topic Re: sample size calculation based on an RxC table in SAS Programming

sample size calculation based on an RxC table

superbibi — Fri, 02 Nov 2018 16:40:44 GMT

Hi Friends,

I have this table from a previous study showing the questionnaire outcomes of two groups of people with chi-square test.

The table shows the percentage of participant response at each score.

How can I use this information to calculation the sample size for a new study in order to gain a significant different between two randomized groups with 1:1 allocation ratio, type I error = 0.05 and power = 0.8?

Thank you.

	score	1	2	3	4	5	6	7	8	9	10
study (n= 25)	percentage of participant of that score	0	0	0	0	0.8	0.2	0	0	0	0
control (n= 25)	percentage of participant of that score	0	0	0	0	0	0.1	0	0.3	0.2	0.4

Re: sample size calculation based on an RxC table

Reeza — Fri, 02 Nov 2018 17:16:43 GMT

I think you want PROC POWER.

https://www.lexjansen.com/wuss/2016/105_Final_Paper_PDF.pdf

Re: sample size calculation based on an RxC table

superbibi — Fri, 02 Nov 2018 17:47:15 GMT

Thank you for the response.
I also know this link is great for RxC table. http://support.sas.com/kb/25/013.html
But it assumes not exactly the method I want. I am wondering how can use all the information in this RxC table to do the sample size calculation.

Re: sample size calculation based on an RxC table

Reeza — Fri, 02 Nov 2018 18:19:47 GMT

That link is a teenager (2005?). If you want to use all the information, then in my experience, no proc will work, you have to simulate the data and CI and power you want.

Re: sample size calculation based on an RxC table

FreelanceReinh — Fri, 02 Nov 2018 18:57:52 GMT

Hi @superbibi,

Many thanks for the link. Apparently I wasn't aware of this macro when I posted Re: proc power for multiple groups in 2016 (includes sample size calculation and a simulation). In that thread the RxC table was transposed (two proportions, more than two groups) and the likelihood-ratio chi-square test was used. It seems to me (at first glance) that the formula from Chow et al. (2008) I applied there doesn't use more information from the RxC table than the powerRxC macro. Couldn't it be too restrictive to use more information?

Re: sample size calculation based on an RxC table

superbibi — Fri, 02 Nov 2018 19:02:14 GMT

Thank you. Do you think I should assume a meaningful difference of probability regarding this kind of sample size assumption? For example, in this case I use the information in question 6, which showed 0.2 and 0.1 at study and control group, and I wrote the code as below.

Do you think this method appropriate?

proc power;
      twosamplefreq test=pchi
         groupproportions = (.20 .10)
         nullproportiondiff = .10 to 0.40 by 0.1
         npergroup=.
         alpha = 0.05
         power = 0.8;
   run;

Re: sample size calculation based on an RxC table

FreelanceReinh — Fri, 02 Nov 2018 22:44:17 GMT

If you asked me I'd say now you go in the opposite direction and use less information from the table. Then your sample sizes will tend to be larger than before.