The population is all Yale students who graduated in 1924 still alive at the time of the survey (the late 1950s). The sampling method was not likely to produce representativeness: there was no attempt to choose people randomly (people either were or were not on the address list Yale kept, and they either did or did not choose to respond if contacted from that list), and there are good reasons (see below) to expect the sampled and unsampled people to differ substantially in income.

You can make a good case for any of the following three kinds of bias -- selection bias, because the address list was composed of people who went to the trouble to stay in contact with Yale all those years, and such people are quite likely to differ in income from those who did not keep up with the college (Why does Yale have their names? Maybe because they are faithful donors to the alumni fund); nonresponse bias, because of the people who got the questionnaire those who chose to respond are again quite likely to differ in income, if the questions are about income (would you send it back if you weren't proud of how well you were doing?); and response bias, because whoever put the survey together may well have had a vested interest in making these people look prosperous and may have worded the question in a way that encouraged people to exaggerate their income (this third source of bias seems less plausible to me than the other two, but still possible). At least the first two of these sources of bias seem likely to have figured prominently in the $ 25,111 estimate.

Given the directions of the biases, it seems quite likely that Time's figure was substantially too high.