Over 70 years after being outlawed, caste-based discrimination and the practice of untouchability against Dalits still regularly make the newspaper headlines. For a recent example, in 2018, in Mumbai, Sudharak Olwe held a photo exhibition on the stories of families who have lost a member as a result of caste hatred. Reported triggers of hate crime range from a phone having a ring tone paying tribute to the famous anti-caste figure B. R. Ambedkar to a lower-caste man speaking with an upper-caste woman. It is usually accepted that caste-based discrimination is a persistent phenomenon within Indian society. An open question, however, is whether the rising consciousness of its extent reflects an increase in the practice of caste-based discrimination or whether it reflects a shift in public attention and sensibility. Answering such a question requires a measure of discrimination to assess how the phenomenon has varied over time and, ideally, across place.

Discrimination is difficult to measure, be it via numbers or by using qualitative approaches. The concept is usually defined as the unequal treatment of equals based on the use of prohibited criteria linked to group identity. Discrimination occurs in a variety of contexts – in the labour market and other markets, during selection processes, or in many social interactions. While the existence of unequal treatment may already be hard to prove, partialling out the causes of that treatment is even harder. The main measurement challenge faced by researchers and policymakers has been the way an outcome is defined and isolated as the result of discriminatory behaviour when other causal factors coexist.

Overcoming the challenge of measuring discrimination is, however, key to be able to assess the extent of the phenomenon, its evolution, and where to target efforts to fight it. Researchers and policymakers in India and beyond have typically resorted to three main approaches. These are decomposition methods, direct data collection, and experimental methods (mainly correspondence and audit studies).

Decomposition methods, coming from Economics and Demography, allow one to analyse differential rates of one outcome (death, income) in relation to the affiliation to a particular demographic group (Kitagawa, 1955; Oaxaca, 1973; Blinder, 1973).