I've got a question, what if your expected count is violated, but your likelihood ratio p-value is statistically significant, is it still ok to look at Cramers V (in case of 2x3) to measure the effect size or not?
Hi Josephine. I would say it depends on how far off it is. You could make a judgement call. The expected counts could be violated by a small magnitude and the result of the test could still make 'sense'. As long as you explain everything in your report and allow for others to challenge your assumptions or judgements, you should be fine. Strictly speaking, expected counts must be met for the test to be statistically valid.
I've got a question, what if your expected count is violated, but your likelihood ratio p-value is statistically significant, is it still ok to look at Cramers V (in case of 2x3) to measure the effect size or not?
Hi Josephine. I would say it depends on how far off it is. You could make a judgement call. The expected counts could be violated by a small magnitude and the result of the test could still make 'sense'. As long as you explain everything in your report and allow for others to challenge your assumptions or judgements, you should be fine. Strictly speaking, expected counts must be met for the test to be statistically valid.