Re: st: Wishlist Table command with standard error

[n j cox <n.j.cox@durham.ac.uk>]

Garry's request is reasonable, and joins the list of several
hundred (guess) equally reasonable requests of the same status.

It's a little trickier than you might guess. -table- is
really a front end for other programs that do most of the
work, in this case -collapse- and -tabdisp-. As I understand it,
you would need to hit -collapse- as well as -table-. In principle,
a competent user-programmer could clone these programs and make
the changes needed. So, in principle, no one need wait for
StataCorp to agree and then get round to doing it: it could
be done today. In practice, another way seems more desirable.
You would only need to change a few lines of code; it's finding
those among several hundred that is the issue.

Svend's reminder of -tabstat-'s flexibility points to one good
solution, but it won't extend so well to more complicated examples
than he gives, even two-way tables.

There is at least one other way that is much easier. Back-tracking, the 
problem has two parts, calculating summary measures and then tabulating
them. We can do all this with standard commands from official Stata.

. sysuse auto, clear
(1978 Automobile Data)

The first trick is to use -egen- functions to get the summary
statistics. (Those who know -egenmore- on SSC will know that
it has an -semean()- function, but we can get there without
it.)

. egen n = count(mpg), by(foreign rep78)

. egen mean = mean(mpg), by(foreign rep78)

. egen sd = sd(mpg), by(foreign rep78)
(1 missing value generated)

. gen se = sd / sqrt(n)
(1 missing value generated)

Naturally, all these functions are smart about missing values.

Now we fire up -tabdisp- to do the table. -tabdisp- is billed
as a programmer's command, but it is fine for many interactive
tasks. The one thing to grasp is that all it does is tabulate,
nicely: what you want shown must be worked out upstream. We
can get two-way tables (and more):

. tabdisp foreign rep78, c(n mean se)

----------------------------------------------------------------------
          |                     Repair Record 1978
 Car type |        1         2         3         4         5         .
----------+-----------------------------------------------------------
 Domestic |        2         8        27         9         2         4
          |       21    19.125        19  18.44444        32     23.25
          |        3  1.328768  .7862783  1.528535         2  1.701715
          |
  Foreign |                            3         9         9         1
          |                     23.33333  24.88889  26.33333        14
          |                     1.452966  .9043789  3.122499
----------------------------------------------------------------------

Clearly, the number of decimal places is dopey. Add a -format()-
(even 2 d.p. is probably silly):

. tabdisp foreign rep78, c(n mean se)  format(%9.2f)

----------------------------------------------------
          |            Repair Record 1978
 Car type |     1      2      3      4      5      .
----------+-----------------------------------------
 Domestic |  2.00   8.00  27.00   9.00   2.00   4.00
          | 21.00  19.13  19.00  18.44  32.00  23.25
          |  3.00   1.33   0.79   1.53   2.00   1.70
          |
  Foreign |                3.00   9.00   9.00   1.00
          |               23.33  24.89  26.33  14.00
          |                1.45   0.90   3.12
----------------------------------------------------

Now the format is wrong for the counts. There is an
easy work-around. Make the -n- variable a string; then
it will be immune to any incantations of numeric format.

. tostring n, replace
n was float now str2

. tabdisp foreign rep78, c(n mean se)  format(%9.2f)

----------------------------------------------------
          |            Repair Record 1978
 Car type |     1      2      3      4      5      .
----------+-----------------------------------------
 Domestic |     2      8     27      9      2      4
          | 21.00  19.13  19.00  18.44  32.00  23.25
          |  3.00   1.33   0.79   1.53   2.00   1.70
          |
  Foreign |                   3      9      9      1
          |               23.33  24.89  26.33  14.00
          |                1.45   0.90   3.12
----------------------------------------------------

That's pretty much what -table- would give you. (In
fact, nicer.) Naturally you can leave out the column for
missing -rep78- if you wish.

There is some discussion of -tabdisp- in my

2003. Speaking Stata: Problems with tables, Part I.
Stata Journal 3(3):309--324.

Nick
n.j.cox@durham.ac.uk

Svend Juul
==========

Use -tabstat-; it is quite flexible:

. tabstat price, stat(n mean sem) by(foreign) col(stat)

Summary for variables: price
     by categories of: foreign (Car type)

 foreign |         N      mean  se(mean)
---------+------------------------------
Domestic |        52  6072.423  429.4911
 Foreign |        22  6384.682  558.9942
---------+------------------------------
   Total |        74  6165.257  342.8719
----------------------------------------

Garry Anderson
==============

The table command does not accept se or sem (the standard error of the
mean) in the contents(clist) option. I think this would be very useful
because a summary of a variable often includes n mean and sem.
*
*   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/