As @paulgp & I incorporate more-recent data and better (?) modeling, even 3.4 million may be conservative.
Could be closer to 4m than 3m.
Hope to post update before official release at 7:30am CT tomorrow.
Could be closer to 4m than 3m.
Hope to post update before official release at 7:30am CT tomorrow.
For now, here's where we've been trying to build out daily data. Thanks @Swazonomics & Dylan Piatt for pitching in on that.
Please let us know if you see errors, have additional reports we're missing, or whatever. Post or email/DM me.
docs.google.com/spreadsheets/d…
Please let us know if you see errors, have additional reports we're missing, or whatever. Post or email/DM me.
docs.google.com/spreadsheets/d…
Great note from Jesse Rothstein via @hshierholz. We've been modeling to match state's reports, which are not seasonally-adjusted (NSA). The seasonal adjustment will add 13.25% to the national NSA number today.
@paulgp & I posted updated estimates (NSA).
We add a new model focusing on the relationship between available daily UI claims reports from states & daily changes in @GoogleTrends. It predicts 3.8m (3.6-4.1), higher than the weekly-data model's 3.4m.
paulgp.github.io/GoogleTrendsUI…
We add a new model focusing on the relationship between available daily UI claims reports from states & daily changes in @GoogleTrends. It predicts 3.8m (3.6-4.1), higher than the weekly-data model's 3.4m.
paulgp.github.io/GoogleTrendsUI…
Unfortunately, the daily model seems to fit really well (R^2 = 0.59). Check out the scatter below. 
The fit is better than the weekly's (R^2 = 0.23).
Soon, we'll learn which performed better. This is all really novel, untested stuff in a dynamic environment. Huge signals happening though.
If there's a discrepancy in blog between text & table, trust the table. Editing fast.
Soon, we'll learn which performed better. This is all really novel, untested stuff in a dynamic environment. Huge signals happening though.
If there's a discrepancy in blog between text & table, trust the table. Editing fast.
Report is in.
2.9m not seasonally adjusted, so our models over-predicted.
The weekly model was closer and the daily was much too high. We will try to learn & improve.
The seasonally-adjusted number was 3.3m. Coincidentally, that's about what our weekly model predicted.
2.9m not seasonally adjusted, so our models over-predicted.
The weekly model was closer and the daily was much too high. We will try to learn & improve.
The seasonally-adjusted number was 3.3m. Coincidentally, that's about what our weekly model predicted.
The CA number (186k) is super-low. We had that penciled in for about 600K.
That may be revised up by a factor of 2-3 next week, I bet.
That may be revised up by a factor of 2-3 next week, I bet.
