In 1998, there were six generic congressional ballot polls in the field entirely during the final week of the campaign: Gallup, Pew, Fox, ABC, and CBS / New York Times. ABC had both a 44% turnout model in the field, and a 36% turnout model in the field. Since turnout in 1998 for 38%, for this study I used their 36% turnout model. CBS had a "very likely" voter model in the field on their own, and also a "likely voter model" in the field in conjunction with the New York Times. Because they are two different models with two different sponsors, I included both. Here is the table:

      Rep       Dem
ABC      48       46
Gallup      45       49
Pew      44       46
CBS      48       49
Fox      48       38
NYT      50       47
Mean      47.2       45.8
Actual      49.1       47.7

(Source for final vote totals.) The mean of these polls predicted the final national vote totals almost exactly, as the Republicans margin of +1.4 was also seen in the final vote totals. Only two polls, Gallup and Fox, were noticeably off the final results, though I think that Gallup might still have been within the margin of error (if only barely). Still, even were those results removed from this mean, the mean is quite close to the final vote margin. So, we can see that in 1998 the final generic congressional ballot polls were pretty accurate.

In 2000, once again there were six generic congressional ballot polls in the field entirely during the final week of the campaign: Gallup, Voter.com, CBS, Pew, Fox and Hotline. Those polls went as follows:

      Rep       Dem
Voter      42       44
Gallup      45       49
Pew      42       48
CBS      38       41
Fox      42       45
Hot      42       39
Mean      41.8       44.3
Actual      48.3       47.9

(Source for final vote totals)  Here we have our first noticeable discrepancy. The final generic ballots showed Democrats with a slight, 2.5% edge over Republicans. However, Republicans won the final vote total by 0.4%. With candidates from other parties receiving 3.8% of the vote (their highest total in recent memory), this means that Republicans scooped up 65% of the undecided vote. However, considering the course of the 2000 campaign at the Presidential level, where the final week showed a strong Gore surge, Republicans winning nearly 2/3s of the undecided vote seems difficult to believe.

The 2000 election thus casts at least some doubt on the accuracy of generic congressional ballot polls. However, it should be noted that even though these polls were off, they weren't that far off. In fact, the final margin in the vote differs from the final margin in the polls by only 2.9%. Further, generic ballots in 2000 had significantly more undecideds than any other election cycle I am examining here. Whereas the other three cycles averaged 7.7% total undecideds, these polls average a whopping 13.8%. Still further, these polls did show Democrats doing better than polls in 1998, 2002 or 2004 showed. In 2000, Democrats in Congress did in fact do better than in any other cycle examined here, both in terms of seats won and percentage of votes received. So, the polls certainly were not a perfect predictor of results, but they weren't useless either. These results were reasonably close.

For those who missed them the first time, here are the results for 2002 and 2004:

Overall, given all of this data, what are we looking at here? I think we are looking at a very reasonable conclusion that congressional generic ballot polls are a pretty good predictor of national congressional vote totals in the House of Representatives. It is important to note, of course, that any single poll may be noticeably distant from the final results,. The best way to gauge the final vote totals is thus to look at the preponderance of polling data, rather than the results of any single public survey.

More to our purpose, what sort of average margin would Democrats need in generic congressional ballot polls in order to be more or less assured of making significant gains in 2006? I think that number is a lead of seven points. Seven would be beyond the margin of error for most, if not all, polls. Also, in 1994, Republicans defeated Democrats by 7.0% in the national vote totals in the House of Representatives, 52.4% to 45.4% If the average lead for Democrats in generic congressional ballots is seven points or greater, and assuming that Democrats have candidates in enough districts for their actual national vote to be representative of their projected final vote in a national public survey, I believe we can all be pretty confident in significant Democratic gains in 2006.

So where are we right now? Over the past four polls, we are looking pretty good:

        Rep      Dem
ABC        38         54
Gallup        43         49
CBS        34         43
Hotline        33         40
Mean        37.0     46.5

We are up 9.5 right now, well over the 7.0 lead that would mean we are definitely, solidly ahead. That is good news, but this far out it doesn't mean much. Things can change fast, and these days almost every election comes out very tight. Further, there is no guarantee that a lead of 7.0% or more would create a Democratic tidal wave similar to the one Republicans scored in 1994. Republican controlled state legislatures have used disturbingly accurate and efficient voting software to create congressional maps that can probably withstand a big Democratic victory much more than their maps in the 1990's could. Still, Tom Schaller tells us that a Democratic victory of 7.0 or greater would leave at least 59 Republican controlled seats vulnerable.

So, keep watching those generic ballot polls. As long as Democrats maintain a lead of 7.0 or greater, feel free to dream about a huge 2006.