My studies aim to evaluate how psychological constructs (type of motivation, habit, self-efficacy, intention) interact to predict physical activity maintenance (weekly frequency). Some participants get temporarily injured at a time wave but come back for subsequent questionnaires where they are apparently healed and back to their usual exercise level.

I would like to know the optimal procedure to control this confound without sacrificing too much data.

[EDIT 22/01/14] After discussion, my problem boils down to the following dilemma:

1) Should I keep all the data in the analyses and statistically control for the influence of injuries with a set of binary variables (not injured/injured) for each time wave (/4)

2) Should I delete only the behavioral data at the time wave where the injury occured and impute it using psychological data from the same time wave that is unaffected by injuries (i.e. type of motivation)?

3) Should I delete all data from this time wave, as if the injured participant was only absent and impute it with data from other waves?

4) Is there anything that prevents me from using data from subsequent time waves if the participant reports not being injured anymore?

My goal is not to test hypotheses regarding the onset of injuries or their effect on behavioral or psychological variables but to control them to obtain unbiased predition coefficients of psychological constructs on exercise frequency, as well as test mediations and moderations.

A secondary problem where I need validation is whether or not I should keep active participants that report dealing with a sports injury at baseline given that their exercise frequency appears to be unaffected. On the other hand I would exclude inactive participants that report being injured (sports or medical condition) at baseline unless they show improvements at the second wave of measurement.

Being guided through the rules and steps of data cleanup and imputation in regard to those temporarily injured participants would be a life-saver.

--- Any advice, general or specific to the scenarios below, would be welcome. I don't know if someone has written on the topic so I would be grateful for documentation on it.----

Thank you very much for your help!

-Pier