One weird trick for speeding up coefplot on large data sets

[David Roodman]

Ben Jann's super-useful coefplot generates temporary variables in order to make its plots. And it leaves those variables behind if you use the generate option. I've found that in large data sets this causes a severe speed penalty. The temporary variables are filled mostly with missing, because the data set is much taller than coefplot needs for its variables.

I found an elegant fix for Stata versions that support frames:

cap frame create cp  // one-time setup

sysuse auto
reg price mpg
frame cp: coefplot

Example with the generate option:

frame cp: coefplot, gen replace
frame cp: `r(graph)' title(Great graph)

In older Stata I would sometimes do

preserve
keep in 1
coefplot
restore

But using a frame should be faster.

[Ben Jann]

Many thanks for pointing this out, David. I fixed the problem in the latest update (ssc install coefplot, replace); there is no longer a performance loss in large datasets. ben