Thread by @ChrisGodlak

#thread
Following recent success of this thread (thanks @FatEmperor i guess :)just wanted to get back to this
i will just take the first graph as exemple: it's reproduction rate vs stringency index
panel (country-time in days) regression with random effects
#COVID19
1/n

https://twitter.com/ChrisGodlak/status/1353838559413866497

i put the initial graph as "benchmark" and i add 3 other graphs
panel regression with random effects
with countries dummies included
with weeks dummies included
with countries & weeks dummies included
and we will compare the R² as these are the only changing in this excercise
2/n

in panel regression what is most important is the R² within
adding country effects changes nothing
adding weeks (i.e. time trend or "seasonality" in this context) increases a lot the R² within (~ x100)
adding countries & weeks => same results as with weeks added only
3/n

so what is the takeway ?
with this available data and variations in specifications, one can argue that most of what goes on in the data is explained by weeks :) aka seasonality !
(are they putting #lockdowns during colder seasons and making things worse ?)
#COVID19
4/4

@threadreaderapp unroll

you can find me & such stata monkey analysis on my TG channel: t.me/chrisgodlak