you're definitely on the right track - but it still would take some more work. you could get a random list of colleges, get a random page from each, then weight each by the total student population so you'd take, say 5 names from the Michigan State list for every one name from the U-M Dearborn list.
However, there's a few more snags -
College directories list every student regardless if they're full-time or part time or grad student. You may be able to get the grad-or-undergrad info from the directory, but you can't always. And I'm afraid that there's no way the student directory will tell you if the student is just taking one class, two classes, or a full load. And that's important because, presumably you're looking only for full-time students, not someone just enrolled in one class.
Then, of course, there's the problems of unlisted numbers (I'd guess many more females would not list their numbers in the student directory), students who are so busy working or studying they can't be reached by phone, and those who decline to participate in the survey (If I was studying for a test, there's no way I'd take time out for this survey), or those who just hang up.
...Simply put, while I agree with your skepticism and laud you for it, I doubt their method of survey was as you claim. I wish we KNEW, however...
The fact that they don't tell us that they carefully selected their sample to represent the entire student population (and I've looked for the info) screams to me that they didn't carefully select anything. I mean, if they went through all the work to carefully select the sample I'd think they would report that. In other words, my bs-detector is going off.
What I did find from looking through their data does not fill me with confidence -
The IOP conducts these surveys once or twice a year. In each of these they claim to survey the opinion of "college students." So I would expect some of these numbers to be close to the same - not wildly different in each survey:
Age of respondent - for the October/03 survey, 20% of respondents were aged 21-24.
for the April/00 survey, 40% of respondents were aged 21-24. That's twice as much. (the other age-percentages were 18-20 77% in '03, 58% in '00).
In Apr/03 the total pecentage of african-american, hispanic, and asian students surveyed was 27%. Six months later, in the Oct/03 survey, the total percentage of the same minorites was reduced by almost half to 14%. (I'd guess that had something to do with their "more conservative" finding).