0

I am testing the hypothesis, "there is no difference between 3 materials", with a small sample size. I have 3 materials and 2 DVs (test 1-2). They also have a different number of replications as seen below. Six material-1, two material-2, and two material-3. Unfortunately this is all of the data that I will get due to the difficulty of the tests.

Material    Test1    Test2
________    _____    _____

'mat1' 5 3
'mat1' 4 7
'mat1' 8 7
'mat1' 8 2
'mat1' 2 2
'mat1' 5 5
'mat2' 5 10
'mat2' 7 4
'mat3' 8 6
'mat3' 8 3

I am using Matlab to run a MANOVA analysis using the repeated measures function with 95% confidence. The output from the analysis is shown below;

Within       Between      Statistic     Value        F       RSquare    df1    df2    pValue 
________    ___________    _________    _______    _______    _______    ___    ___    _______

Constant (Intercept) Pillai 0.11538 0.91304 0.11538 1 7 0.37113 Constant (Intercept) Wilks 0.88462 0.91304 0.11538 1 7 0.37113 Constant (Intercept) Hotelling 0.13043 0.91304 0.11538 1 7 0.37113 Constant (Intercept) Roy 0.13043 0.91304 0.11538 1 7 0.37113 Constant Material Pillai 0.20218 0.88696 0.20218 2 7 0.45359 Constant Material Wilks 0.79782 0.88696 0.20218 2 7 0.45359 Constant Material Hotelling 0.25342 0.88696 0.20218 2 7 0.45359 Constant Material Roy 0.25342 0.88696 0.20218 2 7 0.45359

It seems to clearly show that the null hypothesis holds true. I want to make sure that I went through the process correctly and I used the "best" method for the data I have. Should I do separate t-tests for the materials because of the small sample size? Is there some post-hoc analysis I should be doing? I am giving myself a crash course in statistics over the last few weeks, so any help is greatly appreciated.

Lino
  • 1
  • Do you really mean that there are 3 inndependant variables with an unequal number of repeated measures, or is this a single predictor ("material") with 3 levels, and an unequal number of independant observations? – chl Oct 22 '20 at 16:03
  • @chl A single predictor with 3 levels. I updated the post. – Lino Oct 22 '20 at 17:08
  • Ok, I got close results using R or Stata with their respective manova function. You cannot reject the null in this case. However, the correlation between your two outcomes is zero; combined with the low sample size, it is diffcult to conclude. – chl Oct 22 '20 at 18:21
  • @chl Understood. Thank you for your help. – Lino Oct 22 '20 at 19:33

0 Answers0