Written on 22 October from the forecast and the coverage as they stood at the end of 18 October.

The coverage on 18 October had a gloomy tone. RealClearPolitics and The Hill both carried a piece arguing that concerning economic data posed problems for Republicans in the final stretch. NBC News published a poll describing Americans as anxious, pessimistic and unhappy with the President.

The forecast hardly noticed. The chance of Democrats winning the Senate moved from 57.7% to 58.1%, and the House from 91.5% to 92.1%. The national generic ballot average rose by about a third of a point, to just under six points in Democrats' favour. No competitive race moved by more than a fraction of a point.

Why economic news rarely shows up here

This is the pattern our own research found, and it is worth restating because days like this test it. In September we tested twenty economic measures, from petrol prices to inflation to the stock market, against 2,130 polls over twenty months.

The economy moves presidential approval, by modest amounts in the direction theory predicts. It barely moves the congressional vote. The effects on the generic ballot were four to eight times smaller, and mostly indistinguishable from zero. Political scientists have found the same thing in earlier decades.

The likely reason is that voters read the economy through their existing loyalties. People facing the same prices assign blame according to what they already believe, so bad economic news tends to confirm opinions rather than change them.

So when the forecast's national environment moves, the cause is almost always new polling, not new economic data. Whatever unhappiness voters feel will show up in the polls if it changes their vote, and the polls are already in the forecast.

A quiet day is a normal day

Most days in a campaign look like this one. The model reruns every six hours, and in a typical run nothing changes beyond a few tenths of a point. Large moves come from three places: a batch of new polls in a thinly polled race, a shift in the national generic ballot, or a change to the model itself, which we report when we make one.

Texas is a good example of how stable a heavily polled race becomes. With 26 polls in its average, Ken Paxton's lead over James Talarico moved by five hundredths of a point on the day, to about half a point. A race that close is still a genuine toss-up, but it is a stable one. A single new poll, however striking its headline, can only move it a little.

One cautionary note on days like this: stability is not certainty. The forecast is steady because the polls are steady, not because the polls are right. A shared polling error, of the size seen in several recent cycles, would move every close race at once, and nothing in a quiet day's numbers can rule it out.