Our scorecard tracks the race to control the U.S. Senate using the most recent polls in the 13 most competitive contests. We classify a state as "likely" to go to one party when the average result across the last five public polls shows one candidate with a statistically meaningful lead (for the statistically minded, our standard is a lead of a least one standard error, a modest statistical advantage). We classify as "tossups" those races in which neither candidate shows a significant lead over the last five polls. Safe seats are those for which a candidate consistently has a statistically significant lead (in this case, a lead of more than two standard errors).

We then total up the numbers for each party, including those races classified as "likely," "safe," or in which the incumbent faces no contest in 2006.

The "momentum shift" arrow derives from averaging the recent trends across all 13 of the competitive Senate races. When the recent trend across the 13 states is statistically meaningful, the "momentum shift" arrow will point toward the party it favors for the most recent period.

The table of the 13 competitive races shows the last-five-poll average in each. Blue state names indicate a statistically meaningful lead for the Democrat ("likely" status or better), while a red state name indicates a meaningful lead for the Republican. Black indicates a dead heat and green indicates an advantage for the independent candidate. The momentum shift arrow for each race will show when the polls indicate a statistically meaningful trend toward one of the candidates over the last five polls. Click the link for "complete race details" to bring up a chart showing the most recent results. We have included all polls based on random probability sampling, including those conducted using an automated methodology rather than live interviews.

Why a five-poll average? Results for pre-election polls often vary due to random-sampling error, as well as differences in methodology (question wording, sampling, the survey mode, or the way pollsters define likely voters). While averaging is an imperfect solution, we believe that a five-poll average provides a more reliable snapshot of data available for each race than focusing on only the single latest poll does.