How about not adding scientism to unknown unknown uncertainty? The current result is separated by .2% of the total votes in a few states. A few mildly bad weather incidents... Why pretend we can model something like this—rare event, unreliable data—when we don't have the tools.
-
-
Replying to @zeynep @TJ__Murphy and
These probabilities are generally pretty calibrated! https://projects.fivethirtyeight.com/checking-our-work/us-senate-elections/ …pic.twitter.com/LKMgGCKP1G
2 replies 0 retweets 9 likes -
This Tweet is unavailable.
-
This Tweet is unavailable.
-
This Tweet is unavailable.
-
Replying to @simon_bazelon @HenryPorters and
No they are not, and that's exactly the problem. For weather, we have an enormous amount of data, every day, relentless opportunities for calibration, atmospheric physics and a chance to test things all the time. There are 12-13 presidential elections on which we base models.
1 reply 0 retweets 9 likes -
Replying to @zeynep @simon_bazelon and
OK, but even if true, that's not the same as saying "human events are not probabilistic." Betting odds are exceptionally good at forecasting the outcomes of horse races, for instance, which are the product of human-animal interactions.
1 reply 0 retweets 2 likes -
Replying to @JamesSurowiecki @zeynep and
And they're v. good at forecasting point spreads in NFL games, which are entirely human events. Obviously, there are boundaries around those kinds of events that don't apply to elections, and a lot more data to work with.
1 reply 0 retweets 3 likes -
Replying to @JamesSurowiecki @zeynep and
But the point is that some human events with uncertain outcomes are, in fact, probabilistically predictable. Not entirely clear why elections would be different, if we had more data and better data to work with.
2 replies 0 retweets 4 likes -
Replying to @JamesSurowiecki @simon_bazelon and
"If we had more and better data" is doing a lot of work here. Plus, horses don't have a stake and means of changing/influencing the betting odds. Plus, there is a lot of stochastic but consequential events outside the model (weather? last-minute scandal?) plus winner-takes-all.
2 replies 0 retweets 5 likes
Not everything can be modeled well, is the point. Ideally, you need a better understanding of mechanisms, ways of dealing with endogeneity, good data (!)... Less reflexivity, frequent events to allow calibration, fewer sources of correlated error, less coupled tipping points...
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.