Re: Output from POWER UNIVARIATE LINEAR

Posted by Bruce Weaver on
URL: http://spssx-discussion.165.s1.nabble.com/Output-from-POWER-UNIVARIATE-LINEAR-tp5740814p5740818.html

Thanks very much Richard.  Yes, the other Rich who posted that emailed me that same request about changing .1 to .15.  The issue was that for the example I chose, Rsq for the reduced model was equal to Rsq-change, which could cause further confusion for someone trying to decipher the output.  In light of that, here is another example that avoids the issue.  

* Another example.
* Block 1:  y = b0 +b1*x1 + b2*x2 + b3*x3 + b4*x4 + error
* Block 2:  y = b0 +b1*x1 + b2*x2 + b3*x3 + b4*x4 + b5*x5 + error
* I want power of .8 to detect a change in Rsq from .4 to .45.  
* The relevant test is a test on the change in Rsq when adding 1 variable,
* so NTESTED=1.  And there are 4 variables in the block 1 model,
* which SPSS calls the nested model.  Let's see what the output says!
* First, here is the Stata command:  
* power rsquared 0.4 0.45, ntested(1) ncontrol(4) power(.8)  
* Stata result:  n = 89.

POWER UNIVARIATE LINEAR
  /PARAMETERS MODEL=FIXED SIGNIFICANCE=0.05 POWER=0.8
   TOTAL_PREDICTORS=5 TEST_PREDICTORS=1
   FULL_MODEL=0.45 NESTED_MODEL=0.4 INTERCEPT =TRUE.



As expected, the output incorrectly shows that there is only 1 variable in the nested (or reduced, or block 1) model.  There are 4 variables in block 1 and 1 "tested" variable is added to block 2.  

I hope this example clarifies things for anyone who was confused by having Rsq for block 1 = Rsq change.  

Cheers,
Bruce

Richard M wrote

Yes, I understood the complaint, and I don't disagree. Unfortunately, I also don't do the design and implementation. ;)

I will pass along your response. My guess is that there will be a change to the output effective with version 29.

Thanks for your interest in our product and your willingness to share your thoughts, complaints, etc. While I can't promise that *every one* will be attended to, I can promise that I will pass them along.

Rick M
Quality Assurance
SPSS

BTW: Someone made a request to change the nested effect to .15 and ask if power went up or down. It goes down, precipitously, which I think surprises no one (of course holding N constant at 81 and keeping everything else the same). Here is a chart (from an extension that I am working on - hopefully finished, if I can shake another bug or two out of it):