Harry Enten, CNN's chief data analyst, put a number on the screen this week that should keep every Democratic strategist awake through November: 4.1 percentage points. That's the average amount Republican Senate candidates have outperformed their final polls in states Donald Trump won, measured across every cycle from 2018 to 2024.
Enten walked through the competitive 2026 Senate map and laid out what the current polls actually say. Kansas is tied. Iowa has the Democrat up 1. Maine, Democrat plus 2. Texas and Michigan, Democrat plus 3. Alaska, Democrat plus 3. Ohio, Democrat plus 4. Every single one of those margins falls inside the 3.7-point average polling error in recent Senate races. Every one.
Now apply the 4.1-point historical correction in Trump-won states. Enten's own math produces the result: a 52-48 Republican Senate majority. Not a Fox News projection. Not a Newsmax fantasy. CNN's data desk, running CNN's numbers, on CNN's air.
The gap between Trump states and non-Trump states makes it worse. In states Trump lost, Republican overperformance averaged just 0.2 points — essentially zero. But in states he carried, the polls consistently undercount Republican voters by a margin wide enough to flip every race on that board. The polling industry has known about this for three cycles running. They haven't fixed it.
Democrats looking at a plus-3 in Texas or a plus-4 in Ohio are looking at numbers that have been wrong in the same direction, by roughly the same margin, in the same kinds of states, for eight years. The 2024 Senate polls underestimated Republicans. The 2022 polls underestimated Republicans. The 2018 polls underestimated Republicans. At some point the word "error" stops applying and "pattern" takes over.
Ten days out from the midterms the network that functioned as an unofficial Democratic communications office for the better part of a decade just told its own audience that the scoreboard they're reading is calibrated wrong.
The polls say Democrats are ahead. The polls have said that before. The actual votes said something different — by 4.1 points.
