Skip to main content icon/video/no-internet

Election night projections are made in the United States on the night of major primaries and major elections to help project the winners of key political races (e.g. president, senator, governor) and issue referenda. A small group of survey research experts make up the Election Night decision teams that make these projected calls. The same basic methods are used to project the winners by the National Election Pool (NEP), a consortium of ABC, CBS, CNN, FOX, NBC, and the Associated Press.

Each media organization makes its own projection decisions but relies on common data using this procedure and occasionally supplemented with its own. The method has evolved over the years, but it is essentially the same model developed and implemented in 1967 by Murray Edelman and Warren Mitofsky (then with CBS News) based on their work in probability methods at the U.S. Census Bureau.

Sources of Information

There are four possible sources of information about the election outcome in any given state that are used to make Election Night projections for that state: (1) the actual vote at sample precincts, (2) a statewide exit poll of voters at those precincts, (3) a statewide telephone poll of absentee (early) voters, and (4) the tabulated vote reported by counties throughout the state.

Precinct Sample

Like all estimation, the election night projections start with the sample. In each state where a projection is desired, a sample of voting precincts must be taken. Depending on the size and diversity of a state, a sample of 60 to 120 precincts is taken as the basis of a projection. This procedure for sampling is known as a stratified systematic sample proportionate to size.

Each state is broken into three to five geographic areas (geo-strata) based on their historical voting patterns. For example, Figure 1 shows the geographic areas (strata) for New Jersey, while Figure 2 shows an example of New Jersey's historical voting patterns.

Figure 1 Example of state-level geo-strata (New Jersey)

None

A recent past general election in a state is used as the base race for selection for that state. There are three important attributes in choosing the base race: (1) being recent, so as to minimize the geographic boundary changes that may occur from election to election; (2) being competitive, so that it shows a reasonable distribution of the vote; and (3) being typical, in that it reflects the ideology of the political parties that the candidates represent.

Figure 2 Example of historical voting patterns (New Jersey)

None

A listing of all precincts and vote counts in that past election is obtained and geographic codes are added. The precincts are sorted by geographic area and placed in order of total vote. The precincts in each area are then broken down into two separate size groups (usually at the median of accumulated total vote in the geographic strata). The two size groups are used to minimize the bias that can occur when sampling probability proportionate to size. The precincts in each group within a geographic area are ordered by their percentage of Democratic vote in the base race. A sample of precincts is then selected from each group proportionate to the size of their total vote in the base race with an exception in only one state. There are a couple of precincts in New Hampshire that are larger than the interval of selection, and in those cases the precincts are selected with probability of one (1.0). This procedure gives an excellent probability sample of the state for the past race.

...

  • Loading...
locked icon

Sign in to access this content

Get a 30 day FREE TRIAL

  • Watch videos from a variety of sources bringing classroom topics to life
  • Read modern, diverse business cases
  • Explore hundreds of books and reference titles

Sage Recommends

We found other relevant content for you on other Sage platforms.

Loading