Generic US House polling: What does it tell us?

Charles, at Political Arithmetik, has one of his many must-read, must-view posts–this time, about trends in 2006 and past cycles in polling for US House.

As Charles notes, Democrats tend to perform much better in this polling than in the actual voting, where real candidates–including known incumbents–are running, rather than just the “vote for party” option that the polling simulates. Even so:

In no cycle since 1994 has the Democratic lead approached 10%, let alone exceeded it. So by the standard of “relative lead” in the generic ballot, 2006 reflects extremely strong pro-Democratic forces.

The current trend of several poll shows an average lead for the Democrats of almost fifteen percentage points! And it has been trending mostly upward for some time.

Of course, as Charles also notes, it is not only fraught with peril to go from generic polling to estimating the actual aggregate congressional votes, but even more peril lies in estimating seat swings based on (expected or even real) aggregate votes. But such peril never stops F&V!

I posted some graphs of votes and seats trends over time here three months ago.* If Democrats really had a 15-point lead in national votes, that would mean around 54% of the vote (allowing for the recently typical 5% or so for third parties). Extrapolating from the graph (see link below), that would put the Democrats anywhere from about 240 seats (at the lower responsiveness of recent elections) to about 265 (using the trend over the entire post-1945 period. which is probably less likely to be relevant). (Those are ‘eyeballed’ point estimates off the graph; I think it is pointless to talk about margins of error, as long as we all understand that these are just rough estimates of what would happen if the votes translated into seats in 2006 more or less as they have in the past–whether the very recent past, or the past six decades.)

Of course, we should not assume the polling will accurately translate into votes. As Charles shows, it does not, or else Democrats would have won each of the last six House elections. (Even 1994, Charles shows, the Democratic party had recovered from its earlier generic-polling deficit by election day!). (They did win 1996, in votes, though not in seats.)

The following list shows, for each year since 1994, the final polling difference (always D-R), the actual difference in percentages of the total vote, and the actual difference in the two-party vote. Data on the polling are from Political Arithmetik. Actual votes data are from my files (based on Clerk of the House reports).

The last number, in parentheses, is simply the difference between the final two-party vote and the final generic polling result. (I much prefer to use the shares of the total vote, but pollsters, in their infinite wisdom–and for that matter, most of the country-specialist political scientists–ply their craft as though there were literally only two parties.)

    1994, +3.35, -6.79, -7.05 (+10.4)
    1996, +5.45, +0.30, +0.32 (+5.13)
    1998, +8.03, -0.89, -0.93 (+8.96)
    2000, +5.26, -0.34, -0.36 (+5.62)
    2002, +1.49, -4.62, -4.88 (+6.37)
    2004, -1.07, -2.76, -2.88 (+1.81)

So, on average, the final poll has overstated the Democratic party vote advantage by 6.38 percentage points! (Note that the one year when the polls showed Republicans ahead was also the year with the smallest error of estimate, while the year the Republicans took control of the House from the Democrats was the year with, by far, the greatest error from poll to actual votes.)

If we take Charles’s current average of generic polls (+14.65 Dem) and subtract the average difference between final polling and actual votes (6.38), we would have an estimated Democratic advantage of “only” +8.27. That would be roughly 51.6% votes for the Democrat (again allowing around 5% “other”). Now, doing our perilous extrapolations, we get a range of anything from around 225 seats (a bare majority, assuming the votes-to-seats conversion will be as nonresponsive as it has been in the last decade) to around 245.

Of course, if the generic trend continues upward for Democrats, it would become quite unlikely that they would not be the majority party in the next House. On the other hand, the relative nonresponsiveness of the current House electoral system to votes swings would suggest that even a small recovery for the Republicans in the generic polling by early November could keep the current majority intact.

No, we can’t make any predictions, unfortunately. It looks bad for the Republicans. You already knew that. But Democrats could fail to capitalize on their current good polling. You probably aleady knew that, too. What I hope this discussion shows is that a failure to “capitalize” would not necessarily be from bad tactics (as the media will tell us), but from underlying structural features of the US House electoral system. To review, those are (not necessarily in order of importance):

    1. Voters vote for candidates, not parties.
    2. The party polling tends to overstate voter preferences for living, breathing Democrats, and has done so for many years (even in 1994).
    3. The process of allocating House seats is not as responsive to vote shifts as it once was.

* “Votes and seats in the US House,” and

“Presidential approval and midterm vote and seat loss.”

0 thoughts on “Generic US House polling: What does it tell us?

  1. I wonder if you don’t have your final bullet points upside down– it seems to me that the electoral map-drawing is more important than any other factor.

    In fact, you might say that the polls have it right and the elections understate Democratic support. Vote totals in the least competitive districts tend to be much lower than in competitive districts, so the party that wins more of its seats in highly safe race will end up with a lower share of the total vote.

    As for the “vote for the man, not the party” principle: does this refer to the way ballots are made up, or to voter behavior? Because if it’s the latter, I think there’s room for some skepticism. Incumbents who get out of sync with their districts– like Connie Morella, for example– quickly find themselves in trouble.

  2. JG, excellent point on turnout varying with competitiveness of districts, though I have no idea the extent to which that might affect the translation of the latent party support (as expressed in polls) into actual votes for a party.

    On whether voters vote for party or candidate, of course party remains the buggest predictor of vote. The point I was making is simply the (not very controversial) claim that candidate quality matters. Thus many voters who might generically like one party (and tell pollsters thus) will nonetheless vote for a candidate of the other (for his or her record of constituency service, or personal qualities, or whatever).

    As for the bullet points, they were not meant to be in order of importance. I have edited the post to indicate that.

  3. Pingback: Fruits and Votes

  4. Pingback: Fruits and Votes

  5. Pingback: Fruits and Votes

  6. Pingback: Fruits and Votes

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s