Loop over imputed data.

[Moritz Huth]

Hello everyone,

I work with an imputed panel dataset on a household level. What you see here is an excerpt of my data:

* Example generated by -dataex-. For more info, type help dataex
clear
input double(weight _1_net_wealth _2_net_wealth _3_net_wealth _4_net_wealth _5_net_wealth) float ranknw1 byte pct_percent1
6030.4927 598500 598500 598500 598500 598500 1198  9
1539.1941 598000 598000 598000 598000 598000 1199  9
209.76678 596500 596500 596500 596500 596500 1200  9
1907.0204 595400 564400 581600 565300 588070 1201  9
433.08447 595300 589000 605000 601000 667150 1202  9
9238.9199 595000 595000 595000 595000 595000 1203  9
8029.6475 595000 595000 595000 595000 595000 1204  9
356.36459 592000 803500 592000 600000 612000 1205 10
6368.6235 591600 598200 623600 591600 592500 1206 10
 296.9743 591600 591600 591600 591600 591600 1207 10
625.55798 590900 590900 590900 590900 590900 1208 10
266.18207 590000 590000 590000 590000 590000 1209 10
365.48203 589200 589200 589200 589200 589200 1210 10
1012.8746 588700 588700 588700 588700 588700 1211 10
end

where
_1_net_wealth ... _5_net_wealth are the imputed net wealth variables of each household
ranknw1 ranks the households based on _1_net_wealth in descending order
pct_percent1 shows the distribution of percentiles based on _1_net_wealth
weight are the weights each household observation represents
My data ranges over 100 percentiles and is sorted by _1_net_wealth in descending order. However the ranks and percentile are in ascending order, meaning that the observation with the highest value of _1_net_wealth represents the first rank within the first percentile of the distribution.

The following code I would like to apply to the percentiles 1-10, for all five imputations:

// create ranks and percentiles
egen ranknw1 = rank(-_1_net_wealth), unique
xtile pct_percent1 = -_1_net_wealth [weight = weight], nquantiles(100)
gsort ranknw1

// create CCDF
keep if pct_percent1 == 10
bysort pct_percent (ranknw1) : gen ni = sum(weight)
egen nall = total(weight)
gen ccdf1 = ni/nall
gen log_ccdf1 = log(ccdf1)
gen log_nw1 = log(_1_net_wealth)

// estimate pareto alpha using OLS
reg ccdf1 log_nw1

To my question/problem:
How do I create a loop that applies the code above for all 10 percentiles and for all 5 imputations?

Thanks a lot and sorry for the long post
Moritz

[Rich Goldstein]

I don't follow but my guess is that one place to start is

help mi xeq

and, if you don't yet know about -foreach-, another place is

h foreach

[Moritz Huth]

Thank you, Rich. I checked with mi xeq and foreach before, but I was not able to solve the problem on my own, hence the post...