Steps
1. Decide exactly who is doing the discriminating, and what they are discriminating on the basis of. Is it an employer, a landlord, an industry, a society? Are they discriminating on the basis of race, religion, gender, class, or something else?
3. Based on the above, decide how you plan to proceed. The most important differences between the two methods described are how many people are doing the discriminating, and how easy it will be to collect the necessary data. The statistical method works best when many people each do a little bit of discrimination, but to be reliable it requires collecting a lot of data. The sting method requires little numerical data, but focuses the attention on a few specific individuals and might miss sporadic discrimination.
Method 1 of 2: Statistical
1. Create a specific hypothesis describing as specifically as possible the nature of the discrimination. A good hypothesis might look like, "Hospitals in California offer lower wages to Hispanics than to equally qualified non-Hispanics."
2. Collect data that could provide evidence for or against such a claim of discrimination. You could look for such data from sources like the United States Census or the Bureau of Labor Statistics or you could conduct an independent survey to collect your own. For the example, you would need information on the wages of a random sample of Hispanic and non-Hispanic California hospital employees. Your data should include any other factors that might affect wages such as education and experience. Include other information about the employees that might affect wages even if you think it shouldn't. You don't want to accuse someone of discriminating against Hispanics when in reality they discriminate against Jews. The fact that an ethical company would do neither is no excuse for academic sloppiness.
3. Import all your data into your favorite statistical analysis package. Run a multivariate least squares regression for wages against all the other variables collected.
4. For each variable, look at the coefficient and the p-value. The coefficient represents the magnitude of the discrimination and the p-value represents the likelihood of any discrimination at all. For example if the analysis of your data said that Hispanic had a coefficient of -.54 and a p-value of .02, it would mean that being Hispanic cost an employee $0.54/hr. and there was a 2% chance that the discrepancy could be explained by random chance in the absence of racial discrimination (and therefore a 98% chance that it was due to being Hispanic.)
5. Publish your findings. Expect the accused to look for possible shortcomings in your methodology. Don't make their job easier by omitting obviously relevant variables or choosing a biased sample.
Method 2 of 2: Sting Operation
1. Create a specific hypothesis describing the nature of the discrimination. For example, "The landlord of Evergreen Apartments doesn't like renting to African-Americans."
2. Send the individual that you suspect of racial discrimination applications of two people who are equals with respect to financial stability, but of different races. Most applications won't directly ask for race, but it's often possible to guess someone's race from their name. Use a stereotypical white name for one applicant and a stereotypical black name for another.
3. Wait for the responses. If both applicants are treated equally, there is no evidence of racism. If one gets invited to tour the unit and make a deposit while the other is told that there are no vacancies, there is a very good case that the applicant's perceived race was a factor.
4. The case can be strengthened by repeating this test several times. One case where the responses differ is suspicious but might be explained as a mistake, not necessarily racially motivated. Five cases where the "black" applicant fares better than the "white" one is almost indisputable evidence that race was taken into account.
