Help with percentile rank

[Ruth Shen]

Dear Statalists,

I am running into a problem when generating percentile rank for each individual. Suppose I have sorted by group_id and income, my data set goes like:

group_id inidividual_id income nb

1 11 1000 0
1 12 1100 1
1 13 1100 1
2 21 1350 3
2 22 1400 4
3 31 1450 5
3 32 1500 6
3 33 1600 7
4 41 1620 8
4 42 1700 9

The definition of percentile rank is rank_i=n_b/N
n_b, number of individuals with income lower than income_i
N, total number of individuals in the group.

I have tried count, but it can only return one value and could not get nb correctly. Any suggestions will be welcomed. Thank you!


Ruth

[Rich Goldstein]

what you write is not entirely clear to me; however, assuming no ties, I think the following will give you what you want:

sort income
egen rank=rank(income), track
replace rank=rank-1
gen pctrank=rank/N

I also assume you know what N is

[Nick Cox]

See also http://www.stata.com/support/faqs/st...ing-positions/ for a detailed discussion taking Rich's approach further.

[Stephen Jenkins]

See also Philippe Van Kerm's fracrank module:

net describe sgini , from(http://medim.ceps.lu/stata)

and look also at the (linked) online help

[David Radwin]

I will add only that there is more than one "official" method for calculating percentiles: http://www.itl.nist.gov/div898/handb...on2/prc262.htm .

[shem shen]

See also Philippe Van Kerm's fracrank module:

net describe sgini , from(http://medim.ceps.lu/stata)

and look also at the (linked) online help

This fractional rank command is fantastic. Thanks!