Here's something that an industrious undergrad might be able to pull off (although he or she might need a partner). Go to a local supermarket and sit near the checkout lines. Record the gender and race of each employee in the checkout line and each customer that goes through the line (you probably ought to ask a manager for approval first). Also record the number of people that go through each line in, say, an hour. Repeat this at some other grocery stores.
This data should allow one to determine:
- Whether customers are more likely to go to an employee of the same race or gender than pure chance would predict.
- How much customers are willing to pay, in time, to go through a checkout line operated by someone of the same race or gender (assuming a same-race or same-gender preference exists). That is, you can find out how much they are willing to pay to indulge their racism or sexism.
There may be some confounding factors--self-checkout or express checkout, and the relative attractiveness of the employee, but I think it should be a fairly straightforward project.