Photo-Illustration: Intelligencer; Photos: Getty Images
What’s really misleading about extremely close election contests is that if you stare at the numbers long and hard enough, they grow in stature and meaning, because otherwise you’re stuck with an imponderable situation that leaves little to say with any degree of confidence. That’s where we are in polls about the Kamala Harris–Donald Trump contest.
For the first time in this cycle, really, we are seeing divergences in the polling averages based on how and whether those administering it adjust for the mix of pollsters releasing public data. RealClearPolitics, which accepts virtually all polling data as equally legitimate and conducts a simple arithmetical average, shows Kamala Harris’s national lead now dropping to 0.2 percent (48.7 to 48.5 percent). With the same available data, FiveThirtyEight, which weighs polls for quality and partisan bias, shows Harris leading nationally by 1.7 percent (48.1 to 46.4 percent). Nate Silver, with his own weighting system, has Harris up by 1.3 percent nationally, though in an interesting wrinkle, he actually projects her winning the national popular vote by 1.9 percent (based on extrapolation from state polls). But with this relatively small variation in the national vote, the three sources of polling averages show big variation in the Electoral College outcome. RCP has Trump leading in all seven battleground states and harvesting 312 electoral votes. FiveThirtyEight has Trump leading in four battleground states (Arizona by 1.8 percent; Georgia by 1.5 percent; North Carolina by 1.2 percent; and Pennsylvania by 0.3 percent), and Harris leading in three (Michigan by 0.6 percent, Nevada by 0.1 percent, and Wisconsin by 0.2 percent), which would translate to 281 electoral votes for Trump. Silver has the same battleground-state results with slightly different averages.
But you can find credible polls diverging from the averages in all seven states. Three recent Arizona polls (Marist, Bloomberg–Morning Consult, and AtlasIntel) have Harris and Trump tied there. Marist and TIPP–American Greatness have the two candidates tied in Georgia as well. SurveyUSA and AtlasIntel show Harris up by a point in North Carolina. In Pennsylvania, Harris leads by two points, per Bloomberg–Morning Consult, or by four points, per Franklin and Marshall (at least among registered voters; Trump leads by one among likely voters). Meanwhile, Trump is up in Michigan by two points if you believe Trafalgar, and by three points according to AtlasIntel. In Nevada, Trump’s up a point, per Redfield and Wilton. Finally, in Wisconsin, Trump leads by a point, according to Emerson.
Election forecasts are all very close, with Trump leading slightly. FiveThirtyEight and Decision Desk HQ give Trump a 52 percent chance of winning; the Economist gives him a 53 percent chance, and Nate Silver raises that to 54 percent. These are probabilities, to be clear, not projected margins of victory; it means that in massive numbers of simulations of the results, Trump wins slightly more often than Harris. The reality is that the winner and loser will be determined by the direction of polling error in each battleground state, and we have no way of knowing whether such errors will favor Harris or Trump. And more generally, the winner will be the candidate with the most effective voters-mobilization strategy (with Trump relying on an unusual and outsourced effort to reach low-propensity voters and Harris conducting a more traditional get-out-the-vote operation) and closing “argument” (with both candidates getting more personal, but with Trump counting heavily on anti-transgender ads and Harris stressing the former president’s dangerous unreliability and narcissism).
There’s time enough for a last-minute shift in the polls, but right now if anyone uses polls to predict the winner, the proper question should be this: “Which ones?” The averages continue to make it too close to call.
Source link
