2
$\begingroup$

I would like to predict whether single Reaction-Times (RTs) in a computer task can predict subject performance on a comprehension task. For each subject, I have many RTs but only 1 comprehension score.

Example data structure (my_data):

Subject_ID Trial RT Comperhension_Score
P001 1 250 20
P001 2 260 20
P001 3 245 20
... ... ... ...
P005 1 255 31
P005 2 270 31
P005 3 280 31

I'm using nlme, my model is my_model <- lme(Comprehension_Score ~ RT, random = (~1|Subject_ID), data = my_data)

I want to analyse at the trial-level because my understanding is that this preserves information and can improve statistical power, but is it valid to repeat Comprehension_score participant-wise and run this model? Or should you always reduce the data so that all the measures are on the same scale.

Edit: Alternatively, is there a way to modify the model (e.g., adjusting random effects) to make it valid?

Thank you!

Improve this question
$\endgroup$
4
  • $\begingroup$ How many trials did each participant complete? And how many participants do you have? $\endgroup$
    – awhug
    Commented Jan 22, 2022 at 1:32
  • $\begingroup$ @awhug There are ~45 reaction times per participant, and there are ~30 participants. Although I am attempting to trial-level analyses to improve power, my question is more about the validity of repeating participant-level scores in a trial-level analyses. $\endgroup$
    – Ang
    Commented Jan 24, 2022 at 1:02
  • $\begingroup$ Interesting question! My non-expert intuition is that your model formulation is not correct. Allowing Comprehension_score to differ between participants with a random intercept would absorb all of the outcome variance (as 100% of the variation is between participants). If it makes sense in your experiment, could you swap the predictor and outcome in the model? This would be a more classic mixed model. Hope someone more knowledgeable can help! $\endgroup$
    – Lachlan
    Commented Jan 28, 2022 at 3:29
  • $\begingroup$ @Lachlan Thank you for your input! Your non-expert intuition makes sense to me. I hadn't considered that the random effect might absorb the variance of Comprehension_score. Since there is only a single predictor, I think I can swap the IV and DV. Regarding my OP, I think I will compute subject-wise mean RTs and run a linear model for now. $\endgroup$
    – Ang
    Commented Jan 31, 2022 at 0:40

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.