Godless Capitalist, a co-founder of GNXP, suggests an approach for getting real achievement data by caste background for Indians in North America that's similar to what Nathaniel Weyl and Stefan Possony did in the 1960s: surname analysis. Weyl and Possony made up lists of surnames that tend to belong to one ethnic group, like Clark is English, Wallace Scottish, Sullivan Irish, Lloyd Welsh, Schmidt German, and so forth; then figure out what percentage of the population has each surname from Social Security Administration data; then compare the base percentages to how frequently these names show up among lists of high achievers.

GC suggests you could do the same thing with social networking sites like Facebook where people put down their colleges, degrees, and jobs by comparing achievement levels to names associated with different castes and regions. For details of how to do it, see here.