Negative Cramer's V

[Pallavi Ch]

Hi all,

Can someone please guide me with interpreting ''negative Cramer's V''
I am trying to find the association between the independent variable (gender) and dependent (purchase behaviour (PQ_new) and I run the following code

tabulate PQ_new gender_1_0, chi2 row V

PQ_new	gender
Male	Female	Total
No	30	87	117
	25.64	74.36	100.00
Yes	111	210	321
	34.58	65.42	100.00
Total	141	297	438
	32.19	67.81	100.00
Pearson Chi2 = 3.14 Prob = 0.0765

Cramér's V = -0.0846

A normal Cramer's V ranges between 0 and 1. My concern is this negative value. How to interpret these results? From chi2 value it is clear that gender and purchasing behaviour are not associated.

It would be great if someone could help me.

Thank you

[Clyde Schechter]

I get the same wrong result you are getting:

. tabi 30 87 \ 111 210, chi2 V

           |          col
       row |         1          2 |     Total
-----------+----------------------+----------
         1 |        30         87 |       117
         2 |       111        210 |       321
-----------+----------------------+----------
     Total |       141        297 |       438

          Pearson chi2(1) =   3.1384   Pr = 0.076
               Cramér's V =  -0.0846

. display sqrt(r(chi2)/438)
.08464828

The calculation that I gave after the tabulation shows that the correct value of Cramer's V for this table is +.08464828. So somehow Stata is getting the sign wrong here.

You should report this to technical support; it is clearly a bug.

[Pallavi Ch]

THanks a ton!. Discussed with supervisor and concluded.

[Clyde Schechter]

I interpreted your response in #3 as meaning that you did not notify Stata technical support. So I did. They responded by pointing out that in the PDF documentation that comes installed with Stata, they do state that for 2x2 tables, -1 <= V <= 1. And they explained in their response that they return a negative V when the odds ratio in the table is < 1. While this is not, to my knowledge, the conventional way Cramer's V is handled, it is reasonable enough in its own right, and for those of us for whom V should always be non-negative, it is clear that all we need do is take the absolute value of Stata's result.

[Mike Lacy]

I'd also add that Cramer's V is no longer generally regarded as a very useful measure, particularly not for 2 X 2 tables, where an odds ratio, percentage difference, gamma, etc., would be more commonly used. With all due respect to your supervisor, V is quite out of date. (I say this as a person who is himself chronologically stuck in the 20th century. <grin>)