An example of how I would read Congressional Polls

[Harry Hayfield]

All data from Polling Report.com and Wikipedia:

Generic Polls on Congress in July 2006

Gallup/USA Today Jul 28-30: GOP 40% Dem 51% Others / Unsure 8%
CBS/New York Times Jul 21-25: GOP 35% Dem 45% Others / Unsure 20%
Diageo/Hotline Jul 20-23: GOP 32% Dem 48% Others / Unsure 20%
Fox/Opinion Dynamics Jul 11-12: GOP 34% Dem 42% Others / Unsure 25%
AP/Ipsos Jul 10-12: GOP 40% Dem 51% Others / Unsure 9%
Galup Jul 6-9: GOP 41% Dem 51% Others / Unsure 9%

Based on my experience of polls here in the UK, I would ignore the following polls:

CBS/New York Times Jul 21-25: GOP 35% Dem 45% Others / Unsure 20%
Diageo/Hotline Jul 20-23: GOP 32% Dem 48% Others / Unsure 20%
Fox/Opinion Dynamics Jul 11-12: GOP 34% Dem 42% Others / Unsure 25%

and therefore concentrate on the remainder which suggest an average of:

GOP 41%, Dem 51% Others / Unsure 9%

Now we know that the Others won't poll 9%, so let's look at Congress 2004:

GOP 49% Dem 46% Others 5%

Based on that let's assume that the Others stay on 5% and that the 4% who are unsure won't vote, that gives us the following change on 2004:

GOP -8% Dem +5% Others +3% which equals a swing of 6.5% to Dem.

Is that a likely interpreation, or do I need to start from scratch?

[I spent the winter writing songs about getting better]

The problem with this is the US is not like the UK where you can predict a uniform swing across the board, and because of gerrymandering (we need to have our districts drawn like you do), knowing the seats still can make the actual change in seats tough to predict.

Also I can guarantee there won't be a 3 point gain in the votes cast for third parties.

Still that doesn't sound like a bad prediction in the raw votes, I think 2004 was GOP +3%, so that'd mean about Dems +3.5%, which is reasonable. The problem is even with that it's still tough to predict how the individual seats go.

[Harry Hayfield]

So does the following sound a reasonable idea instead:

Average the Dem / GOP lead and compare with lead in 2004, then halve the swing and reduce by another half if the district has an incumbent?

[7,052,770]

There's a lot of anamolies though.  For example, Gene Taylor from MS-4 is a Democrat.  He carried his district with over 60% of the vote in 2004, but so did Bush.  People who are normally Republicans, those who vote for Bush and Lott and Barbour also vote for Gene Taylor.  Such anamolous districts exist across the country and you can't really factor uniform swings into them.

[jerusalemcar5]

So does the following sound a reasonable idea instead:

Average the Dem / GOP lead and compare with lead in 2004, then halve the swing and reduce by another half if the district has an incumbent?

Personally, I would ignore national polls.  They really are meaningless.  It matters on a district by district basis.  These districts are biased as well (more than 350 seats are in the "safe" category) so though many people want Democrats, their voices are usually slashed through and split up.  I hate how we draw House districts .