For those that wonder why we chose Sat May 16 as our evaluation, the CDC uses what's called an MMWR week/epiweek (https://wwwn.cdc.gov/nndss/document/MMWR_Week_overview.pdf …) that ends on Saturday. Hence, all CDC models provide weekly projections ending on Saturdays: http://cdc.gov/coronavirus/2019-ncov/covid-data/forecasting-us.html …
-
-
Show this thread
-
Due to the reason above, we perform weekly evaluations using data from the end of the epiweek (Saturday). You can find our past evaluations here: http://covid19-projections.com/historical-performance … All projections data from:https://github.com/reichlab/covid19-forecast-hub …
Show this thread
End of conversation
New conversation -
-
-
Is it fair to say that the baseline model (just averaging the past weeks deaths) is outperforming other complex model projections because R has stagnated around 1 for more than a month? (fig from
@XihongLin's NISS talk https://www.niss.org/sites/default/files/COVID-19-05-15-2020-COPSS-NISS-xl.pdf …)pic.twitter.com/iXkDOUONnJ
-
Yes I'd say that's a fair assessment. But of course, we've known that for some time. It's not surprising news.
- Show replies
New conversation -
-
-
I've been confused about one aspect. Why are the models not using the baseline more effectively e.g. focusing the model on predicting the residual over the baseline. Instead most of the models are substantially underperforming the baseline.
-
I run residuals on our projections every week, and that analysis certain helps. I can't speak for the other groups though.
End of conversation
New conversation -
-
-
How are some of these models so bad? What are they even doing? Like, imagine a weather forecast which was consistently being outperformed by a "weather today same as weather tomorrow" model.
-
That's a good analogy. If a region has 50% rain probability, then using the previous day's weather would yield ~50% accuracy (assuming independence). But if a region has only 10% rain probability, then using the previous day would yield ~82% accuracy. So that's tougher to beat.
End of conversation
New conversation -
-
-
I am confused. In may 1 you said it would be a few weeks to know if there would be a spike as States reooen. It has been more than three weeks since GA and FL reopened and no spike?
-
I'm wary of Georgia's reporting: https://twitter.com/Fahrenthold/status/1262029906512404480 … Also worried about Florida: https://www.businessinsider.com/florida-officials-stop-publishing-coronavirus-death-toll-data-2020-4 … Meanwhile, here's Texas: https://www.npr.org/sections/coronavirus-live-updates/2020/05/18/857943858/coronavirus-cases-surge-in-texas-panhandle-as-state-continues-to-reopen …
- Show replies
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.