1
$\begingroup$

I have ratio scale data for 3 related samples. What is the statistical test that I can use to check whether there's any significant difference between the 3 groups? (I tried Friedman, but I'm doubtful whether it can be used for ratio scale data.)

https://statistics.laerd.com/spss-tutorials/friedman-test-using-spss-statistics.php

This cite says Friedman test can be used for continuous scale data as well, but I need to make sure about it as this is the only place I found that fact. In every other place, it speaks about ratio scale data.

NOTE: I collected data from 15 people. i.e. sample size is 15. The same 15 people faced the 3 different situations where we need to compare

|cite|improve this question
$\endgroup$
  • $\begingroup$ related = dependent samples $\endgroup$ – Dovini Jayasinghe Oct 1 at 16:28
2
$\begingroup$

In a comment you clarifies

collected data from 15 people. i.e. sample size is 15. The same 15 people faced the 3 different situations where we need to compare

that is information that really needs to be in the post itself! please edit. So one simple model for your situation is $$ y_{ij}=\mu+\tau_i+\beta_j+\epsilon_{ij} $$ with $i=1,2,\dotsc,15$ indexing the persons and $\beta_j, j=1,2,3$ indexing the groups. As written this is a mixed model and could be analyzed as such, or traditionally as repeated measurement ANOVA. If you use R, you could start with something like

lme4::lmer(y ~ groups + (1 | person), data=your_data_frame)

with data in long format. Also search this site for repeated measures, and read up on mixed models!

|cite|improve this answer
$\endgroup$
  • 1
    $\begingroup$ Thanks a lot. I did the edit as you requested. Mixed models are new to me. Will check further on it $\endgroup$ – Dovini Jayasinghe Oct 2 at 17:57
0
$\begingroup$

Since your data is a ratio scale data why don't you try to use ANOVA for testing. It is a parametric test as compared to the non-parametric Friedman.

|cite|improve this answer
$\endgroup$
  • $\begingroup$ If OP has data collected according to a block design, Friedman might be OK. But you would have to specify carefully what kind of ANOVA to use. $\endgroup$ – BruceET Oct 1 at 2:18
  • $\begingroup$ Cannot use ANOVA, since the samples are not independent. They are related (dependent samples) $\endgroup$ – Dovini Jayasinghe Oct 1 at 16:32
  • $\begingroup$ If you really want to use the Friedman test, then look at the example at the end of the R documentation for that test. You have a design just like the baseball example, except you have 15 subj as 'blocks' (not 22 players) and you have three methods to compare. The hypothesis test will compare your 3 groups. // However, if your data have nearly normal residuals, you can use the ANOVA model suggested by @Kjetilbhalversen (+1). Maybe make a normal probability plot of residuals at the end to verify their near normality. $\endgroup$ – BruceET Oct 2 at 0:08
  • $\begingroup$ Is normality the only requirement for ANOVA? Don't the data should be independent? $\endgroup$ – Dovini Jayasinghe Oct 2 at 17:53
  • $\begingroup$ sites.ualberta.ca/~lkgray/uploads/7/3/6/2/7362679/… $\endgroup$ – Dovini Jayasinghe Oct 2 at 17:54

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.