G. Elliott Morris tackles a pervasive anxiety in political forecasting: the fear that polling data is systematically rigged against one party. While the prevailing narrative suggests a simple correction is needed, Morris dismantles the idea that historical patterns can reliably predict future errors, arguing instead that trying to "unskew" the data often creates more chaos than clarity.
The Trap of Historical Bias
The piece begins by addressing a chart from Republican pollster Patrick Ruffini, which suggests that Senate polls have consistently overestimated Democratic support in August over the last four election cycles. Morris acknowledges the surface-level truth of this observation but immediately challenges the conclusion drawn from it. "Ruffini's argument is right in some ways and wrong in others," Morris writes. He contends that while early polls are indeed imperfect snapshots, the leap to assuming a permanent, directional bias is statistically unsound.
The core of Morris's rebuttal rests on the sheer volatility of polling error. He expands the dataset from Ruffini's four cycles back to 1998, revealing that bias changes randomly from year to year. "Like taking the under on low-probability bets, predicting polling error from previous cycles' misses works until it doesn't," he notes. This framing is crucial because it shifts the reader's understanding of polling from a flawed but predictable instrument to a noisy signal that requires sophisticated modeling rather than blunt adjustments.
Knowing what polls did in the four previous cycles narrows your expectation for the next one by essentially nothing.
Morris illustrates this with stark examples from the past. In 2014, historical data suggested polls were underestimating Democrats, yet that year's polls actually overestimated them by seven points. Applying a historical adjustment would have created a massive nine-point error in the wrong direction. Similarly, in 2016, the data suggested the polls were "clean," yet they missed by nearly six points. The author's analysis here is particularly effective because it moves beyond abstract statistics to show the tangible cost of relying on heuristics. Critics might argue that recent cycles (2018–2024) show a more consistent trend than the longer historical record, suggesting a structural shift in polling methodology. However, Morris counters that even within this recent window, the direction of bias flipped in 2022, where polls actually overstated Republican support.
The Danger of "Unskewing"
The commentary deepens as Morris explores the practical consequences of trying to force polls to fit a historical mold. He describes the process of "unskewing" as a high-risk maneuver that increases the "risk of ruin." By adjusting polls based on past errors, forecasters risk compounding their mistakes across every race simultaneously. "If you're going to put your thumb on the scale for a party, you want there to be a close to zero chance of causing a larger error than if you had left the polls alone," Morris argues.
He points to the 2022 midterm elections as a cautionary tale. Had analysts applied the lagged bias from the previous four cycles to the 2022 data, they would have shifted all polls three points to the right. This adjustment would have increased the error in every single key Senate race and incorrectly predicted Republican victories in four races that Democrats ultimately held. This specific reference to the "risk of ruin" connects to broader themes in forecasting, echoing the sentiment found in deep dives on election modeling where a single miscalculation can cascade into a total loss of credibility.
"Unskewing" also introduces high "risk of ruin."
Morris's critique of RealClearPolitics' 2022 adjustment is sharp and unambiguous. He notes that their procedure would have made their averages more biased than raw data in seven of the last ten election cycles. The argument here is not that polls are perfect, but that the cure is often worse than the disease. He suggests that the better approach is to measure uncertainty rather than predict bias, a distinction that is vital for busy readers trying to parse the noise of election season.
Why 2026 May Be Different
Looking toward the future, Morris offers a nuanced explanation for why the recent trend of August polls overestimating Democrats might not hold for 2026. He identifies two main drivers of the previous "error": the shift from sampling registered voters to likely voters, and the specific impact of the Dobbs decision in the summer of 2022. As he puts it, "Early polls are not meant to perfectly predict election outcomes. Voters' preferences can change over the campaign as they tune in to information and socialize about key races."
The author points out that in 2026, the dynamic may actually reverse, with Democrats performing slightly better in likely voter polls than in general adult samples. Furthermore, the absence of a seismic judicial event like Dobbs removes a key variable that disrupted the typical midterm cycle. This contextual depth prevents the piece from being a mere statistical exercise; it forces the reader to consider the unique political and social landscape of the upcoming election. Morris reminds us that "there is no Dobbs in 2026 disrupting that cycle, so the data points from that summer may be useless in anticipating potential poll bias."
Bottom Line
G. Elliott Morris delivers a compelling case against the seductive simplicity of adjusting polls based on past performance. His strongest argument is the demonstration that historical bias is not a reliable predictor, making any attempt to "fix" the data a gamble that often increases error rather than reducing it. The piece's biggest vulnerability lies in its reliance on historical averages that may not account for rapid, structural changes in how voters express themselves, but Morris effectively mitigates this by emphasizing the need to model uncertainty rather than force a correction. For readers navigating the coming election cycle, the takeaway is clear: trust the raw data's uncertainty, not a forecaster's guess about its bias.
The historical evidence suggests this is a fool's errand.