egen xtile

[James Lang]

Hey! I am trying to do the following,

egen tier=xtile(diff), nq(3) by(id yq)

yq is the year and quarter.

It says

too many values
r(134);

There are 1.56 millions of observations and the dataset is 18MB. How to solve it?

Here are the sample data. Great Thanks!

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(diff id yq)
 .18 7001 216
 .25 7001 216
 .23 7001 216
 .11 7001 216
 .29 7001 216
 .46 7001 216
-.05 7001 216
-.04 7001 216
-.02 7001 216
-.05 7001 216
-.15 7001 217
-.01 7001 218
 .03 7001 218
 .02 7001 218
   0 7001 218
-.09 7001 218
   0 7001 219
 .02 7001 219
 .03 7001 219
 .01 7001 219
   0 7001 219
 .05 7001 220
 .04 7001 220
 .05 7001 220
 .06 7001 220
 .08 7001 220
 .08 7001 220
-.03 7001 221
-.01 7001 221
-.01 7001 221
-.05 7001 222
-.03 7001 222
-.04 7001 222
-.02 7001 222
-.03 7001 222
-.04 7001 222
-.01 7001 222
-.13 7001 223
-.14 7001 223
-.14 7001 223
-.14 7001 223
-.14 7001 223
-.13 7001 223
-.28 7001 223
-.29 7001 223
-.13 7001 223
 -.1 7001 223
-.09 7001 223
   0 7001 224
   0 7001 224
 .03 7001 225
 .04 7001 225
-.13 2422 215
 .03 2422 215
 .14 2422 215
   0 2422 215
 .14 2422 215
 .47 2422 215
 .01 2422 216
-.13 2422 216
-.03 2422 216
 .01 2422 216
-.04 2422 216
-.04 2422 216
-.01 2422 216
 .02 2422 216
 .05 2422 216
-.05 2422 216
-.08 2422 216
-.01 2422 216
-.01 2422 216
-.02 2422 216
-.06 2422 216
-.01 2422 216
-.05 2422 216
-.02 2422 216
-.05 2422 216
 .08 2422 216
-.02 2422 216
-.01 2422 216
-.01 2422 216
-.05 2422 216
-.01 2422 216
   0 2422 216
-.06 2422 216
-.04 2422 217
-.05 2422 217
-.05 2422 217
 .01 2422 217
 .02 2422 217
-.01 2422 217
 .03 2422 217
   0 2422 217
-.04 2422 217
-.05 2422 217
   0 2422 217
-.03 2422 217
   0 2422 217
-.08 2422 217
-.04 2422 217
end

[Joro Kolev]

Try this one

. egen groups = group(id yq)

. summ groups, meanonly

. gen mahtier = .
(100 missing values generated)

. forvalues l = 1/`r(max)' {
  2. xtile temptier = diff if groups==`l', nquantiles(3)
  3. replace mahtier = temptier if groups==`l'
  4. drop temptier
  5. }
(6 real changes made)
(27 real changes made)
(15 real changes made)
(10 real changes made)
nquantiles() must be less than or equal to number of observations plus one
r(198);

It generates an error because in the example data there are even singletons by group, and I am asking Stata to compute 3 quantiles for such data which is impossible.

[Joro Kolev]

Or better yet replace the loop above with the loop below. (You still might have a problem because some by groups are too small, but at least the loop below will not break at the first occurrence of such a group with too few members.)

. forvalues l = 1/`r(max)' {
  2. cap xtile temptier = diff if groups==`l', nquantiles(3)
  3. cap replace mahtier = temptier if groups==`l'
  4. cap drop temptier
  5. }

[Attaullah Shah]

You can use astile as well. astile is extremely fast, even with larger datasets.

ssc install astile
bys id yq : astile  tier = diff, nq(3)

[James Lang]

Thank you both~ It's solved )