LogFAQs > #911868028

red sox 777 posted...

All of those things are the same thing.....the model is not a perfect representation of reality and does not claim to be.

They're not the same thing on a fundamental level. It's true that the model has some total uncertainty, and that if I'm making a particular prediction I might not care where the uncertainty is coming from. But to understand what's going on we can distinguish between several types of uncertainty, e.g.:

1) Statistical uncertainty: maybe I really do have a perfect model for the underlying distribution of voters, but with a limited amount of data I'm fundamentally going to get some statistical noise.

2) Systematic errors: if I do something like screw up my likely voter model and think that youth turnout is going to be 20% instead of 30% when I'm weighting things I'm going to screw up (this is one level removed from 538 since it's a pollster error, but these mistakes are likely to be correlated among pollsters)

3) Time based errors: the final polls have pretty much been conducted already. If undecideds break differently than expected, that will change the results (or maybe a huge news story changes voter preferences in the last few days)

4) Fundamental uncertainty: uncertainty in the same sense as quantum uncertainty, e.g. even with perfect information you could not predict the outcome. This source shouldn't exist for elections, but we could philosophically consider it when we're asking about *why* the polls or models were wrong.
---

Congrats to BKSheikah for winning the BYIG Guru Challenge!