After an absolutely shite day at work, I haveth not the energy to go link-searching for some evidence/date to back up my claim on this one...more to be done on that front. But at any rate, here's the theory:

As has become clear from the last two presidential elections, there is a clear separation, voting preference wise, between the bigger cities (Democratic) and the more rural areas (Republican). Population is supposed to be roughly equal for every House seat; in other words, the one representative from the great state of Montana has about the same number of constituents as one of the representatives from the not-so-great state of Maryland...both around 400K or so. Not sure about other metro areas in the country, but at least here in DC, the Democratic candidates win BIG...not unusual to see 20 or 30 point margins of victory. But it wouldn't matter if the margin was a hundred points...there's only one seat up for grabs. Thus, the most accurate polling for the House outcome AS A WHOLE would involve a statistically random sampling FOR EACH SEAT rather than a nationwide poll that did not take specific Congressional districts into account.

This, of course, accounts for the following:

A) Eve of the 2002 off-Prez election: Nationwide poll shows 48-48 preferences

B) Actual result: Republicans gain 6.

My guess or rough rule of thumb, based on the above, is that each percentage point in the nationwide poll represents one seat -- or, that if a poll showed a six point lead for the Donkeys (let's say, 50-44), that in reality the outcome would be a net zero for both sides -- stasis.

That's one of the things I'm basing my big RGain bet on.

More reasons to follow...