LogFAQs > #940241660

red sox 777 posted...

This is probably in jest, but there's some incorrect math thinking behind this. The model is telling you its view of probabilities, but those probabilities don't necessarily reflect reality. You need to evaluate the accuracy of the model too. For example, if we roll a 6-sided die, we are very confident that each side has a 1/6 chance of hitting. We remain very confident about this even if say, after 50 rolls, it landed on "3" 25 times. While it seems like an anomalous result, we are so confident in our model (that each side has an equal chance of hitting) that we think it's just a run of luck for "3." But what if it lands on "3" all 50 times? What if it was 500 out of 500? At some point, we need to start questioning whether this die is fair or if it has been loaded to always land on 3.

If you have a model that predicts a single election and says candidate A has a 99% chance of winning and they lose, the most likely thing is that the model has some flaws. While it's possible the model just got unlucky, it's unlikely. Now if your model has called 99% of previous elections correctly - then we say that the model probably just got unlucky this time. But if you have no track record? Then statistically probably the model is off.

in order to run true accuracy on the model you would need like hundreds if not thousands of iterations of the election.

As it stands, if the results fell within the range of the model, even if on an extreme end, it still fell within the model.

---

"First thing that crosses my mind: I didn't get any GameFAQs Karma yesterday." Math Murderer after getting his appendix removed.