pull down to refresh
32 sats \ 2 replies \ @Undisciplined 10 Jan
I'm not so sure. It looks to me like the linear specification is just the wrong functional form for that regression discontinuity. An exponential trend line would probably fit the data pretty closely.
reply
60 sats \ 1 reply \ @zuspotirko OP 10 Jan
They y-axis is already in log scale
reply
18 sats \ 0 replies \ @Undisciplined 10 Jan
I know, but it still looks misspecified.
The other way to approach it is to dramatically reduce the time scale and look for a sharp kink or jump in the trend from immediately before to immediately after.
reply