Does wyoung allow for interactions?

[Marry Lee]

Hello,
My goal is to include regressions from the main analysis (X is my key regressors) as well as the heterogeneity analysis (using c.X#i.male) in the same wyoung command to account for multiple testing.
However, I did not succeed.


I tried with this data

sysuse auto.dta, clear
set seed 20
wyoung, cmd("reg price mpg" "reg price c.mpg#c.rep78") familyp(mpg c.mpg#c.rep78) bootstraps(100) seed(20)
local cmd1  reg price mpg
local cmd2  reg price c.mpg#c.rep78
wyoung, replace cmd("`cmd1'" "`cmd2'"  ) familyp(mpg c.mpg#c.rep78) bootstraps(50) seed(20)

It worked but still the results are weird. First, if I change the way I am writing the command, the results change while it should not. Second, the adjuted p-values are not adjsuted for the interaction term: c.mpg#c.rep78

 wyoung, cmd("reg price mpg" "reg price c.mpg#c.rep78") familyp(mpg c.mpg#c.rep78) bootstraps(100) seed(20)
Estimating family-wise adjusted p-values for 2 hypothesis tests

familyp: mpg
    reg price mpg

familyp: c.mpg#c.rep78
    reg price c.mpg#c.rep78

     +------------------------------------------------------------------------------------------------------------------------------+
     | k                     model   outcome         familyp         coef      stderr           p   pwyoung       pbonf      psidak |
     |------------------------------------------------------------------------------------------------------------------------------|
  1. | 1             reg price mpg     price             mpg   -238.89435   53.076687   .00002546         0   .00005092   .00005092 |
  2. | 2   reg price c.mpg#c.rep78     price   c.mpg#c.rep78   -18.756467    8.987922   .04070956       .02   .04070956   .04070956 |
     +------------------------------------------------------------------------------------------------------------------------------+

. local cmd1  reg price mpg

. local cmd2  reg price c.mpg#c.rep78

. wyoung, replace cmd("`cmd1'" "`cmd2'"  ) familyp(mpg c.mpg#c.rep78) bootstraps(50) seed(20)
Estimating family-wise adjusted p-values for 2 hypothesis tests

familyp: mpg
    reg price mpg

familyp: c.mpg#c.rep78
    reg price c.mpg#c.rep78

     +------------------------------------------------------------------------------------------------------------------------------+
     | k                     model   outcome         familyp         coef      stderr           p   pwyoung       pbonf      psidak |
     |------------------------------------------------------------------------------------------------------------------------------|
  1. | 1             reg price mpg     price             mpg   -238.89435   53.076687   .00002546         0   .00005092   .00005092 |
  2. | 2   reg price c.mpg#c.rep78     price   c.mpg#c.rep78   -18.756467    8.987922   .04070956       .04   .04070956   .04070956 |
     +------------------------------------------------------------------------------------------------------------------------------+

[Marry Lee]

Added: I also noticed that the interaction works when using a continuous variable (c.) but does not work with a binary variable (i.).

[George Ford]

It's the seed. You include it in the wyoung command, so drop the stand alone.

sysuse auto.dta, clear
wyoung, cmd("reg price mpg" "reg price c.mpg#c.rep78") familyp(mpg c.mpg#c.rep78) bootstraps(100) seed(20)

local cmd1  reg price mpg
local cmd2  reg price c.mpg#c.rep78

wyoung, replace cmd("`cmd1'" "`cmd2'"  ) familyp(mpg c.mpg#c.rep78) bootstraps(50) seed(20)