If they do a truly random poll, and discover that the results don’t match what they have decided in advance is to be the R/D/I mix (e.g. 2012 turnout numbers), they have to throw away responses to achieve that mix.
Which responses are thrown away and how do they decide? Are they more inclined to toss “anomalous” responses, such as Democrats supporting Republicans?
They have different formulas to make sure the sample matches the population (ie: if population is 49% democrat the sample of democrats is adjusted to match). Which makes the results SEEM to be scientific but there is just no telling who will win on Nov 8th. So it comes down to this: Polls might help candidates make decisions on where to spend money, but otherwise they are useful only as propaganda (voter suppression) which is exactly how they were used in 1996 to demoralize the Dole/Kemp ticket.