What do I do when the chi-square assumption is violated


I am looking at data to see if there is a correlation between an Independent nominal variable (3 level) and a hypothesized dependent nominal variable (2 levels).

SPSS shows a statistically significant chi square and Cramer's V, but an assumption is violated. The expected count is less than 5 for 33% of the cells.

What do I do? I tried looking this up and I found sources saying to use a Fisher's test if it was 2x2, but my data is 3x2. :\

The likelihood ratio value calculated in the same SPSS chi square test is 9.676 with df=2 p value= 0.008

What does the likelihood ratio mean in this case? and is it appropriate to use it in the analysis?


Less is more. Stay pure. Stay poor.
Use Fisher's Exact Test. It may take a little time to run.

Can you post your 2x3 table? The likelihood would be easier to explain with a numeric example.