# Standard error computation with interaction term

I aimed at studying the effects of an exposure (E) on fetal growth estimated by repeated ultrasound measurements (n = 2 measures per participant in the following example). I used interaction terms of gestational age with exposure, in order to compute the effect of exposure at different age (using lincom in Stata). I am surprised because the standard error of the effect estimates computed with lincomest (performed after xtmixed) are similar at 22.5 and 32.6 gestational weeks (both SEs = 8, while Beta = - 1 and - 19 g at 22.5 and 32.6 gestational weeks, respectively). I am wondering if I correctly used lincomest.

    gen expoAG =  expo*AG       /* Gestational Age */
gen expoAGsq =  expo*AGsq  /* squared of Gestational Age */

xi : xtmixed EFW expo AG AGsq expoAG expoAGsq  if go_pen==1 ||n_ident:,  mle
lincomest _b[expo] +  _b[expoAG]*(22.5) + _b[expoAGsq]*(22.5*22.5)  /* Prediction at 22.5 GW */

xi : xtmixed EFW expo AG AGsq expoAG expoAGsq  if go_pen==1 ||n_ident:,  mle
lincomest _b[expo] +  _b[expoAG]*(32.6) + _b[expoAGsq]*(32.6*32.6)  /* Prediction at 32.6 GW */

• In the text editor there is a button with 2 brackets {}. Use that to post your code as code. – D L Dahly Dec 24 '13 at 8:59
• I've edited your post using our editing tools. Rather than Stata output, a small reproducible example would be helpful. – chl Dec 24 '13 at 10:03