# Post Hoc for Significant Chi Square Test

I have applied Chi square test between Gender and Practice of specific norm (4 levels; Daily, Weekly, Monthly, Quarterly) to check the association.

Tabulated statistics: Gender, Practice

Rows: Gender   Columns: Practice

1   2   3   4  All

1      14  69  25   7  115
2       7  25  60  23  115
All    21  94  85  30  230

Cell Contents:      Count

Pearson Chi-Square = 47.187, DF = 3, P-Value = 0.000


My results are highly significant, now I want to examine that which has more influence for this significant association; Male/ Female. For this propose, which test is more appropriate?

• What kind of variable is Practice? Since you applied a chi square test, I assume it is categorical - if yes, how many levels are there? And what do you mean by "which is more associated"?
– Ute
Jul 22 at 16:02
• I agree with @ute. Also, questions about how to do things with software are off topic here. Jul 22 at 20:30
• I second @ute comments, "more associated" needs clarification, in particular when "practice" may be considered as an ordinal variable. By "more associated", do you mean the gender where the cells are making the greatest contribution to the chi-square test result? Or do you mean something along the lines of "Does being a woman increase the frequence of practice, compared to men"? These are different things. Maybe you should explain what your research question is, it would help answering you. Jul 23 at 8:36
• @J-J-J I want this "Does being a woman increase the frequency of practice, compared to men"?
– J AK
Jul 23 at 9:40
• Men ans women are, by definition, equally associated. The variable is "sex" and the question asked by chi-square is whether sex is associated with practice. If your question is, rather "do men or women practice more" then I would do ordinal regression, with practice as the DV and sex as the IV. Jul 23 at 11:03

I don't know if you already got what you need from the above discussion, but you can run chi-square post hoc tests in R using, for instance chisq.posthoc.test, like this

t              #make your count table an R object, let's call it t
1  2  3  4
F 14 69 25  7  #I assumed top row is women and bottom row is men
M  7 25 60 23

library(chisq.posthoc.test)

chisq.posthoc.test(t, method="holm")

Dimension     Value         1         2         3         4
1         F Residuals  1.602430  5.901783 -4.781217 -3.132624
2         F  p values  0.545302  0.000000  0.000012  0.010395
3         M Residuals -1.602430 -5.901783  4.781217  3.132624
4         M  p values  0.545302  0.000000  0.000012  0.010395

#you can choose the correction method from the p.adjust methods


So, the post hoc test suggests that with p < .05 as the critical value there are significantly more women in Practice=2 and significantly more men in Practice=3 and Practice=4. No significant gender differences are found for Practice=1 group (if top row is men and bottom row women, the results for Practice 2-4 are of course the opposite to the above).

• Thank you, @Sointu, great - I think this is actually what OP meant (+1) : "Ute yes I want know that contribution for example: male at different levels of practice. I conclude that Gender n paractice are associated but female practice daily, weekly or monthly "more" "
– Ute
Jul 28 at 18:09