To add some elaboration to Uli's excellent theme: The same
general issue arose at various points in the recent London
users' meeting.
For example, Roger Newson's programs
-parmby- and friends should be useful to you, as Uli
mentioned.
Within official Stata, -statsby:- is always there for
you. Here is a dopey example, not using -svy:-, but
the same point holds.
. sysuse auto, clear
(1978 Automobile Data)
. statsby ub=r(ub) lb=r(lb) se=r(se) mean=r(mean) N=r(N), by(foreign) total : ci price
(running ci on estimation sample)
command: ci price
ub: r(ub)
lb: r(lb)
se: r(se)
mean: r(mean)
N: r(N)
by: foreign
Statsby groups
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
..
. l, nola
+----------------------------------------------------------+
| foreign ub lb se mean N |
|----------------------------------------------------------|
1. | 0 6934.663 5210.184 429.4911 6072.423 52 |
2. | 1 7547.174 5222.19 558.9942 6384.682 22 |
3. | . 6848.6 5481.914 342.8719 6165.257 74 |
+----------------------------------------------------------+
. replace foreign = 2 in 3
(1 real change made)
. label def origin 2 "Total", add
. twoway rcap ub lb foreign || scatter mean foreign, xla(0/2, val notick)
yla(, ang(h)) xsc(r(-0.5,2.5) legend(off)
subtitle(means and 95% confidence intervals) ytitle("Price, USD")
Anyone knowing -ciplot- from SSC will realise that this device renders
-ciplot- pretty well superfluous.
Vince Wiggins added the nice tip (a "Yes, of course!" tip once you
see the point): Even if you have no -by:- variable, you just need
to create a dummy [comforter] to satisfy the infant, and then
-statsby- is useable.
gen all = 1
statsby ... , by(all) : ...
Nick
n.j.cox@durham.ac.uk
Ulrich Kohler
> Honey, Wayne, DOH wrote:
> > I find no reference to a corresponding special set of
> graphics commands
> > for use with complex sample data. In fact, I find no discussions
> > whatsoever regarding the graphing of complex sample data.
> Does anyone
> > know where one can find information that will settle my
> concerns that
> > Stata's graphs will present properly generated proportions/means,
> > confidence intervals, etc., for complex samples?
>
> I have a general note on this. The graphs you are looking for
> can be "easily"
> produced with Stata. Therefore you have to apply the
> "resultsset" concept,
> which is fairly general and not restricted to the complex
> sample statistics.
> It goes like this:
>
> (1) Calculate the statistics with the commands you like
> (2) Collect the statistics in a data set ("resultsset")
> (3) Load the resultsset
> (4) Draw the results using standard graph commands.
>
> Below my signature is an example for producing and graphing
> a resultsset for
> a survey statistic which I accidentely made yesterday. It
> uses -postfile-,
> which is sort of low level. Several helper programs, mostly
> by Roger Newson,
> exist for the concept in general. -findit resultsset- will
> point you to these programs.
<example>
*
* 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/