From the Editor: In a new academic paper, Dr. James E. Campbell, Chairman of the Political Science Department at the State University of New York- Buffalo has analyzed The Cook Political Report's pre-Labor Day House ratings going back to the Report's founding in 1984. In 11 of the 13 elections in which the Cook Political Report published new ratings between July 1 and then end of August (all except 1986 and 1990), 99.8 percent of the 3,387 races rated by the Cook Political Report as Solid Republican or Solid Democratic in July or August of an election year went by way of that party, 94.9 percent of the 641 races rated as Likely Democratic or Likely Republican fell the way the Cook Political Report predicted, and 85.7 percent of the 441 races rated Lean Democratic or Lean Republican broke in favor of the leaning party. Of the 130 Democratic-held seats rated as Toss Up, 49.2 percent went for Democrats, and 55.0 percent of the 160 Republican held seats rated as Toss Up were won by the GOP.

Keep in mind that many race ratings shift after Labor Day to reflect changes in competitiveness, but pre-Labor Day ratings have proven highly predictive nonetheless. After examining the election by election accuracy rate of Cook Political Report ratings, Campbell, currently the President of Pi Sigma Alpha, The National Political Science Honor Society, decided to incorporate the Cook Political Report ratings as part of his Congressional election forecasting model, to be presented in a paper at the American Political Science Association's annual meeting in September. (Ed.)


Election and Date of the Cook Report % of Seats Won by Pre-Labor Day Expected Party or Prior Party in the Case of Toss-up Seats
D Solid D Likely D Lean D Toss-up R Toss-up R Lean R Likely R Solid
1984 (7/31/94) 99.4 (175) 94.7 (57) 69.6 (23) 35.7 (14) 80.0 (5) 88.9 (18) 100.0 (36) 100.0 (107)
1988 (8/17/88) 100.0 (214) 90.0 (20) 100.0 (20) 100.0 (2) 68.7 (16) 93.7 (16) 100.0 (17) 100.0 (130)
1992 (7/24/92) 100.0 (121) 93.5 (62) 88.2 (51) 58.3* (24) 70.0* (20) 82.8 (29) 100.0 (36) 97.5 (79)
1994 (8/19/94) 99.2 (132) 84.9 (53) 69.2 (26) 18.9 (37) 89.5 (19) 100.0 (26) 97.5 (40) 100.0 (101)
1996 (8/28/96) 100.0 (100) 100.0 (40) 90.6 (32) 77.3 (22) 48.3 (29) 86.8 (38) 98.1 (53) 100.0 (120)
1998 (8/28/98) 100.0 (146) 100.0 (30) 100.0 (18) 80.0 (10) 36.4 (11) 94.1 (17) 95.0 (20) 100.0 (182)
2000 (8/9/00) 100.0 (172) 90.9 (22) 90.0 (10) 40.0 (5) 53.8 (13) 87.5 (16) 100.0 (19) 100.0 (176)
2002 (8/7/02) 100.0 (175) 93.7 (16) 83.3 (12) 33.3* (3) 75.0* (4) 94.1 (17) 92.9 (14) 100.0 (189)
2004 (8/20/04) 100.0 (174) 100.0 (12) 90.9 (11) 20.0* (5) 33.3* (6) 100.0 (9) 95.2 (21) 100.0 (194)
2006 (9/1/06) 99.5 (183) 100.0 (11) 100.0 (10) N/A (0) 33.3 (18) 37.5 (16) 75.0 (20) 98.9 (177)
2008 (8/30/08) 99.5 (206) 100.0 (12) 93.3 (15) 62.5 (8) 31.6 (19) 54.5 (11) 90.0 (30) 100.0 (134)
Total 99.8 (1,798) 94.0 (335) 87.7 (228) 49.2 (130) 55.0 (160) 83.6 (213) 95.8 (306) 99.7 (1,589)

* indicates that there were some districts that could not be classified as having been held by either of the parties. This reflects redistricting and reapportionment. The top number in each cell is the percentage of seats that were won by the party to which the seat was classified as being solid, likely, or leaning or had previously held the seat if it was a toss-up. The number in parentheses is the number of seats in the specified category.

In developing a new model to forecast aggregate national seat losses or gains by the parties in House elections, I evaluated the district-by-district assessments of the status of House races made by the Cook Political Report in elections since 1984. The district-by-district assessments rate the state of each House race on a scale as being “Solid,” “Likely,” or “Leaning” for either the Democrats or Republicans or as being a “Toss-up” when “either party has a good chance of winning.” Charlie Cook and his associates generously provided me with copies of their “House Races at a Glance” lists of campaign assessments for the months of July through October in the elections from 1984 to 2008. Since I was concerned about having a forecast prior to Labor Day, I used the latest “House Races at a Glance” assessments that were available between July and Labor Day. Unfortunately, since these ratings were not published between July and Labor Day for either 1986 or 1990, ratings in these two elections were not examined. This left eleven elections in which Cook’s ratings before Labor Day of the election year could be evaluated by a comparison to the actual election results in the districts. In most election years, the rating that was evaluated was made in August of the election year.

In evaluating the ratings, I compared the rating of each district in each year to the election outcome in the district. The summary accuracy of the ratings are presented as the percentage of districts in a rating category that were won by the expected party. Since neither party is an expected winner in the toss-up category, the percentage in these cases is that of the party previously occupying the seat (a hold). Because of redistricting and reapportionment, there are some elections in which there were districts in which the original party holding the seat could not be determined. The details of the evaluations are presented in Table 1.

Several aspects of the ratings’ success should be noted. First, the assessments have a very high accuracy rate. In the eleven elections, 97.5 percent (2,056 of 2,108) of districts that were rated as solidly, likely or leaning to the Republicans were won by the Republicans. On the Democratic side, 97.8 percent (2,309 of 2,361) of districts classified before Labor Day as solidly, likely, or leaning to the Democrats were won by the Democrats.

Second, if a race is classified as solidly favoring a party, that party almost always ends up winning the contest. In these eleven election years, there were 3,387 races called as solid for a party and that party went on to win 3,379 of the elections (99.8%). The rare losers were equally divided between the parties, four shocked Democrats and four shocked Republicans.

Third, to be classified as “likely” is nearly as safe as being classified as “solid.” Over the eleven elections examined, there were 641 races classified as likely one way or the other and the likely party won 608 times (94.9%). An unusually large number of the “likelys” who lost were Republicans in the Mark Foley scandal election of 2006 (5 of 20). This was also the election in which two Republican “solids” surprisingly ended up on the short side of the vote count (Kansas 2 and Pennsylvania 4).

Fourth, the “leaning” category proved to be the one to watch. While the overall win-rate for the favored party was fairly strong (85.7% or 378 of 441 races), the win-rate between the parties’ leaning districts varied quite a bit from election to election. There were years in which a party would carry all or nearly all of its leaning races (Democrats in 1988, 1998, 2000, 2004, 2006, and 2008 and Republicans in 1988, 1994, 1998, and 2002) and then there were other years in which a party would carry less than seventy percent of the districts that leaned its way (Democrats in 1984 and Republicans in 2006 and 2008). The high level of wins for the expected party as well as the variance between years and parties in the win-rates makes the leaning category difficult to interpret from a forecasting perspective. It would seem that one explanation of the variance in the win-rate of these leaning districts is that events intervened between Labor Day and the election that threw some of these leaners into the toss-up category or beyond. This seems quite plausible for the Republican leaning districts in 2006 with the Foley scandal breaking during the campaign and in 2008 with the Wall Street meltdown in mid-September.

The hold-rate (the prior party holding the seat) of the toss-up category also varies quite a bit, but proved to be the most useful for forecasting the national seat change. As billed, either party could win a race in the toss-up category. Where the previous party could be determined, Republicans held 55 percent (88 of 160) of their toss-ups and Democrats held 49.2 percent of theirs (64 of 130). On average over the eleven elections, a race being in the toss-up category really amounted to a toss-up. As a result, the net number of the parties’ seats being classified as toss-ups provided the basis for a “seats in trouble” index for my forecasting model.1 When combined with an index of presidential approval, the model accounts for about 90 percent of the variance in net national seat change in these eleven elections.