Something peculiar is taking place in the world of political polling, and it seems to be underpinned by one key forking methodological decision. In one reality, Kamala Harris appears to be struggling, barely fending off Donald J. Trump in the national popular vote, while barely maintaining an advantage in Northern battlegrounds. This scenario mimics the dynamics of the 2022 midterm elections. In a parallel reality, Ms. Harris somehow has a lead in the national vote, however, the battlegrounds are excruciatingly close, essentially mirroring the 2020 elections’ tension.
There’s one main methodological maneuver at the heart of these disparate universes – the use of ‘weighting on recalled vote’. This technique attempts to adjust based on respondents’ recollection of their voting behavior during the previous election. Then, pollsters use weighting, a statistical technique that reflects each demographic group proportionately to the actual population. The survey approach amplifies or diminishes the influence of Biden or Trump voters in accordance with the last election’s outcome.
However, such an approach was historically viewed as an error, a false step avoided by pollsters for reasons we will delve into further. It’s interesting to note, though, that this technique is being more widely employed today, likely in an attempt to make up for the glaring inaccuracies in polling seen in 2016 and 2020 specifically regarding Trump’s support. This strange turn of events illuminates yet another rift in the ranks of pollsters during this election.
What’s intriguing is that in the past month, about two-thirds of polls opted for weighting by recalled vote. An apparent outcome of this methodology is an eerie resemblance of the poll results to the 2020 election results. Pollsters who avoid this trend, like those conducting The New York Times/Siena College surveys, more often present changes from the 2020 elections, tending to align their results closer to the 2022 midterm election trends.
The variances seen across these types of polls, albeit seemingly minor, ultimately culminate into contrasting narratives. Polls shying away from recalled vote weighting document the gradual fading of Trump’s advantage in the Electoral College in comparison to the popular vote over the last four years. The recalled vote weighted polls, however, seem stuck in a time loop replicating the 2020 election, almost verbatim.
It’s surprising to see that a significant number of respondents cannot accurately recollect their voting behavior. Interestingly, they also have a tendency to recall voting in favor of the election victor, and occasionally recount voting instances when they evidently did not participate. The persistence of this pattern, extending from the dawn of recall vote studies to date, profoundly skews the accuracy of this polling methodology.
Notably, the application of recall vote weighting predictably boosts support for the party that lost the previous election. Despite the multitude of challenges poll data analysis faces, under-representation of the erstwhile losing party does not appear to rank among them. Consequently, recall vote weighting seems to have muddled polling accuracy in every election dating back to 2004, based on a re-analysis of a repository of 70 polls.
Despite this unfavorable track record, encouraging the steep rise in the usage of recall-vote weighting is the idea that it has become more dependable than before. Whether this claim holds water remains debatable. Recent data visually seems to align more with actual election outcomes compared to previous cycles, but there may be more going on below the surface.
It’s possible that due to the surge in political participation and polarization, respondents are better at remembering who they voted for. Alternatively, online data collection could be reducing the typical ‘winner bias’ seen in phone surveys. In this peculiar election, the previous loser, Trump, is running again causing additional complexities when measuring recall-vote accuracy.
Another factor in favor of weighting by recalled vote is the usage of panel-based surveys. Respondents in such panels may have declared their voting preference in 2020, so the reliance on memory is minimized, providing potentially more accurate results. However, as with all things, the devil resides in the details. What does one do about new panelists or the shifts in the electorate makeup since 2020?
Certainly, recall vote weighting can help avoid entirely unrepresentative numbers and push a survey towards results aligning with the last election. This proves particularly useful when the survey design fundamentally lacks validity, like a poll exclusively of users of a single social platform. It’s notable to point out, over 80% of online panel-based surveys utilize recall vote weighting.
An important aspect to keep in mind while discussing all of this, is the usage of recall vote weighting to account for the consistent underestimation of Trump’s popularity in polls over the last eight years. Serving as an effective method to ‘nudge’ survey outcomes to the right, recall vote weighting may be the stop gap measure pollsters use to regain lost credibility.
All toolkits have trade-offs, and in this instance, with recall vote weighting, there is the risk of missing significant shifts in the electorate since the last election. Pollsters’ distrust in their own survey results is leading to the employment of this technique. However, this ‘fix’ could potentially miss out on novel trends and dampen any outlying insights. While pollsters intend to avoid the embarrassments of past underpredictions, they are also edging towards the risk of not seeing any new patterns emerge.
Harris Losing Ground to Trump as Polling Methods Favor Republicans appeared first on Real News Now.
About Author
