Many pollsters say party ID is an attitude, not an attribute. The distinction is key, as attitudes can change while attributes are more stable. And attributes, such as race, age, and gender, are measured regularly by the U.S. Census Bureau, providing benchmarks pollsters can use to tweak their numbers—making adjustments to render samples more representative of the group whose opinion they seek to measure. So if, say, just 6.5% of the people answering a poll are between 18 and 24 years old, but 13% of adult Americans are in that range, each respondent in that group has his or her answers counted twice to make up for the shortfall.

There are no reliable, stable numbers, however, for how many Americans are Democrats, Republicans or independents. So if a poll finds that Democrats outnumber Republicans by 10 percentage points in the sample, after weighting for factors such as age, than many pollsters accept this as a finding rather than a mistake: More people are identifying themselves as Democrats. …

For pollsters who favor weighting by party ID, the tricky part is what numbers to use: Who knows just how many Democrats and Republicans are out there? Registration numbers won’t cut it, as states’ rules on party registration vary widely. Relying on exit polls from prior elections doesn’t work, because even if people haven’t changed their party preference much, some people have reached voting age while others have died.