We normally interpret chi square test with p value and CI ... however can we use Cramer V test in measuring strength of association while using chi square test and highlight it in our study
Yes, it is useful to report both the p-value from the hypothesis test and Cramer's V as an effect size statistic. They convey different kinds of information. ... Out of curiosity, though, when you mention confidence interval, the confidence interval for what parameter ?
That makes sense. The difference in proportions could be used in lieu of something like Cramer's V. It depends what is most helpful for the reader. And it can be helpful to include as much information as possible. Also, it's a good idea to include the confidence interval for Cramer's V. Just performing a quick google search, it appears you can do so by bootstrap (although there are some caveats with that method), or there is a method using Fisher's z that I'm not familiar with.
Suraj Kapoor , differences in the proportions of what? Since you mention Cramer's V rather than the large number of effect sizes discuss with respect to 2x2 tables, I assume you have df>1 in your chi-sq. Do you want differences for all pairwise comparisons of proportions? This will be a lot if df >> 1.
Good Evening Sir , What i know Cramer's phi can be used to measure strength of association when applying Chi-square test for independence in 2x2 table and if it is more than 2x2 table Cramer's V is used with interpretation as small(0.1), medium(0.3), Large(0.5)
Suraj Kapoor , I will repeat my question. Assuming df > 1, differences in proportions of what?
Difference in proportion of the categorical variables like comparison prevalence of smoking based on gender
I have a question. If chi square significance value is more than 0.05, that will mean Null hypothesis is accepted :there is no association between the two variables.
In this case what is the significance of Cramer V Statistics Value 0.2. Please guide
Veena Chavan , I think your confusion stems from a misunderstanding about hypothesis tests.
a) Technically, we never accept the null hypothesis. We just conclude that we don't have sufficient evidence to reject the null hypothesis given the data. Often the phrase is "fail to reject" the null hypothesis.
b) The null hypothesis is about the (unseen) populations we are sampling. Our data is only a sample of these populations.
c) So in this case, our conclusion is that we don't have sufficient evidence with this data to conclude that the association of the two variables in our (unseen) population is not zero.
c2) The true association of the whole population may be greater than zero, but we haven't sample enough of the population to have confidence in this.
d) Cramer's V is a point estimate of this association, based on the data we have. If we re-sampled our population, it's very likely that our Cramer's V estimate would be different. What's helpful is to construct a confidence interval of this estimate. If all things work perfectly, if the p-value for the test of association is > 0.05, then the 95% confidence interval for Cramer's V would include 0. (Practically speaking, depending on methods and assumptions, the confidence interval for the effect size statistic and the hypothesis test may not line up exactly.)
Bec Joyce ,
1) Where did you find this table ?
2) See Table 7.2.3 in Cohen 1988 for converting his interpretations for w to Cramer's V for various values of r, related to "degrees of freedom".
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS