Let's finish on an upbeat note. A British polling firm, FocalData, uses a survey method called "MRP" — which, as everyone knows, stands for "multilevel regression and post-stratification" — but is generally known as a "mega poll" because of its massive number of respondents. In analyzing the U.S. presidential race over the past month, FocalData interviewed more than 31,000 voters — roughly 30 times the respondents most American pollsters interview. The firm revealed its findings exclusively to Politico.
I'll not bury the lead. The pollster's chief research officer, James Kanagasooriam, told Politico that throughout most of the month the MRP model showed "a Trump win." Once down to the wire, however, the "update has it nudged Democrat," said Kanagasooriam, who added, now "we are 'lean Democrat [Harris].'"
Showed a Trump win did not mean predicting a Trump win any more than leaning Democrat means what Politico writes: that the poll is "tentatively predicting a Kamala Harris win." That may sound pleasant but it's wrong; polls don't predict, they survey. And, before anyone pops a cork, the research chief further added that Harris' lead is "only by the barest of margins."
Still, a late-October shift toward Harris that shows her leading by any margin, rather than merely tying with Trump, is considerably encouraging. The MRP model also incorporates FocalData's "large-scale online swing state polling.... It takes a massive nationwide sample and then uses demographic data to estimate seat [in parliamentary races] or state-level outcomes."
The pollster found that "Harris is likely to take Michigan" by almost 5 points, leads in Nevada by 2 points and has "a slight edge in Pennsylvania and Wisconsin." In Arizona, MRP has Harris ahead and state polling has Trump ahead, but only by 1 point in each case. In Georgia and North Carolina, Trump's lead is so slim it doesn't much deserve to be called a lead.
I want to how the state breakdowns mainly because of Harris' better showing in Pennsylvania and Wisconsin's online state polling.
The MRP model has had mixed results in parliamentary polling, but its proponents have higher hopes for its use in the U.S. The reason: In Britain the model faces the challenge of surveying 650 parliamentary seats, while the U.S. has only 50 states. One other interesting aspect of FocalDats versus America's surveys is that analysts suspect that some domestic pollsters are "herding," that is, bunching their results to fit the overall polling consensus so that their findings are not what could be embarrassing outliers, especially just before Election Day.
Explains Kanagasooriam: "We, along with others, have seen evidence of herding among pollsters at this election — likely out of fear of underestimating Trump for a third cycle. The result is the current public polling is … displaying much too high levels of confidence in a close race when we are a normal-sized polling error away from a clear victory for either candidate."