University of Illinois University of Illinois at Urbana-Champaign University of Illinois at Chicago University of Illinois at Springfield Institute of Government & Public Affairs Image Map
Sign Up For IGPA Updates
Connect with IGPA

The Illinois Opinion Monitor includes periodic random-sample surveys of the adult population in Illinois (and, occasionally, the nation); and periodic non-random-sample surveys of policy experts in Illinois.

The first wave of the mass sample, the pilot study for prong (a), was undertaken in the autumn of 2006, as part of the Cooperative Congressional Election Study (CCES), implemented by Polimetrix. The CCES was a consortium effort directed by Professor Stephen Ansolabehere of the Department of Political Science at MIT. It was the largest-ever election study, featuring more than 100 researchers from 39 universities.

The CCES was, in effect, 40 studies, 39 modules designed by individual teams and each featuring about 1,000 respondents plus a grand module of “common content” administered to 38,443 Americans. All surveys were completed in October and November 2006, in pre- and post- election phases, respectively. The common-content sample for CCES is a nationally representative sample, constructed by sample matching in two stages. First, a random sub-sample of 36,501 individuals was drawn from the 2004 American Community Survey (ACS), conducted by the U.S. Bureau of the Census. The 2004 ACS was a probability sample of size 1,194,354 featuring a response rate of 93.1%. Each respondent in the selected ACS sub-sample was then paired to the closest matching active PollingPoint panelist using a measure of distance that incorporates age, race, education, and gender. Ranking was achieved by iterative proportional fitting, and final weights were trimmed to lie between 0.33 and 3.

Polimetrix maintains a panel of public-affairs survey respondents with more than one million members. These members are not a random sample of the general population, but they are diverse in age, race, gender, education level and place of residence within the United States, so that one can generate effectively random samples by means of matching, as described above.

The Illinois sample of the CCES was assembled by the same sample-matching technique, except that Polimetrix panel respondents were matched exclusively to individuals residing in Illinois from the ACS. The resulting sample was weighted to match state demographics (age, race, education, and gender), and a separate set of weights were generated by state senate district for use with selected items concerning local representation.

For both the national CCES sample and Illinois CCES sample, margins of error can be computed in the same manner as for traditional random-digit-dialing phone surveys, so that they depend mostly on the number of respondents, but also (slightly) on the skew of the variables in question and on the level of confidence one desires. Most poll results are reported with margins of error that reflect 95 percent confidence, although this point is routinely neglected in non-technical reports. So, for example, a typical poll-result might be that 45 percent of respondents expressed approval of a public figure, and there is an error margin of ? 3 percentage points. A more precise expression of this result is that we would expect the interval (in this case 42 percent-48 percent) to include the true level of approval in the entire population being surveyed 95 times, and not to include the true value five times, were we to draw independent random samples of this same size 100 times in succession.

All polls based on randomization and sampling, whether undertaken by telephone, face-to-face interview, or Internet, also have potential biases, distinct from sampling error. By comparison with commercial polls (mostly done by random-digit dialing by phone), the CCES was more accurate in forecasting 2006 congressional election results.

For further technical details on the Polimetrix panel and sample matching, see the Polimetrix White Paper series.