Hi,
I'd appreciate it greatly if you could comment on the following two inquiries:
1-) I run this interaction model in Stata 14:
xtreg Y c.l.X1##c.l.X2##c.l.X3##i.l.X4 l.Controls i.country i.year,r
where all variables are lagged and continuous except for the dichotomous X4. After running the regression, I type the following:
margins, dydx(l.X1 l.X2 l.X3 l.X4) force
Then I get the following average marginal effects (AMEs) for l.X1, l.X2, l.X3, l.X4 respectively: -0.012 (-1.42), 0.006 (0.93), 0.022 (4.66), 0.003 (1.06) with z-values in the parenthesis.
Is it accurate to interpret the AMEs as mentioned above in the following manner?
- Y is higher for l.X4s by an average of 0.003 (albeit insignificant at 10% level) compared to non-l.X4s, averaged across all l.X1, l.X2, l.X3 observations in the sample.
- The AME of l.X3 is 0.022 (significant at 1% level), meaning that on average (allowing for the interactions with l.X1, l.X2,l.X4) for this sample, an increase in l.X3 of one unit corresponds with an increase in Y of 0.022.
2-) How can I create a marginal effect plot for each of l.X1, l.X2, l.X3, l.X4 for the above model so that I could exhibit for which values/range of the covariates the AMEs of each variable are significant? If that plot is not so intuitive (if I could code is another problem of course) given many covariates, does it make more sense to report the exact values for which AME of each variable is significant after investigating it with the following relevant code:
margins, dydx(l.X1) at (l.X2==(0.1(0.1)0.6) l.X3==(0.2(0.1)0.7) l.X4=(0 1)) force
where continuous covariates change from their min to max values by 0.1 and X4 is dichotomous.
Best,
Lütfi
I'd appreciate it greatly if you could comment on the following two inquiries:
1-) I run this interaction model in Stata 14:
xtreg Y c.l.X1##c.l.X2##c.l.X3##i.l.X4 l.Controls i.country i.year,r
where all variables are lagged and continuous except for the dichotomous X4. After running the regression, I type the following:
margins, dydx(l.X1 l.X2 l.X3 l.X4) force
Then I get the following average marginal effects (AMEs) for l.X1, l.X2, l.X3, l.X4 respectively: -0.012 (-1.42), 0.006 (0.93), 0.022 (4.66), 0.003 (1.06) with z-values in the parenthesis.
Is it accurate to interpret the AMEs as mentioned above in the following manner?
- Y is higher for l.X4s by an average of 0.003 (albeit insignificant at 10% level) compared to non-l.X4s, averaged across all l.X1, l.X2, l.X3 observations in the sample.
- The AME of l.X3 is 0.022 (significant at 1% level), meaning that on average (allowing for the interactions with l.X1, l.X2,l.X4) for this sample, an increase in l.X3 of one unit corresponds with an increase in Y of 0.022.
2-) How can I create a marginal effect plot for each of l.X1, l.X2, l.X3, l.X4 for the above model so that I could exhibit for which values/range of the covariates the AMEs of each variable are significant? If that plot is not so intuitive (if I could code is another problem of course) given many covariates, does it make more sense to report the exact values for which AME of each variable is significant after investigating it with the following relevant code:
margins, dydx(l.X1) at (l.X2==(0.1(0.1)0.6) l.X3==(0.2(0.1)0.7) l.X4=(0 1)) force
where continuous covariates change from their min to max values by 0.1 and X4 is dichotomous.
Best,
Lütfi
