Of all the races that ever were, the percentage of times a runner that the girl in orange was at that point in the race was X, would give useful information and still wouldn’t predict with 100% certainty the outcome of any given race…
Even if we had all the data from every race of the same event in history, and knew with certainty that runners in her current position at a given closeness to the finish line win, say, 25% of the time...
There are still far too many other variables to assert with any level of confidence that her win probability is anything close to 25%!
It could be - based on variables unique to the race much higher or lower. 5% or 95%.
The historical data is only one input.
The application of group-level data (especially an unscientific metric like "spread") to an individual is a textbook ecological fallacy.
This is heavily relied upon in the field's current analysis
The example you used of a 46-44-10 poll isn’t really what Silver etc are referring to. Most polls push undecideds and end up looking like 51-48.
But aside from that, the argument is basically that no polls are ever wrong, because they’re snapshots. Yet experts you partnered with last year, like Josh Smithley, argued pretty vehemently against certain polls such as the TIPP poll that deleted Philly, while praising other polls. So which is it?
Of all the races that ever were, the percentage of times a runner that the girl in orange was at that point in the race was X, would give useful information and still wouldn’t predict with 100% certainty the outcome of any given race…
You're absolutely right and I want to add to it
Even if we had all the data from every race of the same event in history, and knew with certainty that runners in her current position at a given closeness to the finish line win, say, 25% of the time...
There are still far too many other variables to assert with any level of confidence that her win probability is anything close to 25%!
It could be - based on variables unique to the race much higher or lower. 5% or 95%.
The historical data is only one input.
The application of group-level data (especially an unscientific metric like "spread") to an individual is a textbook ecological fallacy.
This is heavily relied upon in the field's current analysis
https://en.m.wikipedia.org/wiki/Ecological_fallacy
Frank’s First Rule of Probability:
Never bet on anything involving people…
You're talking to a guy who bets on NFL players to score touchdowns 😬🤣
Finally, someone gets this. I had nearly given up.
If you would like to work together - formally or informally - let's chat.
Glad to hear from you
Email me and let's set a time up.
The example you used of a 46-44-10 poll isn’t really what Silver etc are referring to. Most polls push undecideds and end up looking like 51-48.
But aside from that, the argument is basically that no polls are ever wrong, because they’re snapshots. Yet experts you partnered with last year, like Josh Smithley, argued pretty vehemently against certain polls such as the TIPP poll that deleted Philly, while praising other polls. So which is it?
"The example you used of a 46-44-10 poll isn't really what Silver etc are referring to"
Can you tell me what the poll averages were in 2016? Pick a swing state. Or nationally.
“Most polls push undecideds and end up looking like 51-48.”
Well, Silver's poll average in Pennsylvania was 48.1 - 48.2.
And Morris at 538 had numbers even slightly lower than that.
So if you'd like to retract and correct your claim of what “most polls” end up looking like, then we can continue.