When The New York Times’s Upshot announced it was “live polling” this year’s hottest races, I wasn’t a fan.
But my worst fears weren’t realized and I learned some things watching the interviews roll in.
Not really. Twenty-three of the 27 polls where Republicans ended up ahead began with a Democratic lead.
“So what?” you ask.
Random selection is the fundamental principle enabling the relatively small samples found in polls to reflect much larger populations. In this context, in slightly oversimplified form, randomness means that every potential voter has an equal probability of being contacted.
Completing polls too quickly can violate the requirement for random selection. They are not a random sample of the electorate, but rather a sample of easy-to-reach voters.
The Upshot polls suggest that easy-to-reach respondents tend to be more Democratic on the average. Thus, quickie polls may not be very accurate. Calling back potential respondents over and over again to get hard-to-reach folks is a vital element of the process.
We find in our own polling that the people we reach on the first attempt are often different from those we reach on the third or 10th attempt.
On a more substantive level, Upshot provides 54 polls that can be compared with past voting behavior in congressional districts.
Let’s go back a few months when Democrats were thrilled about the fact that special elections were producing margins for Democrats substantially larger than Hillary Clinton garnered in those same districts.
Indeed, on average, Daily Kos calculated the margins in special elections were 10 points greater than in the 2016 presidential contest, a fact trumpeted as evidence of the forthcoming Democratic wave.
We can do the same arithmetic with the Upshot polls.
In these surveys, on average, Democrats are doing 4.5 points better than Clinton did.
It’s meaningfully less optimistic for Democrats than the special election margins for at least two reasons.
Special elections do not feature incumbents. Most of the fall contests do have an incumbent, and incumbents have advantages.
Second, special election turnout tends to be lower than turnout for the November contests.
Democratic performance in the specials was correlated with turnout—the higher the turnout, the worse the Democrat tended to do relative to past presidential performance.
In addition, as more a parlor game than a legitimate forecast, we can see what would happen to the House if every Democrats’ margin was 4.5 points better than Hillary Clinton’s (matching the average difference in the Upshot polls).
It’s an assumption certain to prove false. The change won’t be uniform across districts. While on average, Democratic congressional candidates are besting Clinton’s margin by 4.5 points, the range runs from 19 points worse to 44 points better.
But looking at the average does offer an interesting perspective, which suggests Democrats could pick up 32 districts for a total of 226 seats — and the majority.
While this is not a real prediction, it comes close to other, much more complex projections. CNN’s Harry Enten projects an identical 226 seats for Democrats; G. Elliot Morris’s median prediction is 230 seats; Optimus forecasts 231 and Nate Silver’s 538.com predicts 234 Democratic seats.
Academic models are suggesting gains of 27 to 44 seats, while political prognosticators like David Wasserman of the Cook Political Report and Nathan Gonzales of Inside Elections project Democratic gains of 25-35 seats.
All of which is to say: a. I learned something from Upshot polls, and b. quite different methods are converging on a reasonable expectation of Democratic control of the House with a pickup of over 30 seats.
Mellman is president of The Mellman Group and has helped elect 30 U.S. senators, 12 governors and dozens of House members. Mellman served as pollster to Senate Democratic leaders for over 20 years and as president of the American Association of Political Consultants.