Originally Posted by DJ's left nut
Some do, some don't.
Pollsters can, and often do (especially when trying to make projections based on their believed electorate), filter through results and/or drop randomly selected samples if they don't fit the projection model.
The guys that are trying to really predict will absolutely tailor their sample to fit a projected voter pattern, otherwise it's just random chance.
As far as I know, only one group does this, and they finished almost last in this election. Choosing the numbers for each party is just choosing the outcome you want to see, which defeats the whole purpose of taking a sample.
You say not doing this leaves you with "just random chance", but the whole purpose of taking a large sample is to reduce the randomness to a small and quantifiable amount. When you aggregate this across many samples, that randomness goes away almost completely, hence the derision poll deniers got and also the reason they were so wrong.