triple interaction and lincom

[Karen Arulsamy]

Hi all,

Based on the below equation, I would like to obtain the gender difference (sex_main, 1 is female) in the relationship between dfruit_simp and mh_likert during the pandemic (treat_covid1, 1 captures the pandemic).

global controls_simp i.agecat_sec i.educ_simp i.race_main i.gor_main i.imonth_final

xtreg mh_likert i.dfruit_simp i.dfruit_simp##i.treat_covid1##i.sex_main $controls_simp if wsample, fe vce(cluster pidp)

Could I get some advice on how to do this? I've tried various iterations of the below and the coefficients do not match the coefficients obtained when I have run this equation separately for men and women with the i.dfruit_simp##i.treat_covid1 interaction only.

lincom _b[1.dfruit_simp] + _b[1.dfruit_simp#1.treat_covid1] + _b[1.dfruit_simp#1.sex_main] + ///
_b[1.dfruit_simp#1.treat_covid1#1.sex_main] + _b[1.dfruit_simp] + _b[1.dfruit_simp#1.treat_covid1]

Thank you in advance for your help with this.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(mh_likert dfruit_simp treat_covid1 sex_main agecat_sec educ_simp race_main gor_main imonth_final)
11 1 0 1 0 0 1 7 10
19 1 0 1 0 0 1 7 6
9 0 0 1 1 1 1 7 8
16 0 0 1 1 1 1 7 8
10 1 0 1 0 0 1 5 3
8 1 0 1 0 0 1 5 4
7 1 0 1 1 0 1 8 1
17 1 0 1 1 0 1 8 1
7 1 0 1 0 1 1 5 6
6 1 0 1 0 1 1 5 9
6 1 0 1 0 0 1 10 7
13 0 0 1 0 0 1 4 7
6 1 0 1 0 1 1 5 1
35 0 0 0 0 0 1 2 9
12 1 0 1 2 1 0 1 9
12 1 0 1 2 1 0 1 10
10 0 0 0 1 1 1 4 4
12 1 0 0 2 0 1 8 11
7 1 0 0 2 0 1 8 10
11 0 0 0 0 1 1 10 5
23 0 0 1 0 0 1 3 3
10 0 0 1 0 0 1 3 1
11 0 0 1 0 1 1 7 12
11 1 0 1 0 0 1 11 3
7 1 0 1 0 0 0 12 5
18 1 0 1 0 0 0 12 7
9 1 0 1 2 1 0 12 7
9 1 0 1 2 1 0 12 2
4 1 0 0 1 1 1 10 8
12 1 0 0 3 1 1 3 8
10 1 0 0 3 1 1 3 9
22 0 0 0 0 1 1 11 4
29 0 0 0 0 1 1 11 5
8 0 0 0 1 0 1 11 12
7 0 0 0 1 0 1 11 2
11 1 0 1 1 1 1 7 1
7 0 0 1 2 0 0 7 3
8 1 0 1 2 0 0 7 1
7 1 0 0 3 1 0 1 4
10 1 0 0 3 1 0 1 3
11 1 0 1 1 0 1 1 3
26 0 0 1 1 0 1 1 3
18 0 0 1 0 0 1 1 3
27 0 0 1 3 0 1 1 3
12 1 0 1 2 0 1 1 4
12 1 0 1 2 0 1 1 3
14 1 0 0 1 1 1 1 4
10 1 0 0 1 1 1 1 3
16 1 0 1 2 1 1 1 3
13 1 0 1 2 1 1 1 3
end
[/CODE]

[Clyde Schechter]

When you use an interaction to estimate separate effects in a simple model such as -regress y i.m##c.x-, you will get the same effects of x in each level of m as you would get if you ran a separate -regress y x if m == ...- for each level of m. However, once there are other variables in the model (and you have lots of them) then the two approaches are not exactly equivalent. The difference arises because in the interaction approach, the effects of the other variables are constrained to be the same for all values of m, whereas when you do separate regressions, you get different coefficients for the other variables each time. If you want to replicate the separate regressions approach, you have to interact the moderator variable with all of the other variables, not just with x. So that would look like this:

xtreg mh_likert i.sex##(i.dfruit_simp##i.treat_covid1 $controls_simp) if wsample, fe vce(cluster pidp)

[Karen Arulsamy]

Thank you very much for your help with this Clyde Schechter , appreciate that a lot.