Mark Blumenthal posted yesterday on "Waves and Stability" at www.pollster.com. The thrust of the post was that generic congressional polls (GCPs) seem to portend a wave election.  But, GCPs are really poor predictors of how many seats a party will pick up.  And, the relationship between a party's national vote share and the number of seats it wins has broken down post-1994.  I.e., the allocation of seats in House elections is becoming more resistant to change.  Bottom line:  in spite of the heavily Democratic GCPs, its hard to know if there will really be a wave.

I love these massive polls of House races.  Individually the numbers are somewhat interesting.  But, they get  really interesting when you play around with aggregating the numbers.  Here's my cut.

In a diary last Saturday (http://www.mydd.com/story/2006/9/23/1054 36/148) I discussed trends in generic congressional polls (GCPs).  I noted that in September, GCPs of registered voters averaged +11.3.  I then went on to discuss likely voter GCPs:  "There have been seven such polls in September.  (Note:  there is some overlap with the RV/GCPs as some polls report both RV and LV numbers.)  Here the average is +7.4.  But, things get a lot more interesting if we dig a little deeper.  The seven polls include four polls that range from +10 to +14, and three polls that range from 0 to +3.  The average of the two clusters is +11.5 and +2.0 respectively."

Conventional wisdom is that Bush's fear-mongering has tilted the scales in the battle for the House.  Dems had a slight advantage in August, but now the Repugs have a slight edge.  This view seems to be bolstered by generic congressional polling (GCP) data such as Gallup's recent 48-48 among likely voters.  Zogby and Fox have also shown a very tight race among likely voters.  Mystery Pollster has some very good analysis here on these polls.  He notes some conflicting data and points out the difficulties in constructing likely voter models. In this post, I want to expand on Mystery Pollster's analysis by looking more comprehensively at all the September GCP data to see if there is a clear trend.

I think there is and it is fairly encouraging.  But, it also tells us that it is absolutely critical to fire up the base. There have been eight GCPs (Generic Congressional Polls) of registered voters in September.  The average result is +11.3.  This is remarkably consistent with GCPs dating back to at least February.  On a rolling average basis, the GCP has fluctuated between 10 and 12 for the last seven months.  Remarkably, Rove/Bush/Cheney have not moved the needle a whit among registered voters.  Even, if we look at the last four RV/GCPs, two of which include data from 9/11 itself, the average is +10.8.

Turning now to likely voter GCPs.  There have been seven such polls in September.  (Note:  there is some overlap with the RV/GCPs as some polls report both RV and LV numbers.)  Here the average is +7.4.  But, things get a lot more interesting if we dig a little deeper.  The seven polls include four polls that range from +10 to +14, and three polls that range from 0 to +3.  The average of the two clusters is +11.5 and +2.0 respectively.

It seems pretty clear that there are two types of likely voter models in use.  One model puts the priority on voters' stated intentions.  The other model puts the priority on recent voting history.  If you focus on intentions, the Dems are up 11.  If you focus on recent voting history, which reflects the vaunted GOP GOTV machine, then Dems are up 2.  I suspect that the truth lies somewhere in between.  The opponents' GOTV machine is alive and well.  Will it work as well as it did in '02 and '04?  Probably not.  But, it will work well enough that likely voter models based primarily on intentions overstate the Dems advantage.  On the other hand, some Ds -disgusted and angered by Iraq, Katrina, etc.-- will vote in spite of sitting out recent midterms.  And, some Rs -dispirited by exploding debt and other grievances - will sit out in spite of voting in '98 and '02.  So, I feel pretty certain that models that rely heavily on voting history understate the Dems advantage.  All in all, I think +7 is a pretty good guess as to the state of play.

How predictive is this number?  I think its pretty good.  I don't have links at my disposal, but I've seen evidence that LV/GCP polls close to the election are generally very good at forecasting the actual result.  How would +7 translate?  My guess is that it would translate into a pick up of about 20 seats.  See this post http://www.mydd.com/story/2006/1/11/1644 32/998 for some supporting analysis.  (The bottom line from the analysis is that because of gerrymandering we need to win the aggregate Congressional vote by around 5% to take the House.)

So, I think we're in a pretty decent position.  But, and this is a HUGE but, it all depends on turnout.  The divergence between the +11 cluster of LV/GCPs and the +2 cluster is so large, that it is clearly showing us that IT'S ALL ABOUT TURNOUT.  And more than that, I think its all about the Dems firing up their base and executing on GOTV.  The GOP will do a good job of firing up their base.  They will do a good job on GOTV.  It won't be as intense as '02 and '04, but it'll happen.  So, its really in our hands.

The message is pretty clear:  All Iraq, all the time.  If we do that, we will absolutely take the House.

There are two views floating around on whether we can take the House.  The first view, offered most notably by Charlie Cook (see http://www.cookpolitical.com/...) is that there just aren't enough competitive races; realistically, the Dems can pick up 4 to 9 seats.  The counterargument (I believe Matt has posted on this) says this misses the forest for the trees.  If this is a "tidal wave" election analogous to 1994 (similar dynamics if not the same magnitude), races that don't look competitive now will be, and races that look competitive could go blue en masse.