> Methodology: This study was conducted Oct. 11-16 from a representative sample of 978 students nationwide from 2-year and 4-year schools. The margin of error is +/- 3.4 percentage points. The Generation Lab conducts polling using a demographically representative sample frame of college students at community colleges, technical colleges, trade schools and public and private four-year institutions.

Maybe there was some bias they didn't account for; a thousand students isn't that many.

A thousand is enough for a representative sample.

Could be the people that spend time answering random surveys online and the people that use dating apps are different.

It's not even clear how they were recruited. Generation Lab says only that it was via proprietary methods i.e. they won't tell you. Why not? Is there anything so sensitive how they got in touch with students? Normally, polling firms are more open. It suggests they don't want you to know where their lists came from (could be as bad as MTurk).

It is enough, but only if you collect a good sample. I'm not sure what a good sample is. In college I'd expect a big difference between undergrads and grad students (as students get older they are more likely to settle down with someone and marry, so even seniors vs freshmen should show a difference)

This is absolutely false in lots of scenarios, one of them being when you're dealing with a heterogenous population, which seems to be the case here.

Indeed. Also a larger sample doesn't help with "some bias they didn't account for", it just marginally narrows the margin of error.