I understand it quite well. I also understand the need for it to relate to reality.
Your turn.
Read up on the idea of an “expected value” and get back to all of us.
Sorry. Thought I linked to the The Oregonian’s story about it.
Tell us how 0.2 electoral votes can be an “expected value”.
Here’s a hint: In reality, EV’s are integers.
I suspect her post referred to the GOP’s press release congratulating Pence on his victory two hours before the debate began.
The only person claiming that Silver is predicting 0.2 electoral vote for Gary Johnson is you. Not Silver, not his model, and not “most ardent admirers.” Probabilistic models are used in complex simulations specifically because they’re too complex to be modeled mechanistically. If you insist on taking the results of that suite of simulations, presented as expected values or probabilities, and forcing them into a single discrete scenario, then no, you don’t understand stochastic modeling.
No, expected values are not integers. They’re abstract representations of the outcome of a mathematically defined process.
Need I repeat my point about losing sight of where the math goes from continuous to discrete?
The 0.2 EV projection is right off the 538 presidential polls page. Dunno why you’re denying it, but whatever. Bye.
No, but you should point out where you fail to understand basic statistics.
If you flip a coin one time what is the EV of the number of heads you’ll see?
It’s pretty clear on 538 that the most probable number of EVs for Johnson is 0. I am unsure how this can confuse anyone.
A fundamental misunderstanding of the term expected value.
The most probably outcome is the mode, not the expected value.
Can we take this to GQ already?
Well it is a fair cop on 538 that they present what is typically read as a prediction of results. Reporting the statistical expected value is somewhat foolish in this context as that result has exactly zero chance of occurring.
That’s not GQ albeit may be IMHO.
Would mode be the best thing to actually report? Or the bar that occurs the most frequently? Which is not the mode in this case - in the histogram it looks like the biggest chance is on Clinton 356 to my read but that still an under 1.5% chance
It’s a complex problem to model, and the results of the model are necessarily complex. There’s no way to boil the analysis of 20,000 simulations of 50 states each down to a single number. As soon as you start trying, you lose all the benefits of the probabilistic model.
As you say, even the most frequent result has a less than 2% probability. Put yourself in Silver’s place: you can write for sophisticated readers and report the expected values, or you can present a single projected electoral scenario with the caveat that “there’s a better than 98% chance that I’m wrong.” Who would give any credence to the latter?
Please no!
Is everybody using EV to refer to the same thing? (Electoral Votes, Expected Value)
Indeed, the above posters are using EV to mean two different things.
That said, I think Elvis is misunderstanding Nate Silver when he says:
Silver never claimed 0.2 electoral votes was the most probable outcome for Johnson. He claimed that it was the average number of electoral votes Johnson achieved across his simulations. (This is what “expected value” means.)
I clicked and skimmed and saw that Nate Silver had caused great confusion with something like EV[ev] — the expected value of electoral vote count.
I guess this EV/ev confusion is yet another reason to get rid of the electoral college. 