Horse race: This is the "standard" campaign poll. Likely voters are asked whom they would vote for "if the election were held today." They are also usually asked about their views on specific issues that are deemed important. Most published polling during campaigns is paid for by news organizations, or conducted by academic groups that seek publicity or focus on specific issues rather than on specific candidates.

Tracking: These are horse-race polls that are taken with small samples (sometimes only 100 people) and very frequently (at least weekly, and sometimes nightly). Since they tend to be sensitive to external events, they are used to measure the effects of such events. Candidates and academics tend to use them more than news organizations do.

Exit polling: On Election Day, voters are polled (after they vote) about whom they voted for and why. News organizations tend to ask mainly about how the vote went, so they can report results as polls close (or sooner, in some states). Academic groups and political organizations are more interested in why people voted the way they did, and when they made up their minds. These results are often reported as insight, long after the campaign is over.

Cluster analysis: These are infrequently done, because of the large sample sizes. This type of poll is designed to categorize the electorate rather than the candidates. One of the best examples was commissioned by the Times-Mirror Company for the 1988 presidential elections. The poll divided the electorate into a dozen "clusters," each melding a unique combination of beliefs and desires. God and Country Democrats, for instance, tend to be patriotic, have lower- than-average income, attend church and are hawkish on foreign policy and liberal on domestic policy. The idea is to go deeper than the simplistic labels of "liberal" and "conservative" that we normally apply to politicians.