I saw another study where they found phone numbers for mental health professionals, and they called and left a message, asking someone to call them back to set

I saw another study where they found phone numbers for mental health professionals, and they called and left a message, asking someone to call them back to set up a first appointment. The had other people rate the voices as "sounds black" / "sounds white," and "sounds white-collar" / "sounds blue-collar." That way, they measured the correlation between "sounds black" and "doesn't get call backs." As you can imagine, "white" sounding people got more call backs. And "white-collar" sounding people got more call backs. I don't know if they reported which was more beneficial, sounding white, or white-collar. My point being, if you have a large enough collection of names, and you measure how the candidate appears to them, you can then correlate the appearance with the outcome, showing the bias against that appearance. You don't need to have the researchers generate their own candidates.