The November Metric

For General Elections, we need a metric that does what pWAR doesn't. UGEPM is that.

Over the past several months, we’ve published pWAR scores for races up and down the ballot, using the Political Wins Above Replacement framework to evaluate candidates as something more useful than a name on a yard sign. pWAR measures candidate quality: who outperforms their environment, who bleeds votes their party should be keeping, who is a genuine electoral asset and who is a liability wearing a jersey. It answers the question the party establishments consistently refuse to ask — is this person actually better than a generic placeholder?

The problem with stopping at pWAR is obvious. A candidate with a great score in a terrible structural environment still loses. A candidate with a terrible score in a rock-solid partisan fortress still wins. pWAR, on its own, tells you about candidates. It doesn’t tell you about races. The Universal General Election Probability Model is the bridge.

Here’s how it works. Every input to the model gets converted into margin points, which are summed and run through a logistic probability curve calibrated by race level. The inputs are: structural partisan lean (the biggest driver by design, because most races aren’t competitive), the political environment, the incumbent factor, the funding gap, any scandal or controversy adjustment, the turnout factor, and the pWAR delta between the two candidates. That last one is the direct handoff from the primary analysis. The pWAR delta, multiplied by a level-specific weight, tells the model how much candidate quality moves the needle given the structural realities of the race.

The level calibration matters. At the presidential level, the environment and the race are functionally the same thing, so the environment weight is 1.0. At the local level, name recognition and candidate quality dominate because national waves barely penetrate a city council race, so pWAR weight rises to 1.0 and environment weight drops to 0.20. Governor races sit in the middle: sigma of 7.5 (meaning the logistic curve needs meaningful margin accumulation before probability shifts sharply), pWAR weight of 0.80, environment weight of 0.65. Structural lean still dominates. But in a genuinely competitive race, the other variables can and do move the outcome — and that’s exactly what the model is designed to show.

Let’s run three of them.