RESULTS
It is beyond our scope to present a unified treatment of all empirical results. As argued earlier, the estimates of inequality of opportunity are likely a lower bound of the true figure in all cases and the magnitude of the underestimation is inversely related to the richness of the data set.
Consequently, the importance of the empirical results has to be gauged by considering the number of types that can be defined with the data set. Intriguing issues that may arouse the curiosity of the readers can be easily identified. First, what is the extent of EOp with respect to overall inequality? What is the contribution of effort to inequality, is it larger than that of circumstances? Is the indirect contribution of circumstances through its impact on effort sizeable? Does it make much difference to follow Roemer’s viewpoint in measuring effort, or will using absolute measures of effort give similar results? Among circumstances, what are the most significant? Is there a common pattern among inequalities of opportunity with respect to the objectives of health, education and income? Is there a difference of magnitude in inequality of opportunity between the developed countries and the developing countries? Does the ranking of countries differ when we look at inequality of opportunities versus inequality of outcomes? Do taxes and benefits or other instruments make a large difference when measuring EOp? (i.e., inequality of opportunity for pre-fisc versus post-fisc income.)Starting from a very coarse definition of types (three levels for father’s education, five levels for income), Lefranc et al. (2008) found that Sweden and Norway almost achieve EOp for income, while at the other extreme in the range of Western countries are Italy and the United States, with other European countries in the middle. The qualitative results are similar to those of Roemer et al.
(2003). We will take a closer look at the Nordic countries before reporting the results obtained for Italy and the United States. We will then contrast these results with those obtained for Latin America, Africa, and Turkey.Three thorough empirical studies have studied EOp for income in Scandinavia: Aaberge et al. (2011) and Almas et al. (2011) for Norway, and Bjorklund et al. (2012) for Sweden. Starting with the latter, the authors claim that they have a finegrained typology (1152 types), which partitions the sample into types based on parental-income quartile group (four groups), parental education group (three groups), family structure/type (two groups), number of siblings (three groups), IQ quartile (four groups), and body mass index (BMI) quartile at age 18 (four groups). 7 The random sample consists of 35% of Swedish men born between 1955 and 1967 and the outcome is an average of pre-fisc income over 7 years (age group: 32-38). Looking at the graphs of stochastic dominance reveals something that was already present in Lefranc et al. (2008). The income CDFs of the different educational or parental-income types are quite close. The differences are more pronounced for IQ-types. Parametric results reveal that the three most important contributors to inequality of opportunity are parental income, IQ, and the type heterogeneity of the disturbance (which may be due to effort, luck, or unobserved type heterogeneity, because the parental-income and education groups are still large). Looking at the Gini coefficient (the results are a bit sensitive to the measure, as usual), putting IQ aside, the other “social” circumstances account for between 15.3% and 18.7% of the overall Gini. That means that in the counterfactual situation where the only factors of inequality would be these social circumstances, the Gini coefficient would attain a modest value of 0.043 for the oldest cohort. The contribution of IQ represents about 12% of the overall Gini. So far, these results are very impressive and confirm that Sweden is close to reaching a situation of equal opportunity.
Still, it remains to be seen if introducing parental income in a continuous way and perhaps education of both mother and father, thus refining the typology, would alter the results significantly.The results for Norway obtained by Aaberge et al. (2011) are built on a coarser typology (three educational parental levels, to grow up in a large family or not, to be born in a
47 BMI is measured at a young age. It would be far more controversial to put BMI on the circumstance side for older people. Of course, there are genetic roots of obesity among some subjects, but the main determinant is lifestyle (see the discussion in Bricard et al., 2013).
main city or not, and birth cohort). Tranches are defined by relying upon the Roemer identification axiom. The data come from a rich longitudinal set containing records for every Norwegian from 1967 to 2006, enabling one to build up a permanent income measure. The Gini coefficient of permanent income is as low as 0.17, and the authors graph Pen’s parade (the inverses of the permanent income CDFs) for the three educational groups. These inverse CDFs are quite close. The Gini coefficient corresponding to inequality of opportunity is about 0.05 suggesting that opportunity inequality accounts for about 28% of income inequality when the analysis is based on permanent income. Since the typology is coarser than in Bjorklund et al. (2012) for Sweden, the results so far are compatible with a higher inequality of opportunity and likely a higher contribution of inequality of opportunity to overall inequality. Almas et al. (2010) use a different methodology and the results cannot be easily compared. Nevertheless, we can observe an upper bound for the impact of effort. If we consider the usual candidates for effort variables such as years of education, hours of work (for those who work), working in the public sector, county of residence, and choice of university major, then effort’s raw contribution to the Gini in Norway in 1986 is about 25.5% in pretax income when we do not sterilize effort variables of the impact of circumstances.
However, the impact of parental background on effort variables is quite small. It represents one Gini point over a Gini of 0.26. It is generally observed that the unexplained part (by circumstances or effort) remains quite large and even dominant in all empirical studies of inequality of opportunity.Next, we will review results on the “poor achievers” of the EOp class among developed countries, the United States and Italy. Pistolesi (2009) uses panel data, the PSID from 1968 to 2001, and he considers age, race, education of both parents, the region of birth and the occupation of the father as circumstances. The two responsibility variables are the years of education and the hours of work. Their conditional distributions are estimated nonparametrically against the vector of circumstances. Pistolesi then predicts two counterfactual distributions for both educational and working-duration distributions. In the first, the effect of unequal circumstances is removed, whereas each individual is assumed to have exerted the same effort in the second. The circumstances have a weaker impact on hours of work than on education, a finding quite common across empirical studies, and which makes sense. A presentation of the results with the Gini to allow comparisons with previous studies shows that the share of inequality due to circumstances in the DU sense is about 35% for a 5-year average earnings at the mean point of the distribution. It is indisputably higher than in Sweden, but it follows a quite remarkable decreasing trend over the period. If the results were confirmed, it would mean that the increase in inequality that has occurred in the United States is not due to an increase in inequality of opportunity. Checchi and Peragine (2010) study the inequality of opportunity in Italy. There are three circumstances: parents’ education (five types), sex, and regions (North, South). What is striking is that with such a coarse typology, they find that inequality of opportunity accounts for about 20% of overall income inequality in Italy—that is, higher than the 16% in Sweden with a much finer typology.
Perhaps the sharpest indication of inequality of opportunity for income and wealth is a high elasticity of the income (or wealth) of fathers and sons.[168] Corak (2013) provides an excellent review of the facts for highly developed countries.
The Great Gatsby Curve is a strongly positive relationship between the Gini coefficient of income and intergenera- tional income elasticity. For a set of OECD countries, the United States, United Kingdom, and Italy have both the highest Gini of disposable household income (about 0.35) and of intergenerational income elasticity (about 0.5); Norway, Finland, and Denmark have the lowest of both measures (about 0.23 for the Gini, and less than 0.2 for the elasticity). According to Corak (2013), the main determinants of the high elasticities are the behavior of mobility at the top and bottom of the distribution. In the United States, more than half of sons of fathers in the top decile are placed, as adults, in the top three deciles; similarly, about one-half of sons of fathers in the bottom decile are placed, as adults, in the bottom three deciles. In the United States, high-income families pour private resources into their children; Corak (2013) reports that these “enrichment expenditures” (books, computers, summer camps, high-quality day care, and private schooling) total about $8,900 per annum per child for families in the topincome quintile, while families in the bottom quintile spend $1,300 per annum per child (2006 figures). An equal-opportunity policy should compensate low-income children with similar resources, publicly financed. We recall that private schools hardly exist in the Nordic countries, which surely contributes to the lower intergenerational income elasticities there.Next, we will turn to less-developed countries. The Latin American study by Ferreira and Gignoux (2011) provides results that can be compared with previous studies. Circumstances are defined as ethnicity, father’s and mother’s occupations, and birth region, for Brazil, Ecuador, Guatemala, Panama, Colombia, and Peru. The number of types is more than one hundred for the first four countries and about 50 for the last two countries. The contribution of circumstances to inequality is quite high and it varies quite a lot across the six countries.
If we look at income, Guatemala and Brazil have in common a high value of the share explained by observed circumstances, about one-third, followed by Panama (30%) and Ecuador (26%). The contribution of inequality of opportunity to total inequality is about 28% in Peru and only 23% in Colombia. However, these two countries have fewer types, which biases the estimates downward with respect to the other countries. The authors also provide estimates of the contribution of nonresponsibility characteristics to consumption inequality per capita, which may be more similar to permanent income. The degree to which inequality of opportunity explains inequality is even higher for some countries, over 50% for Guatemala. Ferreira et al. (2011) study the case of Turkey, which has roughly the same level of development as Brazil, and find that on a sample of ever-married women aged 30-49, inequality of opportunity accounts for at least 26% of overall inequality in imputed consumption, which is by and large a lower value that those found for Latin American countries, except for Colombia. For African countries we will refer to the study of Cogneau and Mesple-Somps (2008). The surveys that are selected are the only large sample nationally representative surveys in Africa that provide information on parental background for adult respondents. They cover two countries under Britain’s former colonial rule, Ghana and Uganda, and three countries under France’s former colonial rule, Ivory Coast, Guinea, and Madagascar. The types are defined by a small number of occupational, educational and geographical circumstances. For the two most developed countries, Ivory Coast and Ghana, the Gini inequality of opportunity index is about 0.15 (the triple of what is found in Sweden) and it represents about one-third of overall inequality (0.45). The information is poorer for other countries but, given the results one has on a comparative basis, one can guess that the share of inequality of opportunity is even higher there.48
All in all, it seems that the inequality of opportunity for income is highly correlated with inequality of income. This observation is confirmed by the high correlation (0.67) between these two kinds of inequality, measured by the Gini coefficient for western countries (Lefranc et al., 2008). Moreover, this strong correlation seems a general pattern that does not depend on the outcome chosen. Indeed, working on the Retrospective Survey of SHARELIFE, which focuses on life histories of Europeans aged 50 and over, Bricard et al. (2013) observe a positive correlation of about 0.39 between inequality of opportunity in health and health inequality. Furthermore, since lifestyles are documented in this data set, the authors are able to show that inequalities of opportunity for health status in Europe represent on average half of the health inequalities due to both circumstances and effort (lifestyles). There are, however, large variations across countries. The health indicator in this study is SAH (self-assessed health) but using mortality indicators as in Garcia-Gomez et al. (2012), the importance oflifestyles also comes out as a distinctive feature. These authors use a rich data set for the Netherlands (1998-2007), linking information about mortality, health events, and lifestyles. They estimate a full structural model that reveals strong educational gradients in healthy lifestyles which, in turn, have the expected effect on mortality.
From a dynamic viewpoint, intergenerational mobility is clearly an important measure of EOp. Almost all studies of intergenerational mobility define the classes between which mobility is measured as income classes; see Chapter 9 for a thorough discussion of the literature. We mention only one study here. Lefranc et al. (2007) show that under a loglinear relationship between parent and child earnings, whose slope β is the intergenerational earnings elasticity, and choosing the MLD as inequality index, then the following relation holds:
lf = ~αt + βtpt;
The MLD among descendants, can be written as an affine function of the mean MLD among the fathers’ incomes at date t, Ip, which is a circumstance for children. The constant — αt can be interpreted as residual inequality were there to be no inequality of parental income. We may interpret βtIρ as the inequality of opportunity due to the circumstance of parental earnings. Reduction of inequality of opportunity can derive from either a drop in the intergenerational transmission of advantages (β), or from mitigating income inequality in the parental generation. In the case of France, the authors found that the reduction of inequality of opportunity was a consequence only of a decrease of inequality in fathers’ incomes without any clear contribution of the intergenerational link.
We are at the very beginning of solid empirical analyses of inequality of opportunity. Analysis has been hampered so far by limitation of data sets and the intricacy of the issue. For each recent paper beginning with Bourguignon et al. (2007), the same ritual sentence appears in the introduction, to the effect that “this set of circumstance and effort variables is richer than those used so far in the existing empirical literature on inequality of opportunity.” If this trend continues, we can be optimistic that, in the coming years, data sets will improve, as the stakes become clearer.
4.12.