statistical power for logistic regression

[noetsi]

how do you find out what the power of your test is for logistic regression? I have never run across this in SAS.

 

[staassis]

I believe the likelihood ratio, Wald and deviance tests have slightly different power functions. You can always calculate the power via Monte Carlo on a grid of parameter values and then interpolate in between with cubic splines. But there are probably simpler approaches. Somebody somewhere must have tabulated the power functions already.

 

[noetsi]

I am sure sas has a module that calculates power, I know SPSS does. I just do not know where it is.

 

[Dason]

https://support.sas.com/documentati...fault/viewer.htm#statug_power_a0000001016.htm

I'm guessing this isn't what you want? Because all I did was click on the first Google result for "sas logistic regression power"

 

[hlsmith]

I was surprised, that looked like a pretty good power procedure for SAS. I was gonna recommend a Monte Carlo simulation, but @Dason link definitely seem appropriate.

 

[noetsi]

https://support.sas.com/documentati...fault/viewer.htm#statug_power_a0000001016.htm

I'm guessing this isn't what you want? Because all I did was click on the first Google result for "sas logistic regression power"

I was relying on experts here rather than searching. I don't always do well on searching [I often get confused with SAS documentation]. Given that I have no theory to guess at the distribution or effective size of much of what I do I am going to have problems using that. I work with vocational rehabilitation, and if there is any theory for much of it I do not know it. I work with raw data and little else.

But its useful to find out about Proc Power. Never heard of that.

 

[Dason]

Given that I have no theory to guess at the distribution or effective size of much of what I do I am going to have problems using that.

I'm not sure how you planned on doing any sort of power calculation then.

 

[noetsi]

Software I used before asked only for an estimated effect size, and an alpha level I thought (as well as a sample size which I know).

 

[Dason]

Was that for logistic regression or just a linear regression

 

[noetsi]

For ANOVA which is essentially linear regression.