tabstat with formats

[Shen YANG]

Hi All,

Quick question about using tabstat with format option. I managed to summarize two variables:

1. Group variable: pt15a
2. Date variable: startdate (need to identify the earliest date)

I ran the following code:

 tabstat startdate, by(pt15a) stats(count min)

and the result was like:

Apparently, the startdate value was not formatted, then I add the format option as below:

tabstat startdate, by(pt15a) stats(count min) nototal format(%tdDD/NN/CCYY)

Unexpectedly, it also formatted the number of observations.

Could anyone please advise how should I modify the code? Many thanks.

[Nick Cox]

I would switch to tabdisp here, although that is just one way to tackle the problem.

Most esoteric tip: with some display formats, tabstat shows sample size with decimal places, which I never want. But tabdisp allows a mix of formats, as if all else fails string variables can be used to show numerical results with any desired number of decimal places.

. clear

. set obs 100
Number of observations (_N) was 0, now 100.

. set seed 314159

. gen date = runiformint(mdy(1,1,1980), mdy(12,31,2010))

. egen group = seq(), block(10)

.
. bysort group : gen N = strofreal(_N)

. bysort group : egen first_date = min(date)

. format first_date %td

.
. tabdisp group, c(N first_date)

----------------------------------
    group |          N  first_date
----------+-----------------------
        1 |         10   03nov1983
        2 |         10   07aug1989
        3 |         10   22nov1983
        4 |         10   07mar1980
        5 |         10   17sep1981
        6 |         10   11nov1989
        7 |         10   29apr1981
        8 |         10   15mar1982
        9 |         10   09sep1985
       10 |         10   26jun1981
----------------------------------

.
. bysort group : gen n = _N

.
. tabdisp group, c(n first_date)

----------------------------------
    group |          n  first_date
----------+-----------------------
        1 |         10   03nov1983
        2 |         10   07aug1989
        3 |         10   22nov1983
        4 |         10   07mar1980
        5 |         10   17sep1981
        6 |         10   11nov1989
        7 |         10   29apr1981
        8 |         10   15mar1982
        9 |         10   09sep1985
       10 |         10   26jun1981
----------------------------------

That isn't needed for the example above. But remember the trick: call up strofreal() on the fly with a specified display format.

For example, moments from SSC allows any combination of formats for its results. The default is to show integers for sample size and 3 d.p. for everything else. Here you decide that 1 d.p. is enough for the mean and 2 d.p. enough for the SD, and accept the defaults for skewness and kurtosis.

. sysuse auto, clear
(1978 automobile data)

. moments mpg, by(foreign)

----------------------------------------------------------------------
    Group |          n        mean          SD    skewness    kurtosis
----------+-----------------------------------------------------------
 Domestic |         52      19.827       4.743       0.771       3.441
  Foreign |         22      24.773       6.611       0.657       3.107
----------------------------------------------------------------------

. moments mpg, by(foreign) format(%2.1f %3.2f)

----------------------------------------------------------------------
    Group |          n        mean          SD    skewness    kurtosis
----------+-----------------------------------------------------------
 Domestic |         52        19.8        4.74       0.771       3.441
  Foreign |         22        24.8        6.61       0.657       3.107
----------------------------------------------------------------------

You don't have to care about moments, or moment-based measures, for the example to be useful, as a glance at the code for moments shows how you can mix display formats for yourself.

For more tips in the same spirit, see https://journals.sagepub.com/doi/pdf...867X1201200109

[Hemanshu Kumar]

If you're on Stata 17 or later, you could also achieve this with the table command. Something like:

table pt15a , stat(count startdate) stat(min startdate) nformat(%tdDD/NN/CCYY min)

[Shen YANG]

Many thanks to Nick Cox and Hemanshu Kumar !