GEOM

GEOM

Menu

Opportunities aren't equally distributed. Some people have access to more and better opportunities than others.

What is Inequality of Opportunity?

Inequality of opportunity (IOp) is the share of overall inequality that can be predicted by pre-determined, typically inherited circumstances, such as family background, race / ethnicity, caste, biological sex, and place of birth.

Based on these personal characteristics, different statistical methods can divide the population of a country into subgroups that share common circumstances, and compare their income distributions. While an ‘ex-ante’ approach compares averages across these groups, an ‘ex-post’ alternative accounts for differences at every point in the distribution. Headline measures of inequality of opportunity summarize these differences.

What is Equality of Opportunity?

Equality of opportunity means that people have the same chances in life, regardless of any inherited circumstances. So, it means reducing to zero the extent to which inherited factors (like sex, birth area, parental occupation, and parental education) can explain current income or consumption inequality.

This, in turn, implies eliminating gaps between the incomes of population subgroups with different circumstances, ideally by raising the incomes of the population subgroups with least access to opportunities (the “worst off”).

What is Intergenerational Mobility?

Equalizing opportunities requires increasing social mobility across generations, and measures of intergenerational mobility (IGM) are closely related to our measures of IOp. Like IOp, IGM estimates come in many shapes and sizes, including regression coefficients, elasticities, correlation coefficients etc., and each of those either for levels or ranks.

We hope that future versions of GEOM will also include measures of intergenerational mobility, to be drawn from existing research. Watch this space.

Why Inequality of Opportunity matters?

Although estimates contained in GEOM describe a statistical association and are not interpreted as causal, the extent to which the inequality we experience today is inherited matters in at least three ways.

First, many people – including many philosophers of social justice – regard it as the most unfair part of inequality. Second, it is positively associated with the extent of overall inequality: the more inequality a society passes down generations, the more unequal it tends to be. Third, it tends to lead to wasted human potential and can be a source of economic inefficiency and even lower economic growth.

Glossary & Documentation

Although the basic idea of inequality of opportunity is both intuitive and powerful, measuring it requires struggling with some hard statistical problems. This inevitably creates some jargon. Although we have tried to keep it to a minimum on this site, here is a glossary to help you with any technical words that slipped through the net…
Income and consumption are two different observable outcomes, over which we measure inequality of opportunities. Income is “disposable household income”, i.e. the sum of all incomes going to a household over a certain period, minus taxes. Consumption is “household expenditure”, i.e. the sum of all monetary outflows in the household (e.g. food, clothes, energy, transportation etc.) over a period of time. While income is our preferred outcome, developing countries often have issues when collecting this information as it fluctuates a lot and can come from many different sources. So, consumption is often used to measure wellbeing – and inequality – in many of those countries.

There are different measures of inequality, each of which summarizes, in one number, the dispersion in the distribution. The Gini coefficient and the MLD are two such measures.

The Gini coefficient has perhaps become the ‘default’ measure of inequality, ranging from zero to one. When it equals zero, everybody earns the same income: perfect equality. A Gini of one means that all resources go to one person: maximal inequality. In practice, societies range somewhere between these extremes.

MLD stands for Mean Logarithmic Deviation. It is another way to measure inequality, which is more sensitive to gaps in the bottom of the distribution. The minimum value for MLD is also zero, but it doesn’t have a maximum.

Absolute and Relative measures are different ways to communicate Inequality of Opportunities (IOp). For example, if the GINI of Spain is 0.3 and the absolute Inequality of Opportunities (as calculated by the algorithms) is 0.15, then the relative Inequality of Opportunities is 0.15/0.3 or 50%. In other words, we can say that half of total inequality in Spain is explained by Inequality of Opportunities.

A type is a group of people who share exactly the same circumstances – or inherited, predetermined characteristics. Everyone belongs to one – and only one – type so, taken together, the types add up to the total population.

There are two prominent approaches to defining equality of opportunity:

The ex-ante approach is based on the idea that the set of opportunities open to individuals in a given type, before they exert any effort or make any choices, should be of equal value. Ex-ante inequality of opportunity is then the inequality between the values of these opportunity sets, often represented by the average income for each type.

The ex-post approach is based on the idea that the outcomes of individuals exerting the same degree of effort (or responsibility) should be equal. If one takes the relative position (or rank) of a person in her type’s income distribution to be an indicator of her relative effort, then ex-post inequality of opportunity is just an aggregation of differences across these distributions, at each and every rank.

Since IOp measures are essentially measures of inequality between types, in one way or another (see ‘ex-ante’ and ‘ex-post’ tab), it matters a great deal how the population is divided up into types. Having agreed on what inherited characteristics should be treated as circumstances, if we observed the whole population, we could just divide it up into groups within which everyone had exactly identical circumstances. These are the true types.

In reality, though, we often work with samples – not the whole population. This creates a trade-off: if we include too many types, relative to the size of the sample, we may end up overestimating IOp. If we include too few, we may underestimate it. ‘Trees’ and ‘forests’ are machine learning tools that allow us to pick the ‘optimal’ partition, given the data sets we work with.

We use two types of trees: conditional inference trees (CITs) are best suited to the ex-ante measures, and transformation trees are ideal for the ex-post measures. Because CITs can be sensitive to the specific sample that happens to be drawn from the population, Random Forests (collections of trees) are used to get rid of some of the variance, and obtain more robust estimates.

Nodes or leaves are the final empirical types into which the trees divide up the population. These may be a little different depending on whether a CIT or a transformation tree is used.
The idea of “Equality of opportunity” is a normative philosophical concept that proposes that the opportunities or resources available to people, rather than their final individual outcomes, should be made equal. It draws heavily on work by Ronald Dworkin, Richard Arneson, and Gerald Cohen, for example.  The latter has suggested that this nornative view has become “the currency of egalitarian justice” (Cohen, 1989). In the 1990s, these ideas were formalized and incorporated into economic theory by Roemer (1993), van de Gaer (1993) and Fleurbaey (1994), among others.  These authors expressed the concepts the philosophers had written about in terms of tangible economic quantities that could be observed and measured. Empirical work taking these ideas to data and estimating the first measures of inequality of opportunity began with Bourguignon, Ferreira and Menéndez (2007) and Lefranc, Pistolesi and Trannoy (2009). Important contributions followed with Checchi and Peragine (2010) and Ferreira and Gignoux (2011). Fleurbaey and Peragine, 2013 provide an excellent discussion of what they are and how they are related.  For a survey of the literature, see Ferreira and Peragine (2016). Incorporating some of the most recent methodological advances, GEOM relies on supervised machine learning techniques to define types and to obtain robust estimations of ex-ante and ex-post inequality of opportunity. For the ex-ante approach, we use the estimation procedures proposed by Brunori, Hufe, and Mahler (2023), who employ Conditional InferenceTrees (CIT) and Conditional Inference Random Forests (CForest) developed by Hothorn, Hornik and Zeileis (2006). to define the most appropriate partition into types. More detail is provided here. In the ex-post approach, we use the estimation procedures proposed by Brunori, Ferreira and Salas Rojo (2023), who employ Transformation Trees (Trafotrees) proposed by Hothorn and Zeileis (2021) to define the most appropriate partition. More detail is provided here. Finally, details on the common protocols we followed to prepare the data for analysis can be found  here. Documentation regarding each individual country and data set is available from the Country Profile tab.