My pet peeve about the polling error debate

Key Points:

• FiveThirtyEight’s forecasting models have accounted for systematic polling errors since their inception, which helped assign Trump a roughly 30% chance of winning in 2016, contrasting with competitors who gave him only a 1% chance.
• Unlike outdated models that treated races as independent coin flips, FiveThirtyEight’s FLIPR model incorporates correlated errors across races, resulting in a more moderate mid-70s percent probability for Democrats winning the Senate, compared to a 91% chance under an independent-race assumption.
• While FiveThirtyEight’s Senate odds for Democrats are slightly higher than some consensus forecasts, most competitors’ models are broadly reasonable, reflecting a convergence in election forecasting post-2016.
• Pollsters are aware of challenges like Republican voters’ lower response rates due to social trust and education factors, and employ various techniques such as demographic matching, anchoring to past votes, and likely voter modeling to mitigate these biases.
• Criticism that pollsters have ignored these issues since 2014 overlooks the significant adjustments and improvements made in polling methods following recent elections and evolving political dynamics.