0
$\begingroup$

I made an experiment about UX where the participant has to detect among five differents configurations the one that he prefers. The configurations are slightly different and are presented randomly 3 times during the pre-test and 3 times during the post-test. I have two groups (Control and Experimental). The experimental group follows a training of 10 trials that aims to change perception about the configuration. I want to see if the training has an impact on participant's perception.

At the end, I have the repartition of choices for each configuration: enter image description here

I would like to know how I can compare the two groups (Control and Experimental) to know if the training had an impact on the choices they made.

Thanks for the help !

$\endgroup$
  • $\begingroup$ If nothing else you could use a generalized linear model. $\endgroup$ – user2974951 Jan 10 at 12:02
0
$\begingroup$

I recommend you consider using multivariate analysis to compare the two groups. You could compare the responses of the participants within the groups using PERMANOVA and then visually interpret whether the groups (control vs experimental) are meaningful using ordination. Both of these steps can be completed in the R-package vegan.

You would need to set up your data so that each participant is represented by a row. The columns would indicate the participate information (their ID, experimental or control, and the configuration they selected). You would have a column for each configuration and could represent the participant's selection with a 1 or a 0.

Hope that helps.

$\endgroup$
  • $\begingroup$ Thanks, it helps but I can't represent the participant's selection with a 1 or a 0 because each participant had to do 6 trials (for control group) and 16 trials (for experimental group) so I have to do the mean. $\endgroup$ – Julien Jan 8 at 23:52
  • $\begingroup$ Hi Julien, in this case, I would represent each trial as a separate row, then include trial as a column, therefore each participant has row 1, 2, 3, 4, 5, 6, etc. You may need to see if there is an 'repeated measures' type of permutation anova to avoid pseudo replication because each trail is not independent (they are the same participant). Alternatively, a mixed model that incorporates trial as a random effect might work. $\endgroup$ – bicyclerider Jan 9 at 23:05

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.