Health inequality in the Russian Federation: An examination of the changes in concentration and achievement indices from 1994 to 2013

The collapse of the Soviet Union resulted in large fluctuations in household resources and increased uncertainty for the Russian people. Much academic research has explored the socioeconomic and political ramifications of this transition to a mainstream market economy. The importance of health in times of rapid change in the economy has been highlighted in economic development literature [1, 2]. Preston [3] concludes that factors other than individual income play a powerful role in the determinants of health conditions that are often associated with poverty. This reinforces the idea that individual incomes in isolation from the community in which people are located do not immediately reflect health status. Neighbourhood affluence is a more powerful predictor of health status than poverty, above and beyond individual demographic background, socioeconomic status, and health behaviours [4]. An empirical study using Spanish Living Conditions Survey datasets (2005–2008) revealed that the relationship between health and income operates through comparison information with respect to societal peers; conversely, material deprivation in terms of financial difficulties, basic necessities, and housing conditions exerts a direct effect on one’s self-perceived health status [5].

Vella [6] has reported the prevalence of a considerably high share of dissatisfaction with self-perceived health status among the respondents of the Russian Longitudinal Monitoring Survey, 1993, and Cockerham [7] has found no relationship between income and self-perceived health status among the respondents of the Russian Longitudinal Monitoring Survey, 1995. Another cross-sectional study by Cockerham et al. [8] reported a similar relationship between income and self-perceived health status among the respondents of the Russian Longitudinal Monitoring Survey, 1998. Further, the Russian Longitudinal Monitoring Survey, 1998 reflected a strong negative correlation between the health status of the individual and the desire to return to the former Soviet regime – a reflection of dissatisfaction with the economic and social conditions of the time [8]. A panel data frame study using Russian Longitudinal Monitoring Survey (1994–2000) datasets showed little effect of income on self-perceived health status when the possibilities of reverse causation and incidental association between income and health were accounted for (Stepanyan A. Is Economic Transition a Health Hazard? Self-perceived Health and Income in Russia: 1994–2000. Unpublished working paper 2002). However, a longitudinal study (Jensen RT. Job Security, Stress and Health: Evidence From Russian Privatization. Unpublished working paper 2005) using Russian Longitudinal Monitoring Survey datasets (1992–1994) suggested a correlation between negative health consequences with concern about the likelihood of job loss.

Kraus et al. [9] have shown that chronically experienced negative affects (higher temperamental sensitivity to negative stimuli resulting in feelings of fear, anxiety, depression, guilt, and self-dissatisfaction) explains rather than confounds the link between socioeconomic strata and self-perceived health. The systematic review study of Ostrove and Adler [10] suggests that the pathway from socioeconomic strata to health is through exposure to different environments and adaptations to these environments. Environmental demands and supports shape individual responses to self-perceived health. Unemployment and inflation have often been considered two of the most robust determinants of individual happiness, with unemployment typically having a stronger negative impact than inflation. Eggers et al. [11] found that in the Russian Federation, higher unemployment rates lead to more happiness, as people tend to revise their expectations upwards when their neighbours are suffering. Rodríguez-Pose and Maslauskaite [12] argued that citizens are quite tolerant of interpersonal inequality, and as the economy is still dynamic, open and volatile, the influence of macroeconomic factors on happiness varies substantially over time; therefore cross-sectional analysis cannot provide any robust conclusions.

An empirical study (using eight waves of the European Community Household Panel) reflects a larger effect of activity status on self-perceived health than the effect of income. Inactivity is associated with the largest absolute effect, in part, among individuals claiming disability allowances due to ill health [13]. A cross-sectional study [14] on the evolution of health inequalities in Switzerland found no systematic trend in the changes with the measure of health inequalities between 1982 and 2002.

The ecosocial theory of health distribution, introduced by Krieger in 1994 [15] postulates that (1) population health and health inequities must be analysed in a societal, historical, and ecological context, and (2) neither the forms of social inequality nor their associations with health status are “fixed”, but are instead historically contingent. The relationship between socioeconomic strata (SES) and health changes with age, since the mediators of SES act differently in different stages of the life course. Health in later life is the result of multiple social and biological processes whose effects interact and accumulate over time [16‐18].

The microeconomic perspective assesses health costs (or the economic benefits of good health) at the individual or household level. Macroeconomic consequences are viewed from the level of the national economy, generally considering the level of population health that retards a country’s economic growth. The realization of the economic objectives with a concurrent reduction of poverty (more relevant for transitional economies) necessitates evidence-informed policy development for investment in health. The micro perspective is also specifically important for individuals and households, as freedom from ill health adds different dimensions of economic well-being to individuals and households. Further, the concern for poverty and inequality recognizes that in health status societal averages typically disguise as much as they reveal, therefore, the interest is not in health status that prevail in society as a whole but in health status of different socioeconomic groups [19].

This paper investigates how the difference in the self-perceived health status of Russians has evolved in the context of the transition to a mainstream market economy, and following the approach of Wagstaff, Makinen et al. and Castro-Leal et al. [20‐22], this study examines the trend in the distribution of such difference across the socioeconomic strata of the population from 1994 to 2013. Our data is unique in the sense that we follow the same individuals over an 18-year period.

The data for this study are from the 1994–2013 waves of the Russian Longitudinal Monitoring Survey (RLMS). RLMS (http://​www.​cpc.​unc.​edu/​rlms) is an on-going longitudinal household survey of the Russian Federation. The survey was designed to be representative of the Russian population in the early 1990s. The survey contains an array of information on the economic, social, demographic, and health characteristics of respondents, their households, and the environments in which they live. An initial representative sample of dwellings was drawn in 1992 and surveyed every six months until 1994 (phase 1). A second more broadly drawn representative sample (phase 2) replaced the initial sample in 1994 and is being followed annually since then. For cost reasons, RLMS does not attempt to follow individuals or households who move from their original sample dwelling units (attempts are made to follow individuals and households that move locally); instead, any new household member or household living at that dwelling is included in the sample in each wave. Thus, the RLMS sample remains representative of the underlying population if new residents can be considered exchangeable with those that moved.¹

The RLMS data sets contain post-stratification weights, that is, weights that adjust not only for design factors but also for deviations from census characteristics. Variation in individual weights reflects the geographic effects for households as well as differentials due to post-stratification of the sample by major geographic region, age, and gender. The multivariate distribution of the sample by gender, age, and urban/rural location compared quite well with the corresponding multivariate distribution of the 1989 census (http://​www.​cpc.​unc.​edu/​rlms).

From this unbalanced sample, we chose RLMS participants who had been included in all 18 waves. Thus, after balancing the panel, we followed 1,496 individuals (Table 1), henceforth to be read as study sample. We explored the study sample selection (i.e., difference in characteristics of respondents between the extracted individual and RLMS – 1994 datasets) with the binary (study sample = 1, RLMS – 1994 datasets = 0) using logit model. Hence, with the long (19 year) follow-up of the individuals, we have controlled the variations attributed to the sampling population per se. The use of more than one observation facilitates causal inference in situations where inferring causality becomes very hard if only a single cross section are used and the panel nature of the data allows for the unobserved effect to be correlated with the explanatory variables – thus, ensures a better control for the unobserved effects of the variables [23].

Our dependent variable for the analysis is self-perceived (self-perceived) health (SAH). Individuals were asked, “How would you evaluate your health?” and the response was recorded on a five-point Likert scale with the answers “very good”, “good”, “average – not good but not bad”, “bad”, and “very bad”. SAH variables have been widely used in literatures [24‐27] that analyse the socioeconomic health gradient.

We constructed an SES index by applying principal component analysis (PCA) using a set of variables (adult equivalent household income, ownership of assets, living conditions, and other indicators of material affluence, i.e., having savings and growing crops) with a Kaiser-Meyer–Olkin score of 0.80 and above. With the RLMS – 1994, PCA resulted in two principal components (PC), first PC accounting for 29 % of the total variance captured in seven items (adult equivalent household income, country house, own house, additional apartment, central heating, hot water supply and use of cooking gas), while the second PC accounted for 15 %, captured in four items (adult equivalent household income, country house, own house and additional apartment). With the data sub-set (study sample), the first PC accounted for 33 % of the total variance captured in seven items (adult equivalent household income, country house, own house, additional apartment, central heating, hot water supply and use of cooking gas), while the second PC accounted for 15 %, captured in four items (adult equivalent household income, country house, additional apartment, central heating). The income variable represented the sum of incomes from all sources for the household deflated to the value of June 1992. We calibrated the household income as per adult equivalent using the modified OECD scale² for our analyses. We measured inequality in income by the Gini index.³

We standardized [28, 29] the self-perceived health status by age, gender, and diagnosed chronic diseases, applying the indirect method of standardization.⁴ We estimated the correlation of the confounding variables (age, gender, and diagnosed chronic diseases) with self-perceived health status conditional on non-confounding variables (region, settlement of residence, and SES). This regression-based approach (Eqn. 1) “corrects” the actual distribution of the self-perceived health status by comparing it with the distribution that would be observed if all individuals in the group had their own age and gender characteristics but the same mean age and gender effect as the entire population.where, y _i is self-perceived health status; i denotes the individual; and α, β, and γ are parameter vectors. The x _j are confounding variables (age, gender, and diagnosed chronic diseases), which we want to standardize, and the z _k are non-confounding variables (region, settlement of residence, and SES), which we do not want to standardize but to control for in order to estimate partial correlations with the confounding variables. The Newey–West⁵ estimator estimates (\( \widehat{\alpha},{\widehat{\beta}}_j,\ {\widehat{\gamma}}_k \)) the individual values of the confounding variables (x _ji ), and sample means of the non-confounding variables (\( {\overline{z}}_k \)) are then used to obtain the predicted, or “x-expected,” values of the self-perceived health status ŷ _i ^x .

Estimates of indirectly standardized self-perceived health:where,

In the next step, following the principles of previous analyses [30, 31], we dichotomized the five-scaled measure into a binary variable [32], “self-perceived health” (1 = good, i.e., responded as “very good”, “good”, and 0 = not good, i.e., responded as “average”, “bad” and “very bad”).

The conventional regression-based statistical methods report the magnitude and the direction of association between socioeconomic position and the health status of the individual but ignores the possibility of variance in the effect of explanatory variables across distribution. Further, such traditional methods cannot reflect the extent of health disparity across socioeconomic strata of the population and thus do not allow for comparison over time [30]. Therefore, we used the health concentration index (Appendix: A) as our measure of SES-related inequality. The concentration index (CI) becomes positive if health (i.e., self-perceived health status) is concentrated amongst the better-off, negative if health (i.e., self-perceived health status) is concentrated amongst the worse-off, and zero if no inequality is observed. Thus, CI can also be interpreted as the slope of a line passing through the heads of an army of people, ranked by their SES, with the height for each individual proportionate to the value of his/her self-perceived health status, expressed as a fraction of the mean for the group.

Finally, with the achievement index⁶ (Appendix: B), we measured the average level of health and the inequality in health between the worse-off and the better-off. Achievement is the weighted average of the health (self-perceived) level of the individuals (members) in the study (community) where higher weights are attached to worse-off than to better-off (The mean is clearly not appropriate, since it weights everyone’s health, here, self-perceived health status equally, irrespective of how poor they are). Thus, achievement index is distributionally sensitive measure of population health. CI captures the extent to which ill health is concentrated amongst the worse-off. Achievement index indicates the degree of aversion to inequalities in health between the worse-off and the better-off [33].

This study uses secondary data collected from perpetual surveys. The datasets are anonymously coded with no individual identification identifiable by the user. The user has explicit authorization to use the datasets made available for analysis.