Chris Bowers at MyDD has developed an interesting theory that national polls are inflating Hillary Clinton’s lead in the Democratic presidential race. He explains his theory here, here and here. Then Mystery Pollster comments here.
To summarize, national pollsters are using subsets of their national polls to ask about primary voting intentions, but those subsets are too all encompassing to accurately reflect the primary voting population.
For example, if you were to sample 1,000 voters for a national survey, you might get 400 Democrats and 400 Republicans. Then the primary questions would be asked among those 400 from each party. However, the primary voting population is much smaller than that. Not every Democrat and Republican who votes in a general election will vote in the primary. Primary voters are more ideological and better informed, and would be less inclined to be affected by the draw of Hillary Clinton’s celebrity…according to Bowers.
I tend to agree, theoretically. At Public Policy Polling we go about primary polling completely differently. We don’t use subsets of a general electorate poll, but conduct completely separate primary polls. We sample only households that have a history of voting in party primaries. That way we know we are only reaching those likely to vote in a primary.
Therefore, I suggest that the best primary polling would be those conducted using a sample based on voter history. Unfortunately, that type of sampling is unavailable nationally—only about 3/4ths of states have that information available.