Hello,
I am a trying to apply IPW method to account for (observable) selection bias with an observational data set (a programme with no control group). I have a panel data set with only two points of observations and am interested in assessing the marginal effect of "time" on ipaq_1 (levels of physical activity), while accounting for the issue of missing data, particularly, the missing of a very large proportion of outcome data (ipaq_1) at follow-up.
Here is my specification of the "outcome" and "treatment" models:
teffects ipwra (ipaq_1 time i.agemajorcat gender i.IMD) (TEST agemajorcat PAcat0test LCC ATT cohort, probit)
Now, the issue is that even simplifying the equations, the iterations seem not to stop. Basically it looks like as though convergence can't be reached, even if STATA hasn't given out such warning.
My questions are:
- is this approach valid for my problem?
- how to address the issue of convergence
- any alternative IPW approach
Thank you in advance for any comment on this,
Best wishes
Paolo
I am a trying to apply IPW method to account for (observable) selection bias with an observational data set (a programme with no control group). I have a panel data set with only two points of observations and am interested in assessing the marginal effect of "time" on ipaq_1 (levels of physical activity), while accounting for the issue of missing data, particularly, the missing of a very large proportion of outcome data (ipaq_1) at follow-up.
Here is my specification of the "outcome" and "treatment" models:
teffects ipwra (ipaq_1 time i.agemajorcat gender i.IMD) (TEST agemajorcat PAcat0test LCC ATT cohort, probit)
Now, the issue is that even simplifying the equations, the iterations seem not to stop. Basically it looks like as though convergence can't be reached, even if STATA hasn't given out such warning.
My questions are:
- is this approach valid for my problem?
- how to address the issue of convergence
- any alternative IPW approach
Thank you in advance for any comment on this,
Best wishes
Paolo
