You need to compare the party affiliation cited by respondents, to the known party affiliation numbers in the population. That way, you can make the adjustments (weighting, after the fact being the usual way to do this) to make the sample more representative of the population being polled. WriteOn and marktwain have it right.
Mr. Newport is referring to another technique for attempting to match a sample to the population — by asking for party affiliation first. That way, they could balance the sample by calling until they had sufficient Republicans (or independents, or whatever they were short of). They would only interview people, if they were affiliated with the underrepresented party. They don’t do that — fair enough. However, they should have made the mathematical adjustments, as described in my first paragraph.
He specifically said they don’t do this.
He said they ask party affiliation at the end of the poll questions and their responses are for informational purposes only and they don’t do any weighting by party affiliation.
He we very specific.
Party self-identification and actual registration are very different things. Someone could be registered unaffiliated (in some states they don't register by party) and tell a pollster they're a "Democrat" one week and "Republican" the next.