I am seeking advice on how to handle a specific type of systematically missing data (SMD) variable in a longitudinal survey dataset. From what I understand, SMD in survey-related data is often associated with specific values of preceding variables.
For example, in the Canadian context, a survey may ask participants if they identify as Indigenous. Those who answer “yes” are typically asked a follow-up question about whether they are First Nations, Inuit, or Métis. Non-Indigenous participants would have missing responses to this follow-up question, as they are deliberately not asked it. In this case, the relationship between the previous variable and the missingness is clear.
However, my situation is somewhat different. In the study I am working with, the survey is completed by patients at every hospitalization as part of a longitudinal study. A particular question was entirely removed from the survey partway through the study period. This creates three distinct situations:
Patients who have been in the study since before the question was removed, they have completed surveys both with and without the question.
Patients who participated only before the question was removed, they have complete data for this question.
Patients who joined after the question was removed, they have only missing data for this question.
In this case, the missingness is not related to a previously answered question but to the timing of the patient's entry into and participation in the study. Since the missingness is correlated with when the subject joined or left the study and the survey date, I am wondering whether it may be appropriate to treat this as missing at random (MAR).
Any advice on how to best classify and handle this type of missing data would be greatly appreciated.
