The past month has been a wild ride for voters keeping a close eye on generic ballot polling. By most measures, Democrats have experienced a 3-6% decline in their advantage in generic ballot polling. This has caused quite a bit of alarm among Democratic voters, and rightly so — if there really were a 6 point drop in our expectations of the 2018 U.S. House vote they would be in a world of trouble. As I will show you, that is not the case.
Nonetheless. election observers are alarmed. In slightly fewer words, my inbox has been filled with the following message over the past week:
Subject: POLLS SHOWS DEMOCRATS DOWN?
🚨"SHOULD WE PANIC?"🚨
Democratic Reader 1234
My answer has been, in short: No. Democrats ought not to fret over the recent slide in their advantage in generic ballot polling, though there is room for Republicans to breath an ever-so-slight sigh of relief. The polling average — at least, a well-adjusted slightly longer-term average, like mine — has shown Democrats fall from a 10-point lead in the middle of January to a 7-point lead today. If combined with shifting expectations for movement in the polls from now to Election Day (which have become more favorable to Democrats over the past 3 weeks), their slightly decreasing margin in polls has produced almost negligible movement in probabilistic estimates and a decline of just 1-2% since the New Year in projected Election Day vote margin.
Over the past week, when the alarm bells have really started to ring, movement has been even smaller — just 0.5-1%.
Note that this position, roughly defined as “stop reacting to polls so violently and evaluate projected outcomes in November, not polls today” is exactly the stance I was taking in December when many Democratic commentators were overhyping their own gains in the polls.
While a 3 point slide in the polling average is certainly substantial, it is nowhere near as large as the 6 or 7% dis for which some pundits and political prognosticators have been arguing. Most likely, this is due to my model taking a more conservative approach to averaging polls this early. Justified by the data, my average looks at all polls over the past 29 days, rather than the 1 or 2 weeks some other aggregators have been using. In past cycles, the latter has produced wild swings in our measurement of the environment that hold almost no day-to-day predictability, whereas the former has an ultra-strong (0.9 correlation) to final polls.
But the real problem is that the prevailing examinations of the November national environment have been relying on short-term (exaggerated) movement in a polling average taken forty weeks before election day. What readers really need to know is what polls means for the national environment on November 6, not February 5. That is why my forecasting model exists: to tell us how changes in data today should adjust our expectations for an event 9 months from now. For this scenario, that entails measuring changes in previous midterm election cycles from this point until Election Day. My model estimates that number to be a 3.3% gain for the opposition party from this far out. A 6.6% Democratic margin today plus a 3.3% expected gain equals 9.9% this coming November. That’s down from an 11-12% expected margin in early January.
In short, the math is clear: the congressional polling average today is likely underestimating the Democratic party. I tweeted the following graphic yesterday to illustrate what I mean:
This figure compares the difference between this year’s Democratic margin in the polls with an average of the opposition party’s polled margin at each day in the campaign in midterms since 1946. As you can see, not only is there a clearly discernible downward trend from 273 to 0 days out, but the recent GOP gain in the polls is almost entirely predictable. Not only has my model been projecting the national environment when it actually matters, but it has also been adjusting the expected Democratic gain in the polls over the past three weeks to control for the bounce we expect to see (and are seeing!) based on history.
These adjusted projections have cumulated to produce relatively stable probabilistic expectations of the race over the past month.
In sum, the current dip in the polls is significant for Democrats but is (A) smaller than most are reporting and (B) almost 100% predictable from historical patterns in congressional polling. That’s why my model at TheCrosstab.com has been so stable when polls have been so bouncy. Not only is a longer-term average more predictable, the data say, but forecasts out to election day have almost entirely counteracted movement over the past month. Since January 15, the probability the Democrats win back the House has decreased from a 63% high to a 59% low. A modest adjustment if you’ve been listening to punditry and mainstream prognostication.
I urge you to view the national House polls within the paradigm I laid out above. Take a broad average and actually ask yourself “what does this mean for November?” If you do this, you’ll end the day with a much clearer sense of where the race stands. Ultimately that will prepare you better for what will surely be a combative, raucous, and turbulent campaign season this fall.