Method of power calculation- using sampsi command

[Karl Yesler]

Hello,

I have some older code which used the sampsi command (https://www.stata.com/help.cgi?sampsi) and I would like to know what the method(method) option refers to, specifically, what is the difference between "post" "change"? The seem to give very different results. See below for example:

You can see that the estimated power for post and change is very different results.

(as an aside, I know what the newer stata command is power, how might I implement this power calculation using the power command?)

sampsi 351 363, sd1(405) sd2(405) n1(2000) n2(4000) pre(8) post(8) onesided r1
> (.3)

Estimated power for two samples with repeated measures
Assumptions:
                                      alpha =   0.0500  (one-sided)
                                         m1 =      351
                                         m2 =      363
                                        sd1 =      405
                                        sd2 =      405
                             sample size n1 =     2000
                                         n2 =     4000
                                      n2/n1 =     2.00
           number of follow-up measurements =        8
 correlation between follow-up measurements =    0.300
            number of baseline measurements =        8
  correlation between baseline measurements =    0.300
   correlation between baseline & follow-up =    0.300

Method: POST
 relative efficiency =    2.581
    adjustment to sd =    0.622
        adjusted sd1 =  252.110
        adjusted sd2 =  252.110

 Estimated power:
               power =    0.537

Method: CHANGE
 relative efficiency =    5.714
    adjustment to sd =    0.418
        adjusted sd1 =  169.424
        adjusted sd2 =  169.424

 Estimated power:
               power =    0.827

Method: ANCOVA
 relative efficiency =    6.442
    adjustment to sd =    0.394
        adjusted sd1 =  159.573
        adjusted sd2 =  159.573

 Estimated power:
               power =    0.865

[Clyde Schechter]

I've never actually used that feature, so I'm not certain of this response, but this is my best guess.

The -method()- option is used with repeated measures data. There are several ways one can analyze changes in a variable in repeated measures study.

1. You can do an ANCOVA where the second observation is the dependent variable and the baseline value of the observation is a covariate. That's what the ANCOVA method calculates power for.

2. You can calculate the difference between the first and second observations and do a regression with the difference as the dependent variable. That would be the CHANGE method.

3. You can have a data set with separate observations for each of the measurements, with an indicator variable distinguishing first from second and regress the measurement against the indicator. That, I think, would be the POST method.

[Karl Yesler]

Thanks Clyde,

So for #3:
"3. You can have a data set with separate observations for each of the measurements, with an indicator variable distinguishing first from second and regress the measurement against the indicator. That, I think, would be the POST method."

Would the regression have the form of outcome = post or is it outcome = post*group + post + group?

Would #3 then be the repeated measure ANOVA?

[Clyde Schechter]

Yes, I think it would be outcome = post*group + post + group.