-ipfraking- with aggregated data?

[Tom Moultrie]

As part of an exercise that will eventually involve n-variate raking, I set out to reproduce the raking in example in Swanson and Siegel:

I was able to reproduce their results exactly, by creating an 8 242-record unit file (attached as simpleraking_Swanson.dta, ignoring the unknown ages; 285 duplicates of males aged less than 5, etc.), using the following code
use "simpleraking_Swanson.dta", clear
matrix define total_sex =( 1397073, 1445248)
matrix rownames total_sex = sex
matrix colnames total_sex =_one:1 _one:2

matrix define total_age=(201421,411140,1305492,532944,391324)
matrix rownames total_age = age
matrix colnames total_age =_one:1 _one:2 _one:3 _one:4 _one:5
ipfraking [pw=_one], ctotal(total_age total_sex) generate(rakedwgt2) iter(20)
tab age sex [iw =rakedwgt2]

All well and good.

BUT - the raking I wish to do is not amenable to creating a unit-record data file (it will be too big). So, I have been trying to reproduce the above with a file of 10 records with age and sex, and the count (as above) of elements in each combination of age and sex.
I cannot, however, work out what the weights should be that allow the derivation of the same results ... The closest I have been able to get with the aggregated file (..._weighted.dta), is by

use "simpleraking_Swanson_weighted.dta", clear
matrix define total_sex =( 1397073, 1445248)
matrix rownames total_sex = sex
matrix colnames total_sex =_one:1 _one:2

matrix define total_age=(201421,411140,1305492,532944,391324)
matrix rownames total_age = age
matrix colnames total_age =_one:1 _one:2 _one:3 _one:4 _one:5
ipfraking [pw=_one], ctotal(total_age total_sex) generate(rakedwgt2) iter(20)
gen temp = raked * _one
tab age sex [iw =temp]

Which provides the correct marginals, but the incorrect cells ...

I am sure I am missing something obvious ... so any suggestions welcome!

[skolenik]

You've confused ipfraking by making _one to be... not 1. In your weighted data,

rename _one input_weight
gen byte _one = 1
ipfraking [pw=input_weight], ctotal(total_age total_sex) generate(rakedwgt3) iter(20)

[Tom Moultrie]

You've confused ipfraking by making _one to be... not 1. In your weighted data,

rename _one input_weight
gen byte _one = 1
ipfraking [pw=input_weight], ctotal(total_age total_sex) generate(rakedwgt3) iter(20)

Dear Stas

Thank you (and face-palm!). But hopefully it will clarify matters for others!

Tom