Calculating summary median (or other quantiles) for panel data

[Richard Hiscock]

Hi all,
I have unbalanced panel data and wish to calculate p50,p25 & p75 summary measures at each time the the panel.
The data thins out with increasing time and the last panels have only one or two subjects still remaining
Using egen & pctile to calculate summary measures at each time t produces quantile values even when only one or two subjects remain in the panel

I have replicated this below

I presume my code is incorrect and would appreciate any corrective input

many thanks in advance
Richard Hiscock

clear
input id t bob
1 0 6
1 1 7
1 2 10
1 3 7
2 0 10
2 1 6
2 2 8
2 3 9
2 4 4
2 5 4
2 6 3
2 7 4
3 0 16
3 1 6
3 2 10
3 3 8
3 4 7
3 5 6
4 0 6
4 1 7
5 0 6
5 1 10
5 2 4
end

xtset id t
xtline bob , overlay xlab(0(1)7)
sort id t
egen p50 =median(bob), by(t)
egen p25 = pctile(bob), by(t) p(25)

list id t bob p50 p25, sepby(id)

[Clyde Schechter]

There is nothing wrong with your code, and there is nothing wrong with these results.

[William Lisowski]

For further background on how percentiles are calculated by Stata, a good source is the Methods and formulas section of the documentation for the summarize command in the Stata Base Reference Manual PDF included in your Stata installation.

For a reasonably thorough and approachable discussion of the calculation of percentages in general, I recommend the Wikipedia article Percentile

[Richard Hiscock]

Clyde

Thanks for the quick feedback
I am still confused - at the expense of being declared very foolish

My understanding was that the 25th percentile was that observation for which 25% of upward ranked observations were below & 75% above (or averaged over adjacent values should none exist)

For my data id 2 t = 7 is the only observation yet clearly Stata computes p25 to = this value. Im unclear how this is calculated ? Shouldn't the25th percentile be undefined if observations are <=2?

any clarification would be very helpful

Cheers Richard Hiscock

[Richard Hiscock]

William,
our posts crossed over
Clear now using nearest method

thanks Richard Hiscock