I wonder if it is correct to conduct a mediation test in the following case:

1) Independent variable is anxiety score measured at time 1

2) Dependent variable is cocaine consumption measured at time 2

3) Mediator variable would be anxiety score measured at time 2.

Would a "classical" mediation model be correct (apart from the problem that the outcome is binary whereas the mediator would be continuous)? Or the fact that the independent variable and the mediator are the same (although measured at two different times) precludes this analysis?


Assuming you used "classical" to refer to the Baron and Kenny's approach or simple meditation with x--->M, M---->Y and x--->y, without taking the effect of t (time); an outcome being binary and mediator a continuous, is never been an issue. Neither an outcome being measured twice is also a problem.

I think the most important question is "Why do you want to do the mediation? " If you believe being anxious at t1 causes more or less anxiousness at t2 which in turn increases or decreases cocaine consumption at t2, then you can do it.

Plus Baron and Kenny's approach is limited in many ways and has been already decades since it has been recommended to use the alternative approaches.

  • $\begingroup$ Concerning the first part, I use the khb method, which is not a problem for a binary outcome. For the second part, the main question was whether the effect of anxiety at t1 on the risk of cocaine use at t2 was partly or completely due to anxiety at t2 (to somehow distinguish the magnitude of the trait or state effect). However, I can't find any study that adopts this approach, which makes me think that there is a problem (because of the similarity of the predictor and the mediator). $\endgroup$ – Emmanuel.W May 17 '18 at 14:23
  • $\begingroup$ predictor and the mediator can be similar if it is a repeated measurement. because "time" plays a huge role. Anxiety at t1 is not the same as anxiety at t2 because someone with higher anxiety at t1 might have more risk of having higher anxiety at t2. Your challenge might be the prerequisite to “have the maximum possible separation between mediator and outcome measurements” because you measured cocaine use and mediator at the same time (t2). $\endgroup$ – Million May 17 '18 at 15:49
  • $\begingroup$ I find your answer interesting. However, I am surprised not to find any articles that use this method while data must be commonly available. $\endgroup$ – Emmanuel.W May 28 '18 at 13:11
  • $\begingroup$ Does your study invlolves expriment? $\endgroup$ – Million May 29 '18 at 8:39
  • $\begingroup$ No, it is a "purely" observational study. $\endgroup$ – Emmanuel.W Jun 1 '18 at 11:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.