Polling the Warren Factor

Photograph Source: Senate Democrats – CC BY 2.0

Contrarily, most Democratic primary polls in the last week to ten days, including the average of polls both as I calculate them daily here under the 10at10 label as well as in the RealClearPolitics (RCP) average, show a race between Warren and Biden that has tightened dramatically since the big public revelation on September 24 regarding Donald Trump, Ukraine and the Bidens, and potential impeachment proceedings. My average as of yesterday morning at ten o’clock eastern had just a 0.9% spread between Warren and Biden, while RCP showed just a 0.3% gap between the two. While RCP also shows a 2% drop for Bernie Sanders in the relevant period, my averaging shows no real change (15.8% on September 23 versus 15.9% yesterday) to Sanders’ standing over the two-week period. 10at10 did have Bernie at 17.4% last Monday, which could suggest a brief rise and then dip before and after his bout with hospitalization and heart surgery. It also could just be a noisy reflection of which polling firms were in the field and when.

Under the hood in critical crosstabs relating to race or ethnicity and age, there has also been very little change for Morning Consult. Two weeks ago, Biden led Sanders in second and Warren in third with African American voters at 39.7% to 21.3% to 11.2% respectively (n=3654), while this week’s crosstabs show 42.2% to 20.1% to 11.2% (n=3495) with the same ordering. Among respondents who identify as Hispanic, Sanders leads Biden and Warren 28.8% to 25.4% to 13.6% in today’s results (n=1919) versus two weeks ago when Sanders was at 28.3%, Biden 23.3%, and Warren 14.7% with respondents who identified as Hispanic (n=2051). Warren did advance somewhat with white respondents who do not identify as Hispanic, moving from 22.0% two weeks ago to 25.0% over the past week.

The other fault line, perhaps even more stark, in the Democratic primary is by age. Morning Consult breaks things down, with high respondent levels in each, into five categories (18-29, 30-44, 45-54, 55-64, and 65+) as charted below. The under 45-years-old versus 45-years-old plus categories all told went from 30.2% Sanders, 22.0% Biden, 16.3% Warren (n=7456) for the younger set two weeks ago to 30.6%, 22.7%, 16.7% (n=7000) in today’s release. There was also little remarkable change for voters forty-five and older overall with a 39.2% Biden, 22.5% Warren, 10.8% Sanders split two weeks ago (n=9921) and a very similar 40.0%, 23.6%, 10.4% (n=9529) ordering today

Morning Consult’s steady results (they’ve shown much less dramatic change week-to-week, month to month than other polling outfits throughout 2019) are likely attributable to consistent methodology and weighting for a very large sample sized poll effort. Morning Consult’s overall method is to contact around 5,000 Americans each day online with a long series of questions and then to report those results in various articles and outlets and to particular clients in a weekly format that narrows things down to the relevant subsampling with extremely detailed demographic reporting. The advantage to this way of polling is that each sub-category is itself a reasonably sample-sized product. Morning Consult, for instance, interviewed more people in the Spanish language (n=133) this past week than the entire subsampling of people of color for several recent, highly regarded live caller polls.

In last week’s article, I noted that there is now a significant 10% difference between online and phone only polling in Iowa for Sanders in particular. Is what we are seeing now a similar difference between, for instance, people more likely to watch less television news while more likely to respond to a poll online rather than one by phone, on the one hand, and polls that better capture those who are, on average, older, spend more time following the details of network news, and answer the phone? Unlikely. For one thing, Morning Consult keeps track of support by primary news source each week. Those too show little change, at least in comparison between September 23-29 results and previous similar Morning Consult reporting. Over the last several election cycles, furthermore, there is no clear indication that online versus phone only polls are more or less right or wrong, all things considered.

One possibility is that smaller sample sized polls, some of which change their weighting substantially week-to-week (YouGov) are herding toward a particular outcome rather than capturing a genuine shift in the electorate. At this point, we have very little way of knowing who is right until voting an caucusing begins in just under four months.

For now, there is in fact a rather stark difference, even among relatively good polls. There is no agreed upon set of standards for the best way to weight even a genuinely random sample for a Democratic primary. Reasonably respected polling has had Biden anywhere from 21%-33% in the last few weeks, Warren at anywhere from 15-28%, and Bernie Sanders from 10-22%. Simply averaging national or state polls may tell us very little at this point given how bad polling was for the 2016 Democratic primary.

For these reasons and several others, my modelling of the Democratic primary takes a more in-depth approach to modelling each district and state based on a wide range of factors including national polling, state polling, how the district or state voted in the 2016 primary, how polling fared by state in the 2016 primary, and the demographic make-up of each district, comparing it to what is known from polling, particularly more in-depth polling by age and race or ethnicity, about each of the top candidates.

That modeling continues to suggest a three-way race with last night’s update again showing each of Warren, Sanders, and Biden likely winning hundreds of delegates through Super Tuesday and pushing toward a contested convention.

Doug Johnson Hatlem writes on polling, elections data, and politics. For questions, comments, or to inquire about syndicating this weekly column for the 2020 cycle in your outlet, he can be contacted on Twitter @djjohnso (DMs open) or at djjohnso@yahoo.com (subject line #10at10 Election Column).