I'using a dataset where a sample of 7 subjects edited 2 texts (as in corrected grammar issues, etc.). There were two conditions in the experiments related to the text correction software interface (P and PIO) -- I'm not interested in these conditions. All subjects saw all texts, but the combination condition x text varied -- see screenshot of the study design below. I'm interested in a general view of how measures relating to subjects' behaviour (time spent on correction, number of editing operations, etc. - there will be a few of these) correlate with each other. I'm thinking of doing a by-item PCA (items being sentences in each text, which are many) and doing a biplot with the first two PCs just to illustrate how close to each other different measures are. Would that be my best option? Would it be a problem doing a by-item PCA (i.e. using the average of all subjects per sentence) when the combination condition x text varied between subjects? The odd number of subjects means that there will be one combination which is more frequent. Should I maybe exclude one subject to guarantee that the comparison between the measures I'm interested in is based on the same number of text x condition combinations?

screenshot of df

  • $\begingroup$ Can you describe your data better? It's not clear to me whether there is an individual column vector for each assignment, or you measure multiple features per subject, whether the row vectors are sparse or you have two samples per subject. A screenshot of your datamatrix would be nice :) $\endgroup$ – Ulf Aslak Feb 7 '16 at 22:39
  • $\begingroup$ @UlfAslak I decided to use a slightly different dataset, but my question remains... see screenshot and a (hopefully clearer) explanation above. $\endgroup$ – user3744206 Feb 13 '16 at 12:42

Your Answer

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

Browse other questions tagged or ask your own question.