Adding weight to a procedure testing ICC of 2 groups for equality

[Sabine Schmitt]

Hello,

I'm using a program code in Stata 13.0 that examines, if two ICCs (one per group) significantly differ between groups. It runs for unweighted groups, but when adding a weight variable (I previously computed) the same results as for unweighted groups are produced. No error message occurs.

Code:

capture program drop icc_comp
program define icc_comp

syntax anything  [if/] [iw/] , id(varname)

* 1. argument = outcome 
* 2. argument = group variable

quietly tabulate `2' , generate(`2'___)

mixed `1'  `id'  `2' || `id' : `2'___*  , noconstant residual(independent, by(`2')) nolrtest nolog

estimates store Full

display in smcl as text "rho 0 = " exp(_b[lns1_1_1:_cons])^2 / ///
     (exp(_b[lns1_1_1:_cons])^2 + exp(_b[lnsig_e:_cons])^2)

  display in smcl as text "rho 1 = " exp(_b[lns1_1_2:_cons])^2 / ///
     (exp(_b[lns1_1_2:_cons])^2 + exp(_b[r_lns2ose:_cons] + _b[lnsig_e:_cons])^2)

  quietly mixed `1' || `id': , nolrtest nolog

  local linesize `c(linesize)'

  set linesize 80

  lrtest Full

  set linesize `linesize'

  drop `2'___*  _est_Full

end

One example: I used for the unweighted analysis "icc_comp N group , id(id)" and for the weighted analysis "icc_comp N group [iw=weight], id(id)". As said before, exactely the same values occured without an error message (for every analysis of my 7 outcome variables).

Does someone know how I have to alter the program code so that the weight variable is not ignored?

Thanks in advance and best regards,
Sabine

[Joseph Coveney]

You might have better luck getting replies if you post this in the General forum.

For now, here are two comments:

1. Your wrapper program accepts importance weights, but doesn't do anything with them, doesn't pass it through to mixed. (ditto with the if restriction.)

2. mixed doesn't allow importance weights.

[Sabine Schmitt]

Thanks for your reply Joseph, I posted the question in the General forum as you suggested!