Here's the problem: none of the intuitions from reading political polls carry over to a graph like this. Uncertainties in exponential models are not like those in polling. The model uncertainty is enormous. And model averages are not polling averages; you can't aggregate them
-
-
Show this thread
-
We're in a media context where these models (which are indispensable, in a way
@zeynep has written about https://www.theatlantic.com/technology/archive/2020/04/coronavirus-models-arent-supposed-be-right/609271/ …) have already been misinterpreted as oracles of truth. Now@FiveThirtyEight is systematizing this misuse of models in easy visual form for policymakersShow this thread -
It's fine to write about this stuff, but it requires context, not a front-page visualization that readers can turn to to see how the president is polling today, and how many of us are going to die.
Show this thread -
Uncertainties in structurally dissimilar models of a poorly-understood exponential process are not like the uncertainties in polling everyone is used to seeing on this political analysis website, and presenting the data in this way guarantees misunderstanding
Show this thread -
Looking forward to next week, when
@Nate_Cohn introduces a gently quivering javascript needle showing the current R_e value for New York City.Show this thread -
Here's
@NateSilver538 providing the crucial state-by-state information we need to plan for the next month. There will be somewhere from zero to many thousands of coronavirus deaths in Montana and Nebraska. In their defense, I bet this forecast is not wrong!pic.twitter.com/KItuYF2EqK
Show this thread -
Coronavirus has many weird symptoms. Online it mostly manifests as an irreversible brain wasting disease, with the New York Times an early and dramatic superspreading event
Show this thread -
If you read the fine print on
@FiveThirtyEight, most coronavirus model have the same caveat: they assume no change in social distancing or other behavior. But we know this behavior changes with time and with the severity of the epidemic. These models are not a forecast at allpic.twitter.com/pyxNEGfbt3
Show this thread -
Finally, look at the confidence intervals on some of the specific models, looking a *month* into the future. These certainties are a mathematical artifact of the model. FiveThirtyEight is publishing this innumerate clickbait to replace lost political page views.pic.twitter.com/xhNE41pQgV
Show this thread -
Deleted this devastating dunk because people correctly pointed out that FiveThirtyEight belongs to ABC news, not the NYT. I was so sure in my mind of this ownership I didn't even bother to double-check. I apologize for the stupid error. I'm right about everything else, though!pic.twitter.com/iAKkrp3dqr
Show this thread -
Today we have validation of how one of the front page
@FiveThirtyEight models performs compared to reality. So will it stay up on the site? https://towardsdatascience.com/transparency-reproducibility-and-validity-of-covid-19-projection-models-78592e029f28 …pic.twitter.com/iCdwEcsnkn
Show this thread
End of conversation
New conversation -
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.