Chi-square test, Fischer's exact test failed. What are the alternatives to find out the relation between independent categorical variables?

correlation
python

#1

I am using python and it is not a 2X2 contingency table. My contingency table also has some ZERO frequency cells.

Regards
Akshay


#2

Hi,

If they are strictly categorical variables (not ordinal), one thing you can try is to one hot encode each category and then just have a look at the pearson correlation for each pair of new columns created. This should give you a fair idea of relation between each the categorical variables.

Regards
Ankit


#3

Hi Ankit,

Is this what you are referring to? - https://en.wikipedia.org/wiki/Pearson_correlation_coefficient. I don’t think this is useful for discrete data. Also, it gives a linear correlation which doesn’t make any sense. If that is what you are not referring to, please share the appropriate link.

Thanks
Akshay


#4

Fisher-Freeman-Halton test, an extension of the Fisher exact can be applied for contingency tables that are not 2x2.

Here


#5

Thanks @eagle. I came across the above link you shared but I am looking for a more custom solution contingency table of dimensions mXn. The FFH test is only for 3X3 tables.


#6

Hi @akshay.kotha

as you have n*r use log linear it will prove the independence.

Alain


#7

Hi @Lesaffrea,

Could you please elaborate on this?

Akshay