Mean across methods

(1) The forecast shown at the conference was 49; it was revised upward a few days later after adding another variable to the model. 

(2) The original forecasts were -30 with a midterm elections model and -27 with an all elections model, both models estimated with elections beginning in 1914.  On 9/15/10, he  issued a revised set of forecasts with modified versions of the models, this time estimated with elections held since 1870.  The midterm elections models forecast -34 with all 35 midterm elections and -36 with a trimmed data series that omits several outliers. The median value is entered here.

(3a) From their website: "The Cook Political Report's current outlook is for a Republican net gain of at least 40 seats."  This assessment is not consistent with the result one obtains by following the method used to infer a forecast for CQ Politics or RCP.  The method is described in footnote 4. Applying it to the Cook Political Report's latest "dashboard" yields a loss of 30 seats.  Perhaps Polly is being too generous in giving each party all its "lean" seats.  The Democrats have four times as many seats in this category as the Republicans (32 v. 8).  By assigning to each party a 0.6 probability of retaining a lean seat, one obtains a loss of "at least 40" for the Democrats. But the probability may not be symmetrical. 

(3b) In a 9/25/10 column,"After the Deluge of 2010," Charlie Cook wrote that the best-case scenario for the Democrats is a 30 seat loss (see footnote 3a) and a worst case scenario is a 60 seat loss.  Then he added, "Let the record show that, in my experience, the outcome will probably be something between the best and worst scenarios." Accordingly, the median between the best and the worse case scenario, a 45 seat loss, is entered.

(3c). In "Dems Batten Down the Hatches" (10/9/10) Charlie Cook wrote:

"In the House, it still looks like a very solid bet that Republicans will get the 39-seat net gain they need to capture the majority. I've decided not to play the guess-how-much-the-watermelon-weighs game; this one looks well over 39 to me, but I'll let other people toss around big numbers. Thirty-nine is what's important; betting on more than that is just picking a number out of the sky. . . . "  (emphasis added). 

(3d0.  In "House Flips, the Senate Doesn't" (10/16/10), he wrote, "At this juncture, I am still sticking with a 1994-level outlook: Eight Senate and 52 House seats are the over and under, with a 50 percent chance that Republican gains will be higher and a 50 percent chance that they will be lower." 

(4) CQ Politics and RCP assign "safe, "likely" and "lean" seats to each party, and the rest are regarded as tossups.  The forecast was obtained by adding the Democratic seats in the first three categories plus half the number in the tossup category, rounding up in case of an odd number.

(5) Rothenberg told Polly that this is a projection, not a forecast.  Also, Polly takes the median of the range of seats projected (37-45), rounding up in case of a fraction of at least half a seat.

(6) The contracts are specified in five-seat increments.  The forecast was obtained by taking the median value within the increment with a probability greater than 50%.

(7) Electoral-vote has been remarkably accurate in its predictions in the last several election cycles.  The website includes detailed information on the algorithm, as well as available poll data.