How do I select participants in study results analysis? All participants or only the ones who have answered all questionnaires? I have data on a student training for my master thesis. Variables were measured across 3 time points (before, directly after and 1 year after the training).
Some students have filled out all 3 questionnaires while others have only filled out questionnaires at 1 or 2 time points. 
I am trying to calculate a growth curve model in R fot one variable (entrepreneurial action). 
Should I include all observations? Or only the participants who have filled out all 3 questionnaires? (Why?)
 A: I don't think there's a wrong or right answer, and rather, I think the proper approach here, especially for a thesis, is to explain why you're doing whatever you choose to do, and may be include the other analysis in an appendix or mention it in a footnote. 
To highlight what could go wrong in either case, consider the following case. Suppose that only students who succeeded in their entrepreneurial endeavors, and are thus motivated to be more and more entrepreneurial, answer the followup questionnaires. Thus, by only looking at respondents who answer all surveys, you're selecting into the population of respondents who have higher personal 'growth rates' for entrepreneurial action. The growth rate you measure is correct, but for that population of individuals. However, by including all observations, the growth rate will look ever higher than just looking at the ones who respond to all 3, because first time period will be full of non-entrepreneurial people who then drop out.
It should be clear you can come up with any other hypothetical situation, and each will have different conclusions. But at the core, the trade-off is between measuring growth rate that could be looking at different populations over time, or looking at one with the same population, but with potentially heavy selection bias. Typically, I think the latter is 'cleaner' in effect interpretation, but then the goal is to really understand which population selects into this. The ideal is that drop out is just random, and has nothing to do with entrepreneurial action, in which case looking at the sub-sample of respondents who complete all is fine. But it should be clear that this is rarely true: those who respond to all surveys are very different people to those who don't (maybe more organized, maybe more caring, etc.. all of which could affect entrepreneurial action in any way).
I'm sure there's lots on follow-up survey nonresponse, but some classic ways to better understand this are: 1. do you have other baseline covariates (gender, race, gpa,...) that you can study over time and see if composition of population for those change over time? If they all don't, then maybe those who dropped out just forgot, and they randomly forget regardless of entrepreneurialness. Then it's maybe okay to look at the subsample of those who respond to all. 2. Is the difference in first question mean for those who answer all three versus those who don't statistically insignificant? This is another hint that maybe the groups are the same, though again, time evolution could matter here. 3. Can you try to get follow ups for non respondents? Maybe incentivize it somehow?
In the end, I think showing you carefully studied this issue is the most important here. Whichever you use, support it (definitely use points 1 and 2 if you look at the subsample of all respondents), and include some robustness in an appendix. 
