1
$\begingroup$

I made an experingment were 100 individuals were exposed to 10 different situations (imagine they are watching 10 different paintings). For each individual and painting we measured variables like body temperature and heart rate (bpm). See here an example of data structure: enter image description here

Imagine I want to predict individual nevrotism, based on people reactions to the different paintings. And I would like to design this as a Multilevel Linear Model with fixed or random effects (the mixed command in Stata).

I need help with the model design, as we have repeated measures on same people (not independent among them), but also reactions are driven by the painting content (so there is a dependence of variables inducted by paintings) How would you nest observations and interpret results? Would you make 2 or 3 levels? How?

UPDATE

Images are very different. Body response depends in part on the individual and in part on the image shown. So one image should slow the BPM and the other one should accelerate it, but still I want to use BPM to predict something else which is nevrotism.

$\endgroup$

1 Answer 1

1
$\begingroup$

If I understand your study design correctly, the only random effect you have is for individuals, so you would most likely want a random slope model across the three time period for each individual.

I don't believe you want to treat visual stimuli as a random effect. It is a condition within your experiment, and it seems like you would want to estimate that as a fixed effect. We expect individual to have some random variation, but we expect visual stimuli to have a systematic effect.

So, just 2 levels should be sufficient.

$\endgroup$
2
  • $\begingroup$ Thanks a lot. That was exactly my doubt, as indeed visual stimuli are very different. Imagine a painting is very scary and another depicts a love scene. In such case, should I also control somehow for the kind of visual stimuli? Body response depends in part on the individual and in part on the image shown. So one image should slow the BPM and the other one should accelerate it, but still I want to use bpm to predict something else, which is nevrotism. Hope it is clearer now, I will update the question. $\endgroup$ Oct 19, 2021 at 16:58
  • 1
    $\begingroup$ I would still include the stimuli as a fixed effect. Then, when you are interpreting you just focus on the responses you care about and treat stimuli as a "control" variable. You may hear the terminology "control variable" in regression sometimes. They are exactly the same as other regressors, we just chose not to care about their coefficients, just their inclusion. $\endgroup$ Oct 19, 2021 at 18:34

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.