How accurate are polls of the midterm elections likely to be?
The polling this year has consistently favored Democrats, who could be in position to secure control of both chambers of Congress. But all polls have some level of error. They underestimated support for President Trump in each of the previous three presidential elections.
In the midterms, it is a question of not only the size of the errors, but also their direction on a state-by-state basis.
To understand how things might play out, we looked at general election polls for the House, Senate and governor races between Republican and Democratic candidates from 2016 onward, examining the average polling bias by comparing their projected margins with the actual election results. The analysis was limited to states that saw more than five polls in the three weeks before Election Day.
Direction of Miss Varies From State to State
The polling errors that occurred in the recent presidential elections have raised questions about whether the Democratic leads this year will truly end up a mirage. Polling of recent midterms, however, has not had a similarly consistent bias toward Democrats. Instead, as a result of the different candidates, different campaigns and different electorates, the errors have varied across states and also from one election cycle to the next.
In 2018, polls overestimated the support for Democratic candidates in 18 states and for the Republican candidates in 12. In red states like Tennessee, Indiana and South Dakota, they overstated Democratic support, while in blue states like New York, California and New Mexico, they understated it. Michigan’s polls were particularly accurate, missing the final margin by less than one percentage point.
The results were similarly mixed in 2022. Polls that year overstated Democratic support in states including Oklahoma, Florida and Alaska, but overestimated Republican support in New Hampshire, Colorado and Washington.
These state-by-state errors can also change from one midterm cycle to the next. In Texas, polls understated Democratic performance in 2018 but overestimated it in 2022. In Michigan, the strong 2018 polls were followed in 2022 with ones that overrepresented Republican support by five percentage points. Maine had a modest two-percentage-point Democratic polling advantage in 2018 before swinging six percentage points to the right in 2022.
The reality is that a polling miss in one state has no affect on how accurate the polls will be in another.
How Would Previous Bias Affect This Year’s Midterms?
If polling error in the battleground states followed the same geographic pattern as it did in 2018, Democrats would be positioned to hold their current seats and compete for several others held by Republicans.
In that scenario, Georgia, Texas, New Hampshire, and Minnesota would end up in the Democratic column. Florida would remain Republican.
Several races, however, would be harder to call. Michigan, Iowa, Ohio, Kansas and Maine would be extremely close.
Alaska and North Carolina, which are frequently listed among this year’s battleground states, did not have enough polls in 2018 to reliably judge the polling error.
If the polling error of 2022 were repeated, it would produce a far different map. Democrats would hold Georgia, Michigan and New Hampshire and pick up seats in North Carolina and Maine. Republicans would hold Alaska and Florida.
The key battlegrounds would be Ohio, Texas and Iowa. Minnesota and Kansas, seen as two of this year’s battleground states, did not have enough polls in 2022 to provide a reasonable comparison.
The post Why a Polling Miss in One State Tells Us Little About the Next appeared first on New York Times.
