I am currently running several estimations using PROC GLM, PROC LOGISTIC and PROC CATMOD. I have to perform some associated tests of hypothesis as well. I was waondering whether there is any easy way to look at the statistical power of these tests. For example, is there a way to graph the power curve associated with these tests? Thanks in advance!
I'm not an expert, but my limited understanding is that the power curves aren't known for "more complicated" scenarios, and you might have to simulate those cases. For ordinal logistic regression, see http://www2.sas.com/proceedings/sugi27/p260-27.pdf
It looks like you can use multiple ROC statements in PROC LOGISTIC to overlay curves from different models. See the doc: http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_logistic_sec...
I might not be understanding your question, but you can use PROC POWER to perform prospective power analysis for the ANOVA, logistic regression, and some simple tests: http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_power_a00000...
For more complex analyses, see PROC GLMPOWER: http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_glmpower_a00...
HI Rick:
Thanks a lot for your reply! I had looked at the PROC POWER and PROC GLMPOWER alternatives. I haven't used these two PROCs before and my understanding may not be perfect but it seems to me that these two PROCs can handle some model specifications - like simple linear regression or more complicated linear regression, logistic regression where the dependent variable is binary etc. Now, some of the cases that I am dealing with are more complicated than these - like for example, with PROC LOGISTIC I am estimating an ordinal logistic regression where the dependent variable takes more than two possible values.
I dont know if I am missing anything, but so far I am not able to make out for sure that POWER & GLMPOWER can handle this scenario for example. It will be great if you can share anything in this connection.
Also, I am thinking about the ROC curve as well. Do I need to use the ODS functionality to graph the ROC curves? If yes, is there a way to overlay the ROC curves from two different models in the same diagram for comparision purposes?
Thanks again!!
I'm not an expert, but my limited understanding is that the power curves aren't known for "more complicated" scenarios, and you might have to simulate those cases. For ordinal logistic regression, see http://www2.sas.com/proceedings/sugi27/p260-27.pdf
It looks like you can use multiple ROC statements in PROC LOGISTIC to overlay curves from different models. See the doc: http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_logistic_sec...
Thanks Rick, your replies have been very helpful!! I will definitely try to leverage learnings from these resources that you have shared.
Also, as I think through this question, I am realizing that perhaps my original question probably has an inherent loophole in itself. I mean, in order to calculate the power (by whatever means) you need to specify the alternative hypothesis under which you are computing the power - because you have to "know" or "simulate" the probabilty distribution of the test statistic under the alternative hypothesis - correct?
If we do not specify the alternative hypothesis perhaps the concepts of ROC curve/area under the ROC curve etc are the correct things to focus on.
Thanks again!!
Sounds right. Both H_0 and H_a have to be specified for the test. The H_a typically tells us whether to use one or both tails when computing p-values. Best wishes and good luck.
Build your skills. Make connections. Enjoy creative freedom. Maybe change the world. Registration is now open through August 30th. Visit the SAS Hackathon homepage.
Register today!ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.
Find more tutorials on the SAS Users YouTube channel.