I am working with two variables - variable 'A' is an independent categorical variable and has three levels 'a' 'b' and 'c'; variable 'B' is continuous response variable data I have classified into five (ordinal) bins/levels '0' '1' '2' '3' '4' and '5'. The data are not normally distributed.

My hypothesis is that membership to variable 'A' does not determine performance on variable 'B'.

I have conducted a chi square test and there is a significant relationship between the two variables i.e. the hypothesis is rejected. However. I want to know the relationships between the different levels i.e. is the relationship between between 'a' and '5' significant? Is the relationship between 'b' and '5' significant?

I have looked at multinomial regression, as well as log-linear analysis, nut I am not sure which one will provide the information I am after.

  • $\begingroup$ Why not keep B as continuous variable and then compare with different levels of A ? Why do you have to convert B to an ordinal variable? $\endgroup$ – rnso Apr 7 '15 at 11:44
  • $\begingroup$ If B remains continuous I would use Kruskal-Wallis to compare means, however, the mean is not a good measure given data are highly skewed (pareto distribution) - the bins are used to deal with this. $\endgroup$ – user1222447 Apr 7 '15 at 12:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.