Down Ballot as an Early Signal, and Why Almost Nobody Reads It

Several thousand contested local races run in this country every year. The results are public. Almost nobody adds them up.

They sit on the websites of thousands of separate county and municipal offices, in PDFs, in HTML tables, in scanned images, with inconsistent office names and no common identifiers. Assembling them is slow, unglamorous data work, and that is most of the reason a continuous record of political movement goes unread.

Consider what it would have been worth reading. Democrats took the House and the Senate on 11/03/2026 and flipped five Republican held governorships on the same night, including Texas, Georgia, Ohio, Iowa and Nevada. Movement on that scale does not materialize on a Tuesday in November. It builds, and it becomes visible first in the races that nobody collects.

Local results are a real and continuous source of information about political movement. They are available to anyone. Most federal campaigns never look.

Why local results carry signal

Four properties do the work.

There are a lot of them. Municipal, county, school board and special district contests run by the thousand every year, many in odd years with no federal race on the ballot. Thousands of observations a year, against the handful of special elections most analysts watch.

They are less insulated. A long serving member of Congress is protected by name recognition, incumbency and a large advertising budget. A county commissioner usually is not. When sentiment moves, it shows up first where there is less in the way.

They record behavior. A local result says what people did in a booth. Published national polling asks what people intend to do, which is noisier, and noisier still in low turnout elections where the hard question is who shows up at all.

They arrive early. Off cycle and special elections run continuously, so the information is available twelve to eighteen months before a federal general. That is the window when recruitment and resource decisions actually get made.

Why the data gets ignored anyway

The reasons are boring and practical, which is why the gap has lasted this long.

Nobody aggregates it nationally. Federal results get collected and standardized by news organizations and state election authorities in reasonably consistent formats. Local results come from thousands of separate offices in whatever shape each office prefers.

Each race is individually unimportant. No single school board contest justifies the cost of collecting it. The value appears only in aggregate, so somebody has to fund the entire collection before any of it pays.

Attention follows stakes. Coverage, consultant interest and donor money concentrate on federal races, and analytical capacity accumulates where the attention is. 2026 was the most expensive midterm cycle on record, and almost none of that money went into finding out where the map was moving before the cycle started.

What each cycle cost

AdImpact put the 2026 cycle at a record 11.6 billion dollars, above the 2024 presidential cycle and well above the last midterm.

And the comparisons are genuinely hard. A county race with one name on the ballot tells you nothing. Turnout in an odd year municipal election is not comparable to a midterm. Local party labels are sometimes absent. An index has to handle all of that honestly, and a sloppy index is worse than no index at all, because it produces confident wrong answers that somebody will spend money against.

How to build the index properly

For anyone who wants to do this rather than read about it, the construction matters more than the collection.

Take contested partisan races only, and keep a record of what you excluded and why. Uncontested seats and nonpartisan contests without reliable party identification get dropped, not guessed at.

Compare like timing to like timing. An odd year municipal result belongs in a series with other odd year municipal results. The comparison you want is the same jurisdiction, the same office type, the previous comparable election.

Weight by registered voters or by past turnout instead of counting every jurisdiction once. Otherwise a few hundred small districts drown out the county with a city in it.

Report a direction and an uncertainty band. Anybody turning this data into a single predicted national margin is claiming more than it can carry.

Run it as a series. One reading is close to worthless. The value is in the slope across many contests over many months, which is why the collection has to be running before a cycle starts rather than spun up once the cycle gets interesting.

What an index like this is good for

Not calling the result. Three narrower uses, all of them worth more.

Deciding where to contest. If local returns across a region show sustained movement in one direction, that is an argument for fielding candidates in federal seats there that current ratings call safe. Recruitment decisions get made eighteen months out, when public polling of specific federal races barely exists. Local returns are one of the few real inputs on the table at that point.

Setting defensive priorities. The same signal running the other way is a warning to whoever holds those seats, and it arrives early enough to act on. Resources committed in the final month against a surprise are bought at the worst price of the cycle.

Checking a national story. If a party believes a particular issue is moving voters, local races in the places where that issue is most salient are a cheap test. The test is imperfect. It beats conviction.

The structural point

The 2026 Senate flips spanned four regions and included Kansas, where no Democrat had been elected to the Senate in almost a century, and Texas, where none had since 1988. Five governorships went the same way. A map that wide was building somewhere observable, and the observable places were the county commissions, the school boards and the odd year municipal contests that nobody collects.

The ability to read that is a choice a party makes years ahead of a cycle rather than a tool it buys in October. There is no shortcut and no vendor who will sell you the back catalog.

The data will be there for 2027 and 2028. It will still be badly formatted, still unglamorous, and still largely unread.