cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
JMP is taking Discovery online, April 16 and 18. Register today and join us for interactive sessions featuring popular presentation topics, networking, and discussions with the experts.
Choose Language Hide Translation Bar

Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

Hello everyone, I am interested in using the Kruskal-Wallis test to compare several different categories. I have data from two years, and I would like to block on year to incorporate this variation. However, I do not have equal sample sizes from both years. Is it possible to block on year in JMP (or any other software) if you have unequal sample sizes for nonparametric tests? JMP tells me 'blocking was ignored because cell counts were unequal'. Is there a workaround for this?

Thanks!

2 ACCEPTED SOLUTIONS

Accepted Solutions
Jeff_Perkinson
Community Manager Community Manager

Re: Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

There's no real workaround to allow for the unequal cell counts in Fit Y by X.

 

An alternative analysis for your situation is to save the ranks, using averages for ties from the Distribution platform. Then use Fit Model with the Rank as your Y and your categories and year as effects. It's not exactly the same as the Kruskal-Wallis but some statisticians would recognize this as a valid alternative.

 

Hope that helps.

-Jeff

View solution in original post

Re: Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

In JMP 16 and JMP 17, the Fit Y by X Platform will now automatically invoke an instance of Fit Model within the data analysis output window to provide an option for the analysis of unequal sample sizes within Block.  It will also give the user a warning message to prompt them why the additional output outside of Fit Y by X has been generated:  "Fit Model invoked because unbalanced block sizes not supported by Oneway features."

 

View solution in original post

3 REPLIES 3
Jeff_Perkinson
Community Manager Community Manager

Re: Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

There's no real workaround to allow for the unequal cell counts in Fit Y by X.

 

An alternative analysis for your situation is to save the ranks, using averages for ties from the Distribution platform. Then use Fit Model with the Rank as your Y and your categories and year as effects. It's not exactly the same as the Kruskal-Wallis but some statisticians would recognize this as a valid alternative.

 

Hope that helps.

-Jeff

Re: Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

Aside: please note that in recent versions of JMP (JMP 16 up through JMP 17.0.0), the option to test for Unequal Variances when a Block is specified (with equal sample size within) has been removed (grayed out) in the Fit Y by X (Fit Group) platform.This is because the variances were not estimated correctly to account for the presence of the Block effect. 

 

Please feel free to engage with the JMP Wishlist here on the User Community if you are interested in seeing the Unequal Variances test option added back to this platform (for equal sample sizes within Block) in a future release of JMP:  https://community.jmp.com/t5/JMP-Wish-List/Unequal-Variances-test-option-when-a-Block-effect-is-spec...

 

Re: Is it possible to include a blocking variable with unequal sample sizes for nonparametric tests?

In JMP 16 and JMP 17, the Fit Y by X Platform will now automatically invoke an instance of Fit Model within the data analysis output window to provide an option for the analysis of unequal sample sizes within Block.  It will also give the user a warning message to prompt them why the additional output outside of Fit Y by X has been generated:  "Fit Model invoked because unbalanced block sizes not supported by Oneway features."