medical care among the Disadvantaged of 1.5 percentage points.

In Table 4, the results from identical multilevel models among Others are reported. The large number of Others (1,059,155 respondents) could not be handled by the available MLwiN version, and a random sample of 400,000 Others was selected, as trials showed that the statistical software could handle this sample size. In Table 4, the impact of country-level characteristics on foregone medical care appears as different among Others than for the Disadvantaged subsamples. The coefficient for GDP level was, in all three models, +0.007, implying a statistically significant, but weak and surprising tendency that foregone medical care among Others occurred more frequently the higher the country’s GDP per capita. The effect of a decline in GDP per capita (the variable GDP deviation) was, on the other hand, as expected and indicated that also for Others, an economic downturn would tend to be followed by rising levels of foregone medical care – but the effect was only around half as large as among the Disadvantaged. The countries’ income inequality level 2007–2012 appeared to have no overall effect, but that income inequality could be unfavourable also for Others was suggested by the significant (but small) effect of the interaction between GDP deviation and income inequality level (Model 3). When going through an economic downturn, it seems therefore that also among Others, the consequences for access to medical care would be more detrimental the larger income differences in the country.

Second, independent of the country’s overall economic level (measured by GDP per capita), larger income inequalities were clearly associated with more reporting of foregone medical care during the 2008—2013 period among the disadvantaged subsamples (which constituted, on average, about 15 % of the country samples). This unfortunate association between income inequality and access occurred only among the disadvantaged, however, and the association was not found for the other parts of the analyzed samples.

Third, foregone medical care tended to deteriorate with economic downturns, but such adverse developments were stronger the larger the country’s income inequality. Although small income inequalities did not protect against an economic crisis as such (i.e. against GDP decline), a more egalitarian income distribution was associated with less negative effects on access to medical care when the country endured an economic crisis. In this sense, smaller income inequalities seemed beneficial for protecting access during crisis times – in particular for disadvantaged people, but also for other parts of the population.

In sum, the findings of this study correspond, at least to a considerable extent, to the “afflictions of inequality” thesis: Independent of the average affluence of a country (as measured by GDP per capita), social dysfunctions and unfortunate societal characteristics tend to be worse the larger income inequality in the country. Access to medical care (as experienced by the population) may be added to the list of such unfortunate societal aspects – but what processes and mechanisms could generate this association?

Status anxiety and psychosocial injuries, often invoked for explaining associations between income inequality and inhabitants’ health and well-being [4], seem less relevant for the topic examined by the present study. More relevant is probably the obvious difference between egalitarian and inegalitarian countries. In the latter, a larger part of the population will typically have incomes far below the average. Therefore, the proportion of the population who are sensitive to changes which worsen access may be larger in inegalitarian countries. Barriers to obtaining medical care will typically increase during an economic crisis, because supply is reduced (e.g. closure of health facilities), utilization becomes more costly (e.g. increased co-payments), or households’ economic situation worsens [19, 21, 24]. The larger the proportion of the population who are likely to be influenced by such changes, the larger negative effect on foregone medical care is likely to occur. One explanation for the findings in this study, for instance that foregone medical care increased more among Disadvantaged than among Others, could therefore be that the proportion in the population “at risk” for social misfortunes are typically larger in inegalitarian countries.

This “mechanical” explanation can be challenged, however, by an interpretation which highlights political factors. If policies aiming at limiting income inequalities have wide support, governing politicians may also be more inclined to prevent both reduced access to medical services and larger social inequalities in access. This could imply that proposals for increasing user fees will be rejected, exemptions from co-payments will be more generous, and health budgets will suffer less cuts – and, in general, that the healthcare system is more oriented towards safeguarding access.

This explanation is in line with interpretations of the associations between income inequality and social ills which emphasize that political power is consequential both for income inequality, the generosity of welfare services [7‐9, 47, 48], and for how social policies influence access to medical care [30]. However, the findings in this study cannot be explained simply by types of welfare regimes [49, 50]. Analyses of the Disadvantaged samples (Table 3, Model 3) with fewer countries – first, only the 25 non-Nordic countries, and second by excluding Spain, Italy, and Greece and analyzing the remaining 27 countries – gave coefficients for the income inequality variables very similar to those reported in Table 3.

The socio-political approach could be interpreted as implying that income inequality is primarily an outcome of more fundamental social and economic forces, without a causal impact “on its own”. This conclusion may be challenged, however. Smaller income inequalities in a country may enhance social solidarity and social cohesion, and these factors may in themselves facilitate policies which protect access to medical care in times of crisis.

A potential problem is the representativity of the surveys [51], but it seems unlikely that the results are grossly misleading because of sample bias. To what extent the questions used by EU-SILC provide appropriate measurements of accessibility to medical care is debated [52]. One problem which has been given little attention is that measuring access by self-reports of foregone medical care may potentially underestimate access difficulties. When afflicted by severe medical conditions, most people will somehow manage to get medical assistance, no matter how much they have to pay or how far they have to travel, because abstention is not a viable option. Thus, they have not abstained from medical care and may decline to report foregone care, regardless of how difficult access actually was.

The results of multilevel analyses may depend on how the models are designed [44]. The somewhat surprising result that among Others, levels of foregone care were higher the higher the GDP level in the country (Table 4), appeared only when the individual-level variables gender, age, and low education, were included as random slope effects. Often, multilevel studies take only random intercepts into account, and if the present study had dropped random slopes for the individual-level variables, GDP level and income inequality would have had significant effects on foregone medical care also in the Others subsamples, although weaker than the effects observed among Disadvantaged.

A limitation in this study may be that only two country-level characteristics – GDP per capita, and income inequality – were analyzed. More country-level indicators might give a better understanding of processes and mechanisms, but the limited number of higher-level units (30 countries) limits the number of country-level indicators which could be analyzed.

This study has in particular examined a subsample termed Disadvantaged, motivated by the argument that their access will be of considerable health policy interest and particularly relevant for judging the performance of the healthcare system. An objection could be that the Disadvantaged are relatively few, making it dubious to use their experiences as a basis for more general assessments. This objection is relevant, but it should be noted that although the Disadvantaged constitute only 10–20 % of the samples, they stand for 41.0 % (in the entire sample analyzed here) of all those who reported foregone medical care.