In Reply to: It's not just the CPI heading up posted by mh on May 13, 2026 at 11:26:27
I have seen polls in the last week ranging from D+3 to D+15. It makes no sense to me. I don't see this big a variation in other polls.
So I did what anyone would do. I asked Claude. Here is what it said:
Generic ballot polls have some structural quirks that make them noisier than, say, presidential approval ratings or head-to-head candidate matchups.
The question itself is vague by design. Asking "would you vote for a Democrat or Republican for Congress" doesn't map cleanly onto a real race. Voters don't actually vote in a national congressional election — they vote in one specific district against specific candidates. So the question measures something fuzzy, and different people interpret it differently.
Aggregation across wildly different districts. A national generic ballot averages together safe seats, competitive toss-ups, urban districts, rural districts, etc. Small shifts in who gets surveyed or how the sample is weighted can swing the number noticeably, because the underlying districts are so heterogeneous.
Likely voter screens vary enormously. Pollsters make different choices about who counts as a "likely voter," and those choices matter a lot for congressional polls. Midterm electorates are smaller and harder to model than presidential ones, so two pollsters using different screens can get results 4-5 points apart even from similar raw samples.
The translation to seats is highly nonlinear. Even when generic ballot numbers are consistent, they don't translate predictably to seat counts because of geographic sorting and gerrymandering. This makes people scrutinize the numbers more intensely and notice the variance more.
Less anchoring to a known quantity. Presidential approval polls have a concrete subject — one person — which gives pollsters a stable thing to measure and allows easy calibration against past results. The generic ballot measures a more abstract preference with no single anchor, so house effects across polling firms tend to be larger and more persistent.
The upshot is that a 3-point swing in the generic ballot can reflect a genuine shift in opinion, a different likely voter model, a different sample composition, or just noise from a poorly defined question — and it's often hard to tell which.
=============
Meanwhile the aggregators say it's D+5 or D+6.