The Referee Test: Three States, Three Polls, Three Smaller Democratic Leads

Five weeks before Election Day, there are two wholly different questions a poll can answer. What do registered voters think? And what do the people most likely to cast ballots think?

Three new Senate polls give us both answers. In Michigan, Iowa, and Texas, the Democratic candidate leads among registered voters. Apply the likely-voter model, and every Democratic advantage gets smaller. Nobody has to guess what changed because these aren't three competing pollsters using three different recipes. They are three surveys conducted under the same polling operation.

Start in Michigan. Democratic Senate nominee Abdulrahman Mohamed El-Sayed leads Republican nominee and former Rep. Mike Rogers 51% to 48% among registered voters. Among likely voters, the race tightens, 50% to 49%, where a three-point Democratic advantage becomes one.

From Fox News:

Both candidates receive near-universal backing from their party. Rogers receives 95% among Republicans, while it’s 93% for El-Sayed among Democrats. In addition, Rogers gets 99% of MAGA Republicans and 88% of non-MAGA Republicans, while El-Sayed garners 98% of DSA Democrats and 90% of non-DSA Democrats. The small number of independents narrowly favor El-Sayed.
This year’s political landscape favors El-Sayed. The Democrats hold a 5-point advantage when voters are asked which party they want to control the Senate, 52% to 47%. But to capitalize on that, he’ll need to consolidate Democratic support after a contentious primary because 11% of likely voters who backed his primary opponent Haley Stevens say they plan to vote for Rogers.

Iowa repeats the pattern. Democratic State Rep. Josh Turek leads Republican Rep. Ashley Hinson 50% to 46% among registered voters. The likely-voter result is 49% to 47%. Four points become two, and both results remain within their respective margins of error.

Texas provides the biggest movement. Democratic State Rep. James Tala-freak-O holds a seven-point advantage over Republican Texas Attorney General Ken Paxton among registered voters. Among likely voters, Tala-freak-O leads only 51% to 49%, a two-point race that falls within the sampling margin.

From Fox News:

All this gives Talarico a slim 2-point edge over Paxton (51-49%) in the first Fox News Poll among Texas likely voters (so no comparisons are available). These matchup results are within the margin of sampling error.
Among the larger sample of registered voters, Talarico’s advantage widens to 7 points (just outside the margin of error). That’s up from a 3-point edge in July.
GOP ambivalence is further reflected in Senate preferences because non-MAGA Republicans are more than five times more likely to back Talarico than their MAGA counterparts (22% to 4%). The crossover echoes lingering divisions from the competitive GOP primary and runoff: 20% of likely voters who backed Republican incumbent Sen. John Cornyn in the March primary back Talarico.

Michigan: three points become one; Iowa: four becomes two; Texas: seven becomes two. The direction never changes, but the size of the race does.

The likely-voter model isn't a crystal ball. It uses factors including past voting history, interest in the election, intention to vote, age, education, race, ethnicity, and marital status. Those assumptions can be wrong; registered-voter surveys are legitimate, too. The problem begins when political coverage treats the two populations as interchangeable evidence of electoral momentum.

A Friday polling roundup declared that the week's numbers brought "more good news for Democrats." It accurately identified the Fox results as likely-voter polling but then folded in larger registered-voter leads from another pollster, including Abdulrahman Mohamed El-Sayed by seven in Michigan and former Democratic Sen. Sherrod Brown by eight in Ohio. The same article noted that a likely-voter Ohio poll had Brown ahead by only three.

The Referee Test makes sure the same standards — the same rules — are being applied equally across the board. It favors consistency over selective outrage.

Let's put these poll reports to the Referee Test: If the headline is about who is winning an election, shouldn't the distinction between registered voters and likely voters receive more than fine-print status as Election Day approaches?

The Fox numbers don't prove Republicans will win Michigan, Iowa, or Texas. They don't prove likely-voter models will nail turnout. Democrats lead all three Senate contests in those surveys, and Republicans have weaknesses the polling clearly documents.

But they do prove something simpler. In three battlegrounds, using the same polling organization and the same surveys, moving from everyone registered to those modeled as likely to vote shrinks the Democratic advantage every single time.

A poll can tell us what registered voters think. An election tells us who showed up. Five weeks out, coverage should make the difference impossible to miss.