topic Re: proc power setup in SAS Programming

proc power setup

GS2 — Thu, 18 Jun 2020 12:31:10 GMT

Using SAS 9.4

I am running a post-hoc power analysis (I understand it is not a great method but it is a request so I have to complete it). I have 2 groups with a total n=62 (group 1 n=49, group 2 n=13). With an alpha of 0.05 and the power set to 0.8, I want to find out what n would be needed in group 2 to show significance between the groups. I found the code below in the SAS help center but I am not sure my setup is correct. Specifically, I am not sure if I have the group proportions correct. Any help would be appreciated. Thank you

proc power;
twosamplefreq test=fisher
groupproportions= (.79 .21)
alpha= .05
npergroup=.
power= .8;
run;

Re: proc power setup

PGStats — Thu, 18 Jun 2020 17:44:00 GMT

The groupproportions= option should specify the proportions of events in both groups, NOT the proportions of the group sizes to the total size. For a post-hoc analysis you would most likely write the observed proportions as the group proportions. If for example you observed 31/49 events in group 1 and 6/13 events in group 2, you would write groupproportions=0.63 0.46

Note that you cannot keep one of the group sizes fixed and ask for the size of the other group in proc power. One way to do it anyway is to ask for the power and adjust iteratively the size of the second group until you get the power that you want

proc power;
twosamplefreq test=fisher
groupproportions= (.63 .46)
alpha= .05
groupns= (49, 13) /* <- adjust the size of group 2 until calculated power = 0.8 */
power= .;
run;

hth

Re: proc power setup

GS2 — Thu, 18 Jun 2020 19:07:34 GMT

My groups are the event, the 49 is no for the event and 13 is yes for the event. That is why I took it out of the 62. Is this not correct?

Re: proc power setup

PGStats — Thu, 18 Jun 2020 19:19:18 GMT

So, you have a proportion of 49/62 events in your (single) sample. So you must be looking for the power of a one sample frequency test?

Re: proc power setup

GS2 — Thu, 18 Jun 2020 19:30:31 GMT

Between the 2 groups that I have (1 vs 2) I want to say I need x number to achieve significance. So group 1 has 49 and group 2 has 13 so how many in group 2 are needed to show significance.

Re: proc power setup

PGStats — Thu, 18 Jun 2020 19:51:03 GMT

Significance of what? That group 1 is greater than group 2? Or what?

Re: proc power setup

GS2 — Thu, 18 Jun 2020 19:54:05 GMT

Yes, significant difference between the 2 groups

Re: proc power setup

PGStats — Thu, 18 Jun 2020 20:19:43 GMT

proc power provides a calculation of the power of the exact binomial test for the proportion of events in a single sample. To calculate the power of detecting a difference in the number of events and non-events within a sample of size 62, when the event probability is 0.21 (= 13/62), you can do:

proc power;
onesamplefreq 
    test = exact
    ntotal = 62
    nullproportion = 0.5
    alpha = 0.05
    proportion = 0.21  /* = 13 / 62 */
    power = .;
run;