Why have the polls and those reporting them underrated Trump again? - Poynter

Postelection media attention has turned toward covering Donald Trump and what lies in store under a second Trump presidency, yet an important journalistic question still looms: Did pollsters and their reporters miss a significant aspect of reality again? Apologists claim the polls didn’t fare too badly. They accurately identified seven critical battleground states; after factoring for margin of error, results for each fell within what most had predicted for those states. That was significantly better than 2020 when polls predicted Joe Biden would win but grossly exaggerated by how much. One of three expert political scientists I spoke with said, ‘it all depends what you mean by close.” No doubt about it: the polls’ performance wasn’t good. While votes still need to be counted, Donald Trump appears to have won all seven battleground states — as well as surprising everyone by winning both popular votes and all seven battleground states! That hardly sounds like the tight race that polls and campaign reporting predicted would play out, nor did it last days after polling closed; rather it happened within hours — hence why we call it election night. “There’s no question the polling and reporting was inadequate,” Josh Clinton, an assistant professor of political science from Vanderbilt University who chaired their self-study group of pollsters said during our phone conversation. They may have come closer than in 2020; yet were still off by an estimated 3-4% margin.” No consolation prize in politics exists: What counts is how much of that failure was measured accurately. Trump and his followers avoid this issue altogether while Clinton and other specialists see something entirely different underneath. Given the difficulty associated with gathering an accurate survey panel in today’s media-saturated world, polling firms employ numerous modeling adjustments in an attempt to construct as representative an overview as they possibly can; still only 1-2% of their target population actually sign up as participants in any one polling round. Michael Bailey of Georgetown University published this year an insightful book on polling systems which shows these adjustments produce more reliable outcomes than traditional methods would. Realistically, however, with so many new dimensions added to the mix come additional ways of missing the mark. I asked Bailey whether journalistic coverage despite adding qualifiers about margin of error may foster a false sense of exactitude in viewers. Though understanding of the difficulty associated with explaining math to non-specialist audiences, he insisted that poll coverage must also take some of the blame for any disparate effects seen between congressional districts. Bailey proposes taking what may seem a radical approach when making election predictions: do away with polls and margin of error terminology altogether. He suggested a “data-driven model” would better represent findings derived from new best practices, preferring terms like “prone to some errors” over the term “margin of error.” (Margin of error can best be understood when applied mathematically to large random samples – say 1,000 in total – closely aligned to what’s being investigated (which election polls no longer do.). As I expected, identifying all correctable 2024 flaws will take months or years of analysis; I was nonetheless delighted to discover solid insights had already been taken on potential danger before Election Day. In late October, The New York Times, Scientific American, and Wall Street Journal all published versions of that story with The Wall Street Journal’s being headlined – Will Pollsters Be Wrong Again in 2024?” Nate Cohn, The Times chief political analyst, highlighted a dispiriting quote from Vanderbilt professor Clinton’s study for the American Association for Public Opinion Research: “Identifying conclusively why polls overstated Democratic-Republican margin relative to certified vote appears impossible with current available data.” Polls offer some additional opportunities for mischief when combined with campaign coverage conventions. David Karpf, a political scientist from Georgetown University told me reporters and editors are increasingly using polls as fodder for discussion in election coverage coverage. Each new poll offers a daily marker. Furthermore, portraying small changes as major advances – like Kamala Harris’ launch and post-convention momentum – makes for better copy than simply saying the poll consensus remained relatively constant over time; which was indeed what occurred. Karpf, who specializes in the internet’s impact on political information rather than polling, suggested that 2028 might bring challenges that we have yet to consider. “Voter suppression or deportations could increase significantly and pose measurement challenges; but, at its heart, election night brought other losers and winners in terms of forecast games.” Allan Lichtman was among those taking a hard hit from this election cycle, due to his widely publicised system for selecting presidential winners based on 13 keys like whether an incumbent candidate is involved in scandal or whether there is conflict raging across the globe. His system now relies heavily on popular perception, rather than reality. Lichtman had been accurate in 9 of 10 presidential contests he forecast prior to 2020; but, not this one. With some humility, Lichtman told USA Today he felt embarrassed over his miscalculation (predicting an easy Harris victory), but would look into revamping his system accordingly. As I reported earlier this week, The New York Times “Needle,” an interactive real-time tracking of election outcomes across states and electoral votes was one of many triumphant examples of success this election cycle. Even as tech workers went on strike it appeared likely that its functioning would remain undisrupted by any means necessary; that question however has since been answered! As planned, that process went off smoothly and The Needle provided an easily understood graphic display of increasing certainty regarding Donald Trump’s victory. By my bedtime of 11:30 p.m. we were at 90%. Good enough for me. Cautious decision desks made their calls later – some as early as 1:30 am or as late as 5:30. Now, polling establishment must undertake an examination into 2024 with an aim of ironing out any remaining wrinkles by 2028. Clinton recently informed me of a successor study group to his, already gathering data before election night – “I won’t be involved,” he noted, offering his best wishes as they began gathering their data collection efforts. “I wish them good luck!”