+ Reply to Thread
Results 1 to 1 of 1

Thread: Correlations computed within subject vs collapsing across subjects

  1. #1
    Points: 6,200, Level: 51
    Level completed: 25%, Points required for next Level: 150

    Location
    UK
    Posts
    53
    Thanks
    13
    Thanked 0 Times in 0 Posts

    Correlations computed within subject vs collapsing across subjects




    Hi everyone,

    I have model predictions (x) that I want to test against subject ratings (y). There are 20 stimuli per subject and I have data from 30 subjects. What I did initially was to collapse all (x,y) pairs of values and compute the Pearson correlation for the 30x20=600-element vectors x and y.

    However, given the variance in the ratings is lower within- than between-subjects, I am wondering whether I should in fact first be computing the model-rating correlations subject-wise, and then report the confidence interval corresponding to the distribution of subject-wise correlation coefficients?

    Possibly relevant is the fact that both x and y have discrete values, with 20 and 7 values respectively, i.e. 20 model predictions for the stimuli, and a 7-step rating scale (see scatter plot below).


    I am of course open to suggestions in case a Pearson correlation is in fact not the most appopriate measure of assessing the model's prediction accuracy across subjects.

    Many thanks for any help.
    Last edited by wildetudor; 07-02-2016 at 05:39 AM.

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats