The results of the Nov 3 national election remain unclear. We do not yet know who won all races.
But we do know that pollsters and the media lost as the election results are quite different than recent forecasts.
In Portland (I live in eastern Oregon), only a few weeks ago, one poll projected the Mayor’s race would be won by the challenger to the incumbent – and by a double digit margin. Instead, the incumbent won by +6%.
Polls are inherently uncertain due to the difficulties of obtaining a representative sample and routine sampling error. Just as with uncertain models, we give too much credence to polls. We have an innate desire to predict the future, but the real world frequently takes its own path.
A poll – and the underlying model used to adjust for sampling problems – produces an hypothesis that is tested by real world data (the votes).
A model is a guess about how we think the world works.
The output of a model is also a guess – and not reality.
And that is a point we must keep in mind when considering model output. A model enable us to play “what-if” games to evaluate possible future scenarios. Models of complex chaotic behaviors can only generate possible scenarios – but cannot tell us which scenario is going to occur.
A model does not confirm anything at all.