Re: 🧵 on Treating Statewide Elections as Independent Events

Hi everyone! It is clear that clarifications are needed behind the meaning of our 99.5% chance of Biden winning the election. It is understandable that you are all familiar with polling survey based methods and
resulting Presidential models, but this is not one of those! So, our model involves taking the theories of negative partisanship and demographic realignment to predict voter behavior. We then determined the probability for each state to be won by the candidate that is leading
the calculated 2-party vote share margin. For example, Biden has a 71% chance to win Wisconsin if our theory holds true because he is up by 4 pts in our MOVs for Biden. What this 99.5% probability means is that if NegPar/Realignment is true then there is a 99.5% chance he'll win.
It's not considering the error involved that our NegPar/Realignment model is wrong. For example, if Biden wins KS (which we have 0% for Biden) and the rest of the map is shifted more Democratically accordingly in Nov, then our model would be wrong! That's b/c our model gives this
event happening a 0% probability! The same goes for Biden undperforming. It's been brought up that if Biden underperforms in MI than he'll likely be behind in similar Rust Belt states. We agree! However, NegPar/Realignment would be wrong if that happens.
Further to reiterate, this model isn't meant to take polling data errors and incorporate a broad range of situations possible within those errors! It would be impossible for us to do that because most of our data inputs don't come with associated errors (demographics, previous
election results, etc.) What this 99.5% means is that if we are right in our methodology of being able to predict elections then Biden wins 99.5%. It is not an assessment of error on our methodology because we don't predominantly use polling surveys that tell us the "true" answer
should be between these error bars. This 99.5% means "If we are right about how elections should be predicted" then Biden wins 99.5% of the time. Therefore, probabilities of states are assessed as independent events because our methodology captures the predicted natl environment
and any failure for that to materialize is an entirely state-level situation. If Biden is winning in places like KS or losing the entire Rust Belt, then the way we constructed this model is not adequate for assessing the outcome of the 2020 Election.
The last thing we are trying to do is leave enough uncertainty in our predictions to suggest that we were right no-matter what because technically each candidate has a probability of winning.
We hope this clears things up.
You can follow @sam_epstein.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: