[Ben Jann]

I now posted an update at GitHub that should finally fix the problem. dstat should now also work if set dp comma is on. To install, type

. net install dstat, replace from(https://raw.githubusercontent.com/benjann/dstat/main/)

[Ben Jann]

The problem should now be solved. dstat should now also work if set dp comma is on. To install the fixed version, type

. net install dstat, replace from(https://raw.githubusercontent.com/benjann/dstat/main/)

[Ben Jann]

sorry for the double post

[Marc Kaulisch]

Ben Jann : Thanks for the quick bug fixing - it works now.

[Jesse Wursten]

Maybe a silly suggestion, but it could be worth mentioning in the help file that if you want the min and max values, you can use p(0) and p(100). Might even be worth to add min and max in the same way median.

[Ben Jann]

Hi Jesse, thanks for the suggestion. I am reluctant to add min and max because I do not know how to estimate the standard errors. I dimly remember that estimating the maximum is a difficult problem (but maybe it is about bias, not variance; can't remember). It is true that p0 and p100 return the min and max, but they apply standard VCE methods for quantiles, which will not be valid at the extremes (for example, for p100 the standard error from these methods is 0 by construction). Does anyone have any thoughts on this? For example, on how to obtain valid influence functions for the min and max?

[Ben Jann]

On GitHub is now a version of dstat that supports min and max (and also range and midrange). dstat does not compute standard errors/influence functions for these statistics; it just sets the standard errors to zero.

See https://github.com/benjann/dstat/ for instructions on how to install the update from GitHub (as well as the list of changes; there have also been some other changes/additions). Alternatively, wait until the changes make it into the "official" version in the SSC Archive.

ben