How much can we trust rolling poll numbers — especially with roll cycles of months rather than days? The battleground series from Morning Consult might be the most audacious polling of this cycle, but its broad scope might make it the least useful in the final days of the midterms. The top-line results for the generic ballot don’t look good for Republicans, but we’re also looking at some old data … in fact, a lot of it.

The good new for Democrats: The topline results shows them up by seven points in toss-up districts. Furthermore, Republicans are only up three in GOP-leaning districts:

At this point, though, the methodology becomes critical context. Morning Consult used a massive amount of respondents to build this mega-survey, but it takes time to compile 388,721 responses, which is the sample size for its October report. How much time did it take? Three months, which has been the rolling window for MC’s midterm generic polling all along, as the methodology for the last four reports shows, emphases mine:

On a daily basis, Morning Consult surveys registered voters across all 50 states on their elected officials, voting preferences, and more. The results for the 2018 Midterm Wave Watcher were generated by compiling those daily surveys into three month rolling averages for each data set. For example, the most recent data set (July) comes from the results of surveys taken between May 1, 2018 to July 31, 2018. …

-July: Surveys with 352,836 registered voters from May 1, 2018 to July 30, 2018

-August: Surveys with 352,469 registered voters from June 1, 2018 to August 27, 2018

-September: Surveys with 361,630 registered voters from July 1, 2018 to September 25, 2018

-October: Surveys with 388,721 registered voters from August 1, 2018 to October 30, 2018

It’s a fascinating model for public-opinion polling, but it seems particularly unsuited for election modeling. Assuming that the data came into Morning Consult on a fairly consistent basis, roughly a third of the results reflected in today’s report on October contains response data from before Labor Day, when voter engagement was much lower than today. Two-thirds of the data was collected before the second Brett Kavanaugh hearing, after which Republicans experienced a burst of enthusiasm among its voters. Three-quarters of the data came before migrant caravans, Donald Trump’s remarks about birthright citizenship, and other issues that may — or may not — impact voter turnout for Tuesday.

This model seems better suited to track the evolution of public opinion on sustained issues — immigration, maybe, or budgets, economic policies, and other policy arenas. In that sense, some of today’s report actually does matter in setting the context of the election:

Even before the migrant caravans, security issues topped the list for Republican voters, but the economy wasn’t far behind. The economy topped the list for all other partisan demographics and in the overall sample as well, as is usually the case in American elections. Republicans should have been emphasizing this point all along. Among Democrats, health care is tied for the #1 spot but it seriously lags behind the economy among independents. Democrats, however, have been hammering on health care while Republicans prioritized security issues, essentially talking amongst themselves rather than to the whole country.

It’s also interesting to see what doesn’t register as a priority. Gun-control doesn’t show up, and neither does immigration on its own, although by now it’s definitely part of the ‘security’ argument. “Women’s issues” only gets cited as a top priority in the election by eight percent of Democrats. Energy issues barely show up, and the budget and government spending don’t show up at all. All of those protests over gun control and “believe the women” may have seriously missed the mark, as may have that strange “grab them by the ballot” Democratic ad that emerged on Thursday.

I’d love to see Morning Consult refine this polling model for issues like these. As predictive electoral polling, though, you’re better off sticking with series with much narrower scope and focused reporting … even if the news isn’t really all that much better from them.