Power analysis for spearman rho

[Joe Tuckles]

Hi, I want to conduct a power analysis for spearman, and reading some instructions I have been asked to input a power for sample size estimation value. The value must be a single value between 0 and 1. Does this mean the rho value I would consider a good finding such as >0.7? Thank you.

[Ken Chui]

Power is the chance of rejecting the null hypothesis when the alternative hypothesis is true. That is not the same as the detectable correlation coefficient, which is considered as part of the "effect size." Assuming we're using -power onecorr-, to estimate the sample size, the basic syntax chart is:

power onecorrelation r0 ra [, power(numlist) options]

r0 is the null hypothesis correlation coefficient (which the question didn't say so I am not assuming. Usually it's 0, but it does not have to be); and ra is the alternative hypothesis correlation coefficient. The 0.7 you mentioned in post #1 goes to the "ra". For power, there isn't a rule, but generally in behavioral and biomedical sciences, 80% (0.8) or above is the convention when computing sample sizes; it's also the default used by Stata if you omit that power() option. So if you don't know what power to aim for, see the definition and come up with a number, but try not to go below 0.8 or the results may raise some eyebrows and questions triggered.

Also, check out the 5 examples starting on page 328 here: https://www.stata.com/manuals/pss.pdf.

[Joe Tuckles]

OK thanks! I want to use Fisher's asymptotic method, applying a two-sided test with a Bonett and Wright estimator of asymptotic variance, I just want to check the language 'a power for sample size estimation value' does refer to ra, and so when describing it if I say "I applied a two-sided test with a Bonett and Wright estimator of asymptotic variance, with a power for sample size estimation value of 0.80", that makes sense?