Let’s assume some Democrats go to trump rally’s for the experience. If you take say 100 people chosen randomly from a rally you could estimate how common that was. However, if you look at interviews of people at Trump rallies the interviewer may have been more or less likely to include such people. As such what your examining is people the interviewer found interesting who where also at a rally which is a very different group.
The surest way to get sampling bias is to some other criteria after the initial selection that can exclude samples from your examination. This might be as obvious as people who hung up on political polling or it can happen before the selection process. Aka I want 1,000 random Americans very often excludes people living outside America at say embassy’s, the homeless, prisoners, children, etc depending on how people are selected.
For a given experiment you could look at how people where selected, but when combining random interesting examples you have no idea what’s being excluded which is why their considered worthless.