I am conducting a study into the transmission of time preferences in spouses through their smoking behaviours. I have a balanced dataset where N=1464 and T=3. I am using a random effects probit model with errors clustered at their personal ID numbers.
The model is as follows:
xtprobit smokes_f c.sctimeimpulse_f i.smokes_m_1##c.sctimeimpulse_m c.age_f c.age_m i.employment_f i.employment_m i.good_ghealth_f i.reduc_f i.reduc_m c.rellength i.child_under15 c.loghhincome i.rural i.deprivation_scale i.state##i.year, re vce(cluster pid)
sctimeimpulse_f = Female time preference measure
sctimeimpulse_m = Male time preference measure
smokes_f = dummy variable of female's smoking status at t
smokes_m_1 = dummy variable of male's smoking status t-1
I have 2 problems which I would appreciate advice upon. I have attached the STATA outputs within a notepad file since the line size was too large.
Model 1 double interaction.txt
1) I know that there is collinearity between smokes_m and sctimeimpulse_m, yet what I am interested in is whether the likelihood of a female smoking given the partner smokes is increased by his time preference rate. Would I be making a mistake to include sctimeimpulse_m as an independent variable as well as being in the interaction, is it more correct to not do so due to the collinearity? As an independent variable, sctimeimpulse_m is insignificant.
I perform the margins command:
margins r.smokes_m_1, dydx(sctimeimpulse_m) atmeans
The result is insignificant also yet the interaction term is significant in the xtprobit output. Why is this if in the xtprobit output the interaction is significant?
2) I performed margins, dydx(*) out of general interest for the other variables in the model. The p values change a fair bit from the xtprobit output. Why is this? Which output should I refer to for the general significance of the variables: xtprobit or margins?
0 Response to Inclusion of interaction terms and interpretation of margins output in RE xtprobit
Post a Comment