Polimetrix uses a matched random sample. The firm begins with two lists, a list of all consumers in the United States, which covers approximately 95 percent of the adult population, and a list of people who have agreed to take surveys for Polimetrix as a part of their PollingPoint panel. All Polimetrix surveys are conducted on-line using this opt-in panel of respondents. For each list, Polimetrix has an extensive set of demographics.

First, a random sample of consumers is drawn. For each person drawn from this sample a list of key demographics is recorded. In essence, each individual drawn is represented as a cluster of demographic characteristics, including age, income, education, race, gender, longitude and latitude, etc. Second, Polimetrix uses a matching algorithm to find the PollingPoint panelist who is the closest match to the person drawn off the consumer file. In this way an entire matched random sample is constructed for all people in the sample.

[Schema slide from Doug’s MPSA talk.]

Learn more about this sampling methodology. [.pdf 131KB]

The sample drawn for the CCES will be a stratified national sample of registered and unregistered adults. The choice of strata will guarantee that the study achieves adequate samples in all states. There are three sorts of strata in the sample: Registered and Unregistered Voters, State Size, and Competitive and Uncompetitive Congressional Districts.

By stratifying on registered and unregistered voters we can create a nationally representative sample of US adults using appropriate sample weights. Because the preferences of voters is of particular interest to many researchers we will oversample registered voters. Approximately three fourths of US adults are registered to vote. In midterm elections approximately half of registered adults vote. The oversample of registered voters will mean that the actual voters in the sample approach one-half.

Stratification on state size is required to guarantee adequate sample sizes in small states. There will be four strata for state size: one Congressional District states, two Congressional District states, three Congressional District states, and four or more Congressional District states.

Stratification on competitive congressional districts will guarantee an adequate number of districts in which there are very active political campaigns in the fall election.

All told this sampling scheme minimizes the number of strata, so as to prevent mistakes, while guaranteeing adequate coverage of all relevant jurisdictions. There will be 16 strata.

Each state, then, will have sufficient coverage so that any team interested in the general politics of a given state will have a state-level survey of approximately 60 questions on their state. In addition, if a sufficiently large group interested state politics emerges they may be able to trade questions across groups in such a way as to augment the Common Content. If a large enough number of teams agree to swap content in this way, then they can trade questions such that every time a respondent from a particular state is chosen in any survey within the Group then the question relevant to that particular state is used.

For example, suppose that 15 teams have particular state level questions that they would like to ask. Say, Ohio wants to ask two questions about Ohio propositions, Michigan wants to ask two questions about a hot contest for Secretary of State, Florida wants to ask two questions about the 2000 election, etc. Every time an Ohio respondent arises in any sample from among this Group’s members’ surveys, the Ohio questions are asked. Every time a Michigan respondent arises in any sample from among this Group’s surveys the Michigan questions are asked. And so forth.

In this way groups can exploit the sample design to develop unique state-level surveys. This strategy seems particularly attractive for larger states from which a disproportionate number of cases will likely be drawn.