Poll methodology

Over the course of four weeks, Penn Schoen Berland is polling for The
Hill in 42 key toss-up races that will be critical in determining
control of the U.S. House of Representatives. This project,
accomplished by a team led by Mark Penn, founder of the firm, will help
The Hill’s readers attain unique insight into the thinking of voters
who will decide which party will likely control the next Congress.

Polling is divided into four weeks so that each week we conduct statistically significant polls in 10 to 12 different target races across the U.S. The first week is focused on districts with first-time incumbents, the second week on open-seat districts, the third week on districts with two-term lawmakers and the final week on districts with long-term incumbents.

{mosads}Data for each district is based on 400 completed phone interviews with likely voters (those who say they will definitely or probably vote in the upcoming election for Congress and are or will be 18 or older and registered to vote in their state at the time of the election). Interviews are conducted over three days for each district and the margin of error for each district is plus or minus 4.9 percent, larger for subgroups.

In aggregate, each week of polling will consist of 4,000 to 4,800 respondents, which carries a margin of error of less than plus or minus 1.6 percent of the areas sampled that week.

The sample for each district undergoes two modes of randomization before calls are made. The composition of data is scrutinized for each district, and if necessary, key demographic information is adjusted based on information from past elections, current voter registration and census data.

When reading the results, it is also important to keep in mind what this project is not.

When aggregated, results will only represent the aggregate of likely voters in the particular swing districts tested each week. They will not represent a national sample, but a sample of a group of key battleground areas. For example, the samples will not include strongly Democratic or strongly Republican districts, which are unlikely to change party in November. They will likely contain many fewer minority voters, who often reside in those districts. The samples will only represent the opinions of those likely voters who reside in toss-up congressional districts that will have an impact on control of the House.

Additionally, we are polling only among likely voters who say they are definitely or probably voting in this November’s elections for Congress. This sample is therefore going to differ not only from the general population but also from overall registered voters in those states and also from the 2008 presidential election, in which turnout will likely end up being much higher.

Penn Schoen Berland