3
$\begingroup$

I have two groups, with 100 people in one group(1) and 80 people in the other group(2). 42% of group 1 love cake. I am wondering what proportion of the 80 group 2 members would need to love cake for me to have 80% power to reject the null hypothesis that both of these groups have the same proportion of cake lovers (I hope I worded that right).

I did a power analysis in g-power (z test of proportions, proportion 1=42%, alphpa=.05, power=80%), and it suggested that about 63% of people in group 2 would need to be cake lovers for me to have 80% power to have statistically significant findings.

I was curious and wanted to see how many participants I would need to do a chi-squares test. I needed an effect size so I imputed my proposed proportions (41%, 63%), which appears to be a w value of about .43 (effect size). When I put this in the power analysis in g-power (chi-square goodness of fit), it indicated that I would need 70 people to have 80% power to detect a statistically significant difference of .43 magnitude.

This confused me. On the one hand I have 180 people and apparently I need proportions resembling 42%/63% to have 80% power. On the other hand, these proportions seem to reflect a w effect size of .43, which requires only 52 people to have 80% power.

What explains this difference? Does it have to do with the fact that on one hand I am trying to determine a proportion, and on the other hand I am trying to determine required sample size? These seem to be different questions, but I can't wrap my head around why the findings differ.

$\endgroup$
0
$\begingroup$

The reason you're getting a difference is the corresponding chi-square to a two-sample z-test of proportions is not a goodness of fit, but a test of homogeneity of proportions; specifically, you should get the same results if you condition on the margins (as a chi-square test of independence) as long as you treat both tests the same with respect to any continuity correction (either doing one for both or for neither).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.