4

my students were involved in three different simulations. I was asked them to rate their satisfaction with these simulations on a 5-point Likert scale (several items). Unfortunately, I have 60 questionnaires for the first simulation, 57 for the second and 33 for the third simulation (reason- winter, and they didn't come). I plan to make the composite variable (non-normaly distributed) for each simulations and to compare them with Friedman test in order to conclude is there difference between the level of satisfaction of students with the simulations. But, I have a problem with unequal samples sizes. Can I use the Friedman test in this situation i.e. random exclude questionnaires down to 33? Or I can use some another test? Thanks a lot in advance. I'm not an expert in statistics and any help will be great,

Lena

Lena
  • 41
  • 1
  • 2
  • 2
    Since your design is that 60 students would answer 3 questionnaires (repeated measures) but some didn't go to all of the sessions, I'd probably have described this as having missing values rather than different sample sizes. (But maybe that's just me.) – Glen_b Jul 22 '15 at 01:26
  • Dear Glen, thank you very much for your comment. Can you give me any idea what to do? Whether it makes sense to use Friedman? – Lena Jul 22 '15 at 09:57
  • If I had a good answer for you yet, you'd have it already. – Glen_b Jul 22 '15 at 10:20
  • 1
    Friedman test is designed for complete observations. Since you think that drop outs are due to winter (and not e.g. because they were unsatisfied with the last simulation), you could (A) run Friedman on the complete observations and (B) use a nonparametric repeated measures ANOVA à la Brunner-Langer: "Brunner, E., Domhof, S., and Langer, F. (2002). Nonparametric Analysis of Longitudinal Data in Factorial Experiments, Wiley, New York." on all values. And (C) check if the results are in line with each other. – Michael M Jul 22 '15 at 10:28
  • Dear Michael, thank you very, very much. Your comment is very helpful for me. I will probably use option A, even that result in 33 respondents. Of course, I'm going to read mentioned article in option B, now... – Lena Jul 22 '15 at 10:44
  • @MichaelM Would Kruskal-Wallis fit option (B) despite considering one factor instead of two? – Jabro Oct 20 '20 at 12:23

0 Answers0