Hacker News Reader

Whenever I see the IHME charts[1] (which are great) I think “I wish I could see the what the model predicted before today’s date so I can assess it.”

This shows laudable transparency, in my view.

(Also the model was directionally correct, it’s not like it was entirely the wrong shape.)

[1] https://covid19.healthdata.org/united-states-of-america/cali...

View on HN

The IHME model is proving to be wildly inaccurate on the downslope due to bad data coming out China. Their projections will continue to creep up in terms of deaths over the next few weeks because the predicted drop offs won’t be happening, sad to say.

If you want to see examples, check out their predictions on Italy. Their figures have been underestimating by over 50%, and the gaps will widen each day over the upcoming week. Heck, the ranges for the last few days have been outside their confidence interval.

A significant portion of this probably is compliance, but I'm with you completely on "the numbers out of China, particularly at the tail end are farcical". There was a drop off a cliff in cases.

Could you share how you would assess the model?

Do you have a background in disease modeling or anything?

I'm curious.

Reply on news.ycombinator.com