Skip to content

friedman.test error: not an unreplicated complete block design

5 messages · lindsay hanford, John Kane, Michael Dewey +1 more

lindsay hanford
# 
Hello R Community,

I am using the friedman.test() function to test differences in a non-normally
distributed dataset, with a dependent variable that either a
continuous variable or a ratio and has 2+ groups.

I am using the friedman.test instead of a repeated measures ANOVA because
my dataset violated the assumptions for using an ANOVA. I am looking to
compare response means on an emotion-labelling task, between groups (HR,
HC) and emotions (Happy, Sad, Angry, Fearful) where these variables are my
group and block variables, respectively.

When I use the following command:
I get the following error:
Error in friedman.test.default(c(1L, 1L, 0L, 0L, 0L, 1L, 0L, 2L, 0L, 0L,  :
  not an unreplicated complete block design

I believe I have set up my dataset correctly.. where Subject ID is repeated for
the four categories of emotion. The variable Error contains the number of
incorrect response corresponding to each emotion.


*Subj  Group Emotion Response*94    HR    Happy  2
119   HC   Happy 0
....
3   HR   Sad   4
61 HC   Sad   2
64  HC  Sad   0
....etc

I think the error c(1L, 1L, 0L, 0L, 0L, 1L, 0L, 2L, 0L, 0L,... ) corresponds
to my Response variable and might not be happy about is the number of 0's that
appear in that variable. However, this is the reason my dataset is not normally
distributed and I cannot use rmANOVA.

Any ideas how to deal with this error? Or whether I should be using a
different statistical test?
Thanks,

Lindsay
John Kane
# 
We really need " commented, minimal, self-contained, reproducible code' as asked for in the note at the end of each R-help message.

Have a look at http://adv-r.had.co.nz/Reproducibility.html and/or http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example  for some hints.

In particular, in your case we almost certainly need some data.  Please use dput() to produce a useable data set. See Hakley's discussion at http://adv-r.had.co.nz/Reproducibility.html for an example of how to to this.
John Kane
Kingston ON Canada
____________________________________________________________
Can't remember your password? Do you need a strong and secure password?
Use Password manager! It stores your passwords & protects your account.
Michael Dewey
# 
Dear Lindsay
If the problem is that you have an excess of zeroes you might look at 
the vignette for the package pscl which is called something like 
Regression models for count data.
On 13/04/2015 17:17, lindsay hanford wrote:

  
    
  


  


      

    

    

      
      

        


  

        

      Peter Dalgaard
    

    

      
      #
    

  


  

      
By coincidence, there actually _is_ enough info to pinpoint the issue:


*Subj  Group Emotion Response*94    HR    Happy  2
119   HC   Happy 0
....
3   HR   Sad   4
61 HC   Sad   2
64  HC  Sad   0
....etc

An unreplicated complete block design has exactly 1 observation for each combination of the two grouping factors. The above clearly has 2 observations with "HC, Sad". So Friedman's test does not apply.

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      lindsay hanford
    

    

      
      #
    

  


  

      
Hi Peter,

I thought it was possible to have multiple subjects in each group, while
still specifying another factor, as is the case for repeated measures
ANOVA. If this is not the case, I guess I should look into Regression
models for count data as suggested by Micheal.

        
On Mon, Apr 13, 2015 at 1:13 PM, peter dalgaard <pdalgd at gmail.com> wrote: