- Thread starter mathprof
- Start date
- Tags chi square sample size

Hi Acer,

Chi-square test isn't only goodness of fit.

In the goodness of fit test: expected value > 5.

Maybe the total number of cells multiplied by 5 is good for approximation, but calculating the power based on the required effect size gives accurate results.

Chi-square test isn't only goodness of fit.

In the goodness of fit test: expected value > 5.

Maybe the total number of cells multiplied by 5 is good for approximation, but calculating the power based on the required effect size gives accurate results.

Grande was describing chi-square test of independence when he explained the rules. I assume they apply only to that test. Seems goodness-of-fit tests usually do not have enough contingency cells to apply those rules.

The goodness of fit can be used to test for independence

I assume Grande describe correctly the rule of thumb. I learned that all cells must have expected value > 5. but 80% is probably also okay. at least easier to meet...(more than one rule of thumb)

In order to calculate the sample size, you need to determine the test's power (usually 0.8, say the probability of 0.2 not to reject incorrect H0 )

and the power depends on the effect that the test needs to recognize. the smaller the effect the weaker the power and you need to increase the sample size to meet the required power