Polling Accuracy

By Jim Ellis — Tuesday, Nov. 19, 2024

Polling

There has been a great deal of controversial discussion about the 2024 political polling. Many continue to raise questions about the major polling firms’ cumulative accuracy rating, but is the criticism fair?

The perception that the polling industry missed President-Elect Donald Trump’s “landslide” win is not particularly accurate. Though Trump swept the battleground states, the margins were close, and virtually all of the 2024 presidential pollsters forecast close races in the critical states, which is the end result.

A typical polling margin of error is three percentage points. Using this as the accuracy scale, the Real Clear Politics polling average suggests that across the battleground board, the average survey fell within such a range.

In national polling, the ending Real Clear Politics average found Kamala Harris cumulatively leading by the smallest of margins, at 0.1 percent. These numbers include 24 polls from 14 different polling organizations.

While they did miss the projection of which candidate was leading, the cumulative margin was well within the margin of error. Since Trump is winning the national popular vote by 1.8 percentage points, the polling community average of 0.1 point separating the two candidate was a miss of 1.9 percentage points, but well within the margin of error.

In the battleground states, the least accurate polling came in Arizona. There, the Real Clear Politics average suggested a 2.8 percent lead for Trump, but he carried the state by a 5.5 percent margin. This was a 2.7 point miss.

The other states where the polling mark was two points off or more came in North Carolina and Nevada. The NC polling average suggested a Trump win margin of 1.2 percentage points. In actuality, the president-elect carried the state by 3.2 percent. In Nevada, the miss was a bit worse.

Pollsters pegged Trump with a lead of 0.6 percent, but he carried the Silver State with a 3.1 percent margin, thus the cumulative miss was 2.5 points.

The most accurate of the state presidential polling came in Georgia. The average projection suggested a Trump lead of 1.3 percent, but he won the state with a 2.2-point margin. Thus, the average poll figure missed the actual result by 0.9 points, again well within the polling margin of error.

In the seven battleground states, the polling community correctly projected five of the seven Trump wins. The two misses were Michigan and Wisconsin, but both fell only a half-point or less off the pace.

The one consistent error point in all of the battleground states, and nationally, was underestimating the Republican vote strength. This has been a consistent pattern during the Trump era, and it happened again in 2024 but to a lesser degree.

The numbers in the 11 competitive Senate races were not as accurate as the presidential projections.

The least accurately polled state was Florida where Sen. Rick Scott (R) recorded almost a 13-point victory, yet the polling average for his race against former Congresswoman Debbie Mucarsel-Powell (D) suggested a much closer outcome. The two were separated by an average of 4.6 points in Sen. Scott’s favor.

A similar pattern, but to a lesser degree, occurred in Texas. For most of the race, the polling average showed a close race between Sen. Ted Cruz (R) and Rep. Colin Allred (D-Dallas), but the end result proved different. While polling found Cruz extending his small lead within the race’s final stage to an average of 4.4 points, the end result yielded the two-term senator an 8.6 percent win, meaning a polling miss of 4.2 points, well outside the polling margin of error and far beyond the early race pattern.

The most accurately polled Senate race was Arizona, where the cumulative average found Rep. Ruben Gallego (D-Phoenix) leading former news anchor Kari Lake (R) by an average of 2.6 points in the later stages of the race, after he posted larger early advantages. Gallego would win the statewide race by 2.4 percentage points meaning a polling miss of only 0.2 percent.

As in the presidential race, the consistent polling flaw was under-counting the Republican support. In 10 of the 11 monitored races, the Republican candidate exceeded the suggested polling support figure. The only contest where the Republican candidate failed to top the polling average came in Montana where challenger Tim Sheehy (R) defeated Sen. Jon Tester (D) by 7.1 percentage points. The polling average in the race’s latter stage was 7.6 percent.

The polling community correctly projected the Senate winner in 10 of the 11 monitored races. The only contender to defy the polling projections was Pennsylvania Republican David McCormick (R) who many media outlets have projected as a winner opposite Sen. Bob Casey Jr. (D). The race is in a recount where McCormick maintains a lead of better than 25,000 votes.

For the most part, the repeated polling criticism is exaggerated. Overall, the pollsters had a relatively good year, and did correctly see the Trump polling trend in that he won the battleground states and the national popular vote by largely predicted close margins. The consistent underestimating of Republican support, however, does indicate future methodology correction is warranted.