The theory of equality of opportunity

In the previous section we discussed equality of opportunity. Given this definition, there is a notion to make the individual responsible for the consequences of his/her decisions because we assume that these decisions are free and well informed. Ultimately, the most interested in getting the greatest welfare is the individual himself/herself.

The main objective of society is to give freedom of choice to the individual as long as equality plays an instrumental role, where everybody should have the same freedom to take decisions. Because inequalities may come from circumstances and individual effort, conjectures about justice arise. Inequalities caused by circumstances can be considered illegitimate and researchers can place special emphasis on these, instead of focusing on inequalities in general. With this new normative judgment, we are interested in measuring unjust inequalities rather than total inequalities. According to ^{Fleurbaey and Schokkaert (2011}), there are two principles that an unjust health inequality measure must meet:

(i) To give equal benefits to individuals equally responsible (Compensation).

(ii) To disregard legitimate inequalities (Reward).

The first principle is related to the compensation principle and the second one relates to the reward principle (^{Fleurbaey, 2008}). The compensation principle disregards individual disadvantages arising from circumstances, while the reward principle allows individuals benefits associated with higher effort (^{Arnsperger, 1994}; ^{Kolm, 1995}; ^{Varian, 1974}).

Nevertheless, these two principles are mutually exclusive. For example, if unjust inequality measures satisfy the reward principle, it does that because it does not consider individual effort. But the compensation principle considers individual effort. Thus, if the inequality measure satisfies the reward principle, it cannot fulfill the compensation principle (or vice versa). Legitimate inequalities are related to individual effort. Then, the research needs to choose one of these two principles as the basis principle of unjust inequality measures. ^{Fleurbaey and Peragine (2013}) show that, beyond having an excluding restriction between the compensation principle and the reward principle, there is an excluding restriction between an ex-ante and an ex-post unjust inequality measure.

The equality of opportunities approach has been applied in different fields of economics. For example, see Roemer et al. (2003), ^{Brunello and Checchi (2007}), ^{Ferreira and Gignoux (2011}), ^{Lefranc, Pistolesi and Trannoy (2008}, 2009), Pistolesi (2009) and ^{Schütz, Ursprung and Wößmann (2008}). The theory of equality of opportunity has been applied also to health issues. For example, see ^{Fleurbaey and Schokkaert (2009}), ^{Jusot, Tubeuf and Trannoy (2010}, 2013), ^{Rosa Dias (2009}), Rosa Dias and Jones (2007), Trannoy et al. (2010) and Jusot, Mage and Menendez (2014).

^{Jusot et al. (2010}) find Gini coefficients for inequalities of opportunities in health ranging from 0.09 to 0.16. Sweden reports the smallest inequality, while Spain reports the greatest inequality. Europe as a whole reports a Gini coefficient of 0.10. ^{Trannoy et al. (2010}) confirm the presence of inequality of opportunities in health in France, with a Gini coefficient of 0.155. In addition, they find that parental socioeconomic status is the most important circumstance to explain health inequalities on descendants’ health through adulthood. Also, ^{Rosa Dias (2009}) finds inequality of opportunities in health in the United Kingdom of about 21.5%-26.3% of total health inequality using a pseudo-Gini coefficient as the inequality measure. There is little literature about this topic in developing economies. An exemption is a study about Indonesia. Jusot et al. (2014) show that parental health is the major transmission mechanism of inequality of opportunities in health. Circumstances account for a share of 10% in total health inequality.

It should be stressed that there is no clarity about which characteristics must be considered as circumstances. However, when it comes to health, socioeconomic characteristics through childhood are the first option because they accomplish the definition that theory gives to circumstances. An individual does not control or does not decide which socioeconomic characteristics should have when he/she is born. There is evidence that children from vulnerable households report the worst health status and this has a lasting impact throughout life. Likewise, children with the worst health status have lower educational attainment and hence lower social mobility through adulthood (^{Case, Lubotsky & Paxson, 2002}; Case, Fertig & Paxson, 2005; ^{Strauss & Thomas, 2008}).

^{Felitti et al. (1998}) find an association between childhood circumstances and the adoption of risk factors that affect health in adulthood (e.g., alcoholism and substance abuse). The childhood circumstances considered are those situations that reflect any household dysfunctionality or adverse situations (physical and sexual abuse or related events). Felitti et al. (1998) present evidence that individuals that have a difficult childhood have the worst health status and are more likely to get sick. For example, the probability to be a smoker is twice for individuals suffering an adverse circumstance. Something similar happens with the prevalence of heart disease, cancer, stroke, chronic bronchitis or emphysema, diabetes, bone fractures, hepatitis and self-reported health. However, Felitti et al. (1998) emphasize the need to substantiate more complex relations that could highlight a direct causality between childhood circumstances and health in adulthood beyond the instrumental role that risk factors have.

Health inequalities in Colombia

^{Bernal and Cárdenas (2005}) study the existence of health inequalities attributable to race. They find that individuals belonging to ethnic minorities are less likely to have medical insurance, have worse perceptions about their health status and are more likely to get sick, although they use less hospital services. The authors conclude that there are not significant differences in health outcomes between ethnic groups. Access to health services and self-reported health do not have a statistically significant relation with race when they control for other socioeconomic characteristics such as labor force status, educational attainment, geographic location, and the like. This implies, according to the authors, that worse health perceptions by ethnic minorities are related to relative disadvantage over other socioeconomic characteristics.

Acosta (2014) analyses regional inequalities in self-reported health. She uses the methodology of ^{Allison and Foster (2004}) to get an approximation to health inequality indexes based on qualitative data. As could be expected, the most prosperous regions report better health status. The conclusion from Acosta (2014) is consistent with estimations of life expectancy at birth reported by the National Department of Statistics (Departamento Administrativo Nacional de Estadística, in Spanish) based on the census conducted in 2005. According to Naga-Yalcin indexes, regions with the highest health inequalities in 2010 are the East Central region and the West Central region. Between 2005 and 2010, health inequalities have increased. Women report higher use of medical services, according to the author, consistent with the fact that women report the worst health status.

^{Camacho and Flórez (2012}) analyze equity in maternal and child health services conditioned by geographic area and wealth quintiles. The implementation of the subsidized regime encourages the use of health services, thus contributing to reduce health inequalities by region and socioeconomic characteristics. However, quality has not improved given its focalization and financing. Indicators of maternal and child health exhibit improvements with increases in coverage of maternal and child health services. In addition, they observe a drop in child morbidity without implications for inequality. The authors suggest the possibility that other social characteristics determine these results.

^{Fresneda (2010}) performs a retrospective analysis on the dynamics of health inequalities in Colombia between 2000 and 2007. He relies on self-reported health status, perceived disability and risk of communicable diseases, external injuries, morbidity and demand for health services. He finds that self-reported health is better in urban areas relative to the reports of rural areas. Likewise, men report better health compared to what women report. Regarding the health regime in which an individual is affiliated, those who are affiliated to a special regime and the contributive regime report better health status compared to those who are affiliated to the subsidized regime. Something similar happens when individuals live in higher socio-economic strata and have more years of education.

Finally, ^{Flórez and Soto (2007}) identify inequalities at the geographical and gender levels and in the use of services in 1990, 1995, 2000 and 2005 using information available from the Demographic and Health Survey. Regarding insurance, there were significant gains in universal enrollment, particularly between 1995 and 2000. The authors associate this fact with the inclusion of relatives in the contributive regime. Also, the largest affiliations occur across richer regions. Between 2000 and 2005, inequalities in the use of medical services increase by region, gender and wealth quintile. Nevertheless, prenatal care inequalities are reduced between 1990 and 2005 for all socioeconomic characteristics. As for child mortality, inequalities are reduced between 1995 and 2005; however, individuals in rural areas and poor regions and boys still report higher mortality rates. Inequalities in chronic undernourishment increase, except when the authors control for gender.

Considering all the previous results, insurance coverage and health service availability have increased remarkably since enactment of the health bill currently in place (Ley 100 de 1993). These improvements were focused on the vulnerable population through the subsidized regime. Even so, quality of services depends on regime affiliation, where better regimes (i.e., contributive and especial) report better health indicators. In addition, it can be observed a higher use of medical services associated with the increased availability of services, but inequalities in the use of services by wealth quintiles or region (rural vs. urban) are still evident. Richer regions report better distributions of health states, always favoring the central regions. There is no evidence of health inequalities between ethnic groups, but inequalities conditioned to other socioeconomic characteristics (education, occupation, wealth quintile, region, and the like) are often found.

All the previously cited literature has been focused on total health inequalities. Nevertheless, ^{Fajardo-Gonzalez (2016}) conducts an analysis of equality of health opportunities from a normative perspective related to the analysis attempted herein. Using data from the 2010 Living Standards and Social Mobility Survey, she calculates dissimilarity and Gini-opportunity indexes to provide different measures of inequality of opportunity in adult health. Additionally, she applies Shapley-value decompositions to estimate the contribution of circumstances on the dissimilarity index. Her findings suggest that about 10 percent of circumstance-driven opportunities enjoyed by those who are healthier should be redistributed among those less healthy and thereby attain equality of opportunities. The most important circumstances of inequality of opportunity in adult health are household socioeconomic status during childhood and parental education.

Often, health inequalities in Colombia have been analyzed from an institutional perspective, i.e., studies place emphasis on access to health services and how much individuals use them. Nevertheless, health cannot be measured only as use of health services and access to them. This paper uses data from the Colombian Longitudinal Survey under the framework of the equality of opportunity theory and, rather than emphasizing on total health inequality (of self-reported health), it focuses on unjust health inequalities under a normative framework of social justice. It could be said that this is the first analysis of this kind in Colombia, and therefore this paper contributes to the literature on unjust health inequalities.

Estimations

According to the theory of equality of opportunity, inequality of final outcomes is attributable to both circumstances and effort. Because the dataset has more information related to childhood circumstances, our measure of inequality of opportunities in health conforms to this fact. For this reason, the measure as presented here is an ex ante measure of health inequality (see section I).

The theoretical basis for measures of unjust inequalities is Roemer (1998). According to this author, the general case is that in which the final outcome is affected by circumstances and effort; and equality of opportunities requires that individuals be compensated for any difference in circumstances that could affect the final outcome, but not for any difference in effort. For Roemer (1998), a “type” is a group of individuals with similar circumstances. Clearly, individuals of the same “type” could have different final outcomes. For this reason, each “type” has a distribution of individuals based on their final outcomes. Roemer (1998) assumes a monotonic relation between final outcomes and individual effort. This means that within a “type” an individual makes more effort if he is in the top percentile relative to other individual of the same “type”. Then, for Roemer (1998), the effort to be considered is the effort that does not depend on circumstances. Two individuals of different “types” make the same effort if both are on the same percentile of their respective “type”.

Having said that, there is a discussion regarding the correlation between circumstances and effort. ^{Barry (2005}) argues about the need to ignore the conditionality imposed by individual circumstances on future effort, that is, individual effort (taking care of his/her health) must be considered independently of the influence of past circumstances. ^{Jusot et al. (2013}) show the empirical implications of different normative conjectures. They rely on three premises and measure the inequality of opportunities that different assumptions involve. At the end, their results suggest that the relative participation of circumstances and effort on health inequalities are very similar regarding their three premises. See also ^{Balia and Jones (2011}).

According to the above, we specify two linear models regarding inequality of opportunities in health:³ one considering Roemer’s normative principle and the other considering Barry’s one. We get

() 2

In this model, we assume a linear relation between self-reported health S (UVAS, AVAS or EQ5D) and socioeconomic characteristics(demographics (Demog), circumstances (Circums), effort (E f f o) and economic status (Stat)(at an individual level, hence the subscript i. Equation (2) identifies Barry’s normative principle, in which only direct effects of circumstances on health are considered. Now, given the conditionality of circumstances on effort, we get

() 3

. ()4

Replacing from Equation (2), we get

()5

()6

()7

where

()8

()9

()10

(8) ()11

Equation (7) identifies Roemer’s normative principle. Equations (2) and (7) are estimated in the next section. ^{Ramos and van de Gaer (2015}) carry out a meta-analysis about the different ways of measuring inequality of opportunities. They find that the literature distinguishes between direct and indirect measures, ex ante and ex post measures and parametric and nonparametric measures. Using their terminology, the measure that we present below is a parametric, direct and ex ante measure of inequality of opportunities in health in Colombia for 2010.

Measuring inequality of opportunities in health

There is no consensus on the best way to measure inequality of opportunities. According to ^{Ramos and van de Gaer (2015}), on many occasions measures of inequality of opportunities do not have any theoretical basis and are ad hoc choices by researchers. Under these conditions, we apply the methodology proposed by ^{Shorrocks (1982}) to establish the contribution that every income source (earnings, capital income, transfers, etc.) has on total income inequality. Shorrocks (1982) poses the principles that must be met by any decomposable index, resulting in the following decomposition rule:

()12

where ^_Sj is the share of total inequality associated to ^_Xj source. This decomposition rule is well behaved (^{Shorrocks, 1983}). However, in this paper, rather than having different components that generate a final health status, there is a data generating process (an OLS estimation) that aims to approximate the performance of individual health status. Therefore, we need to adapt Shorrocks's (1982) methodology to an analysis of inequality based on regressions. This is what ^{Fields (2003}) precisely does, showing that the new decomposition rule is

()13

where ^_aj is the estimated parameter in the regression of Y on ^_Xj and ^_Sj is the share of total inequality associated with factor ^{_{j (Xj)}} . This decomposition rule is characterized by

()14

that is, all shares from the linear regression model (including participations from intercept and residuals) must sum 100%. When we exclude the share of residuals, we get

()15

The participations from the explanatory variables are equivalent to the goodness of fit of the model. Then, we can establish that the percentage of total inequality explained by factor j is

()16

The methodology adopted by ^{Fields (2003}) allows us to compute the participation of unjust inequalities on total health inequalities.

Estimations

We present the OLS estimations of Equation (2) through Equation (7). Given that we have three dependent variables (UVAS, AVAS and EQ5D) and two normative principles, we estimate six regression models. Nevertheless, we present the estimates concerning UVAS only (see Table 2). Results for AVAS and EQ5D are similar, both qualitatively and quantitatively (see Table 2A and Table 3A, respectively, in the annexes). In Table 2, Model 1 identifies Roemer’s normative principle (Equation (7)), while Model 2 identifies Barry’s one (Equation (2)). Women report worse health states in comparison with men. This relation is statistically significant. Having been born in urban areas is associated with positive changes in self-reported health relative to rural areas. Regarding household composition, it is ideal to have lived with both parents. People that lived with one parent only, lived with individuals different from their parents or lived alone during childhood have the worst self-reported health status. This relation is statistically significant. As for parental health, having parents alive and without any chronic illness at the time of the survey is important for descendant’s health, a relation that is positive and statistically significant. Finally, that parents have no education shows a negative relation with a descendant’s self-reported health.

Table 2 OLS estimations of unjust health inequalities for UVAS 

Note: standard error in parenthesis; ***p<0.01. **p<0.05. *p<0.1.

Source: own calculations based on ELCA 2010.

Three situations arise that should be noted. First, the coefficient of mother’s health is higher than the coefficient of father’s health (and, as will be seen later, this is true independently of the health measure). Second, coefficients estimated for Model 1 are higher than coefficients of Model 2. This is so because Roemer’s principle (Model 1) accounts for total effects (direct and indirect) of circumstances on health, while Barry’s principle (Model 2) takes account of direct effects only. This enables us to confirm, in addition, an association between individual circumstances and present socioeconomic status. Finally, the use of health services for preventive reasons shows a negative relation with self-reported health. A possible explanation for this is related to the endogeneity of variables; that is, individuals that report the worst health use health services more often.

Estimations for Model 1 could be biased due to endogeneity issues. Given that we do not control for effort and because of the correlation between circumstances and effort, we could be violating the assumption of exogeneity for OLS estimation (i.e, E(u|X) ≠ 0). In this case, the coefficients of circumstances are biased because they capture both direct and indirect effects (absorbed from effort). However, we are fulfilling Roemer’s principle. Thus, we do not correct for endogeneity in Model 1 for any of the health measures.

Inequality of health opportunities in Colombia

Table 3 shows the participations associated with Equation (16) for UVAS, AVAS and EQ5D applying the two normative principles. Inequality of opportunities in health represents almost 30% of total inequality in self-reported health if we consider Barry’s principle. Roemer’s normative principle exhibits a higher degree of inequality of opportunities against Barry’s principle because the former takes into consideration both direct and indirect effects of circumstances on the final outcome (i.e., self-reported health status), while the latter principle only takes direct effects into account.

Table 3 Inequality of opportunities in health in Colombia 

Source: own calculations based on ELCA 2010.

Contextualizing results from the first and second columns of Table 3 on an inequality measure, the third and fourth columns of the same table reports Gini coefficients associated with inequality of opportunities in health for all three self-reported health measures. According to these Gini measures, total self-reported health inequalities are around 0.1214, 0.0521, and 0.0799 for UVAS, AVAS and EQ5D, respectively. Considering the proportions reported, those Gini coefficients of inequality of opportunities in health are 0.0538, 0.0244 and 0.0344 for Roemer’s principle and 0.0268, 0.0153 and 0.0178 for Barry’s principle.

These reported participations in Table 3 are conditioned on observable information related to individual circumstances, therefore the values that are shown should be considered as a lower bound for the true inequality of opportunities in health. Table 4 shows the participations of observable individual circumstances in unjust health inequalities for the three health measures.

Table 4 Circumstances and unjust health inequalities in Colombia 

Source: own calculations based on ELCA 2010.

For the three health measures, the most important circumstance for total inequality of opportunities in health is mother’s health. Birthplace and gender have considerable participations for UVAS only. Their participations in inequality of opportunities in health range between 15% and 18%. Also, household structure during childhood has a significant participation (after mother’s health) for AVAS and EQ5D. The normative approaches of Roemer and Barry have some implications for measuring inequality of opportunities when health status is measured by AVAS. For Roemer’s principle, the third most important circumstance affecting unjust inequalities is mother’s education (14.28%); in contrast, the third most relevant circumstance applying Barry’s principle is father’s health. There is a particular result with respect to UVAS using Barry’s principle. The participation of father’s education is negative, meaning that inequality of opportunities in health is reduced for higher levels of father’s education by a factor of -6.63%.

Based on a geographical approach, inequality of opportunities in health is higher in rural areas than in urban areas.⁴ Table 5 shows the participations of circumstances on total health inequalities in rural and urban areas.

Table 5 Circumstances and unjust health inequalities in Colombia: rural and urban areas 

Source: own calculations based on ELCA 2010.

Table 9 shows the participations of observable circumstances in unjust health inequalities in rural and urban areas. Unlike our findings at the national level, household structure during childhood does not have a relevant share in unjust health inequalities. For all the health measures considered, gender and mother’s health are circumstances that account for a significant proportion of inequality of opportunities in health. For instance, according to Roemer’s principle, gender has a share of 20.52% and mother’s health has a participation of about 27.51% in EQ5D. The relative significance of father’s education and race depends on the normative principle that is used. For Roemer’s principle, father’s education is the third circumstance with the highest participation, whereas race is the third most influential circumstance for Barry’s principle. This is the case for UVAS and AVAS. For EQ5D, the third most important circumstance is race regardless of the normative principle considered.

Table 6 Circumstances and unjust health inequalities in Colombia 

Source: own calculations based on ELCA 2010.

As mentioned above, mother’s health represents the highest share of unjust inequalities independently of the health measure. Gender has the second highest proportion. The trade-off between circumstances conditional on a normative principle is between father’s health and race for health measures UVAS and AVAS in rural areas. For EQ5D, the third circumstance with the highest participation in unjust inequalities is race.

Observed patterns in urban areas are very similar to what is evidenced at the national level (Table 3). In all cases, mother’s health has the highest participation in inequality of opportunities in health. This circumstance is followed by gender and household structure during childhood. This is conditional on the health measure. For AVAS, normative principles condition the significance of the shares of mother’s education and father’s health. Finally, in both rural and urban areas, inequality of opportunities in health applying Roemer’s principle is higher than with Barry’s principle. Moreover, mother’s health is the circumstance with the highest participation in unjust health inequalities in rural areas, as opposed to urban areas.

In general, mother’s health is the circumstance with the highest participation in inequality of opportunities in health regardless of the health measure, geographical area and normative principle. The opposite occurs with the rest of the circumstances. Their participations depend on the geographical area and the health measure. While household structure is the second most important circumstance in urban areas, in rural areas gender is the second most influential circumstance. Observable circumstances explain between 30% and 40% of total self-reported inequalities in health status conditional on the normative principle considered.