If I have a panel data survey that collect 100 patients' EQ5D scores by app for their health. One patient can submit their scores more than one times as time passing by. The time in this model is defined as the interval between the data they register the app and the data the submit their scores. Some patients may fill this survey several times but some of them may only fill it one to two times. Thus, we could view the latter as drop out if we want to study the trajectory of the EQ5D scores. In this situation, how to use IPW to weight the sample, as early drop out may be a bias in the model because they feel better thus do not want to continue to record their quality of life.
why are you thinking on weighing instead of an indicator variable for censored data (observations you know they have attrition)?
I would not know how to weight observations that only report their perception just once or twice relative to others more responsive.... If you weight by the number of times they answered you are assuming that dropping out if relatively random I believe.
Applying a censored model you are using estimators that take into account that participants who drop out the sample are not similar to the ones that remain
- Badges
- Science topic