I have 10 means and I want to test if there is a significant difference between the groups. The means are composed of cost values that can be negative and positive.

I read here the kruskal test can be used on "ratio" variables but everywhere I read like on this page: https://statistics.laerd.com/spss-tutorials/kruskal-wallis-h-test-using-spss-statistics.php

It seems they must be "positive" ratio scale. The example at that link says you can run the kruskal test on a variable 0-100.

Assumption #1: Your dependent variable should be measured at the ordinal or continuous level (i.e., interval or ratio).

...continuous variables include revision time (measured in hours), intelligence (measured using IQ score), exam performance (measured from 0 to 100),

But can you run Kruskal wallis on NEGATIVE and positive variables? Say each of my 10 variables have a range of ~ -5 to +10?

######## UPDATE:::: THEY defined "ratio" as

Ratio variables are interval variables, but with the added condition that 0 (zero) of the measurement indicates that there is none of that variable. So, temperature measured in degrees Celsius or Fahrenheit is not a ratio variable because 0C does not mean there is no temperature. However, temperature measured in Kelvin is a ratio variable as 0 Kelvin (often called absolute zero) indicates that there is no temperature whatsoever.

from this link https://statistics.laerd.com/statistical-guides/types-of-variable.php

In my case 0 means 0 cost so I can use Kruskal wallis.

  • 1
    $\begingroup$ What they actually say is "Your dependent variable should be measured at the ordinal or continuous level (i.e., interval or ratio)"... which means three options ordinal, interval and ratio. (They're wrong to imply interval and ratio are necessarily "continuous" though -- you can have discrete variables that are either, and if they think the KW applies to ordinal variables then it should certainly work just as well for discrete interval or discrete ratio variables.) $\endgroup$ – Glen_b -Reinstate Monica Jul 28 '16 at 2:09

In Kruskal-Wallis test only rank of observations matter, not values. Therefore, adding any constant to your values shouldn't affect the results of the test (e.g. using temperatures in C or K should give the same results).

And about the assumption you cited, please notice that it says "continuous level (i.e., interval or ratio)". That is, your dependent variable can be any continuous variable, and continuous variables are intervals and ratios. I think your question forgets the intervals part, because your variables with a range of ~ -5 to +10 can be seen as intervals according to https://statistics.laerd.com/statistical-guides/types-of-variable.php.

Therefore, I think that using negative values in dependent variable in a Kruskal-Wallis test doesn't conflict either with the citation you provided or the definition of the test.


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.