|From||Gary Longton <firstname.lastname@example.org>|
|Subject||Re: st: program to measure missing values of one variable againstothers|
|Date||Wed, 17 Nov 2004 09:31:44 -0800|
David Kantor wrote:
I will need a program to take a varlist, and for each variable, see how badly it is populated with missing values with respect to all the others.Jeroen Weesie's -mvpatterns- (STB 61, 2001) should do what you want - it takes a varlist and provides the frequencies of all missing value patterns for variables in the list. "findit mvpatterns" will point you to it.
The purpose is to see which variables are the spoilers -- the ones that cause the most damage in terms of limiting the usable set of observations in an analysis. Thus, it is not enough to say that x has 98 missing values. If they are mostly the same cases as have missing values in other variables, then it's not a problem. If they are largely in cases different from those with missing values in other variables, then it is a problem.
I can do this with not too much effort, but I ask in case someone knows of an existing program for this purpose. (Or is there something in standard Stata?)