Read on Twitter
Don& #39;t know who needs to hear this, but:
Knowing forecasts about P(Trump_wins (nationally)) is primarily useful only for making informed decisions now about stuff *after* the election e.g. calculating P(event | Trump_wins) - P(event | ¬Trump_wins).
Knowing forecasts about P(Trump_wins (nationally)) is primarily useful only for making informed decisions now about stuff *after* the election e.g. calculating P(event | Trump_wins) - P(event | ¬Trump_wins).
National forecasts shouldn& #39;t be useful for deciding whether to change courses of action to try to decrease P(Trump) as the disutility of Trump winning shouldn& #39;t be particularly affected by whether his forecasted probability is say, .2 or .8.
For deciding to change courses of action, you primarily want to know, for each vote you could cause/change:
- the Pr of the vote& #39;s state’s electoral votes being necessary, and
- the Pr of the state being tied, given that its electoral votes are necessary
- the Pr of the vote& #39;s state’s electoral votes being necessary, and
- the Pr of the state being tied, given that its electoral votes are necessary
- the expected cost of causing the vote.
This let& #39;s you estimate the cost of influencing the election by a certain probability.
This let& #39;s you estimate the cost of influencing the election by a certain probability.
http://www.stat.columbia.edu/~gelman/research/published/probdecisive2.pdf">https://www.stat.columbia.edu/~gelman/r...
Like if you have no preparations/decisions to make between now and the election about post-election events, knowing the ongoing cost of a microprobability of Biden winning (a µBiden??) should be ~∞ more valuable than knowing forecasted national win percentages.
