Re: st: svy: mean and descriptive tables

[jpitblado@stata.com (Jeff Pitblado, StataCorp LP)]

Claire Kamp Dush <ckamp-dush@ehe.osu.edu> wants to produce a table of means
and standard deviations from variables in a survey dataset:

> I am running several svy: mean commands for a descriptive statistic
> table I am creating.  Here is the code I am running:
> 
> After svy setting the variables
> 
> foreach var of varlist var1 var2 var3 {
> svy: mean `var'
> di sqrt(e(N) * el(e(V_srs), 1, 1))
> }
> 
> I am using Stata 9.  This code is working fine, and the first line
> calculates the weighted mean and the second the standard deviation.  My
> question is:  
> Is there a way for me to write a code to create a table for my means and
> standard deviations automatically?  For instance, I would like a table
> as follows:
> 
> Var 1		mean1		sd1
> Var 2		mean2		sd2
> Var 3		mean3		sd3
> 
> 
> Any suggestions on how I could do this?  Here is my first attempt at a
> program:
> 
> prog svymean 
> 	local mean = svy: mean `1'
> 	local sd = di sqrt(e(N) * el(e(V_srs), 1, 1))
> 	display e(varlist)	`mean'	`sd'
> end
> 
> Then I thought I would run this for each of my variables and copy and
> paste the results into a text file and then convert into an excel or
> word file.  But, this didn't work because it wouldn't calculate the svy:
> mean of `1'.  
> 	
> I searched Statalist and I did not see a ready solution, but I could
> have missed something.
> 
> In advance, I never know whether to e-mail thank yous and fill up
> everyone's in boxes that much more, so in advance, a whole-hearted thank
> you to anyone who replies.  I really appreciate it.

In Stata 10 there is a new -estat sd- command that does this, see the section
of this email titled 'Stata 10 Users' for an example.

Here is a rewrite of Claire's program that works.

prog mysvymean 
	quietly svy: mean `1'
	local sd = sqrt(e(N) * el(e(V_srs), 1, 1))
	display e(varlist) " " _b[`1'] " " `sd'
end

-----------------------------------------------------------------------------
Stata 9 Users

There are two ways I can think of building a table from the results of the
-svy: mean- command.

1.  Let -svy: mean- perform the estimation all at once, then pick the pieces
    I want from -e(b)- (via _b[...]) and -e(V_srs)-.  This will restrict the
    estimation sample to the set of observations that contain no missing
    values in the variables of interest.

2.  Collect the information from -svy: mean- by looping over the variables.
    This will allow the full set of observations to be used for computing the
    mean and sd for each varaible.

I wrote a program for each case and named them -mean_sd_table1- and
-mean_sd_table2-, corresponding to the above two descriptions.

***** BEGIN:
program mean_sd_table1
	syntax varlist

	tempname v
	quietly svy: mean `varlist'
	matrix `v' = e(N)*vecdiag(e(V_srs))

	local k : list sizeof varlist
	display as txt %12s "Variable"		///
		as txt %10s "Mean"		///
		as txt %12s "Std. Dev."
	forval i = 1/`k' {
		local var : word `i' of `varlist'
		display	      as txt %12s "`var'"	///
			" "   as res %9.0g _b[`var']	///
			"   " as res %9.0g sqrt(`v'[1,`i'])
	}
end

program mean_sd_table2
	syntax varlist

	tempname sd

	local k : list sizeof varlist
	display as txt %12s "Variable"		///
		as txt %10s "N"			///
		as txt %12s "Mean"		///
		as txt %12s "Std. Dev."
	forval i = 1/`k' {
		local var : word `i' of `varlist'
		quietly svy: mean `var'
		scalar `sd' = sqrt(e(N)*el(e(V_srs),1,1))
		display	      as txt %12s "`var'"	///
			" "   as res %9.0g e(N)		///
			"   " as res %9.0g _b[`var']	///
			"   " as res %9.0g `sd'
	}
end
***** END:

Here are some example results

***** BEGIN:
. webuse nhanes2

. svyset psu [pw=finalwgt], strata(strata)

      pweight: finalwgt
          VCE: linearized
  Single unit: missing
     Strata 1: strata
         SU 1: psu
        FPC 1: <zero>

. mean_sd_table1 smsa height weight bpsystol bpdiast tcresult tgresult
    Variable      Mean   Std. Dev.
        smsa   2.50289    1.235087
      height  168.6915    9.613919
      weight  72.05973    15.50079
    bpsystol  126.1283    21.16074
     bpdiast  80.38098    12.92431
    tcresult  211.3975    47.08961
    tgresult   138.576    95.90575

. mean_sd_table2 smsa height weight bpsystol bpdiast tcresult tgresult
    Variable         N        Mean   Std. Dev.
        smsa     10351    2.496329    1.239231
      height     10351    168.4599    9.699111
      weight     10351    71.90064    15.43281
    bpsystol     10351    126.9458    21.40073
     bpdiast     10351    81.01726    12.79357
    tcresult     10351    213.0977    48.45305
    tgresult      5050     138.576    95.90575
***** END:

Note that I include a column for the sample size in -mean_sd_table2- to show
that the estimation sample can differ between variables.

-----------------------------------------------------------------------------
Stata 10 Users

There is the new -estat sd- command allowed after -svy: mean- that produces
a table similar to the one produced by -mean_sd_table2-.  So a mean/std. dev.
table could be produced using something like the following example

	. webuse nhanes2
	. svyset psu [pw=finalwgt], strata(strata)
	. svy: mean smsa height weight bpsystol bpdiast tcresult tgresult
	. estat sd

The table resulting from -estat sd- is

-------------------------------------
             |       Mean   Std. Dev.
-------------+-----------------------
        smsa |    2.50289    1.235087
      height |   168.6915    9.613919
      weight |   72.05973    15.50079
    bpsystol |   126.1283    21.16074
     bpdiast |   80.38098    12.92431
    tcresult |   211.3975    47.08961
    tgresult |    138.576    95.90575
-------------------------------------

Note that -estat sd- also handles results that employed the -subpop()- option
of -svy- and the -over()- option of -mean-.

--Jeff
jpitblado@stata.com
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/