If you don't select the voters by party affiliation then the affiliation they indicate is what the sample says the electorate will be this time around. Comparing the numbers to what you expect based on past elections is a good way to double check your methodology, but you don't throw out the results based on that. You just take a bigger sample to reduce the chance of a skewed party affiliation sample the next time around. If you keep getting a consistant split it means the electorate has shifted over say sometthing like preferring a party known for a strong defense posture.

Regards,
Mark