0
$\begingroup$

I am looking for ways of doing power analysis for ordinal logistic GEE (currently in SPSS, but open to attempting R or something else). My understanding is that this will not be possible in software like G*Power, and might require simulations - something I am unfamiliar with. I wonder whether anyone can direct me to some simple resources?

Details:

I am planning an experiment (psychology). Participants listen to a short scenario about some hypothetical events. These scenarios are manipulated in two ways:

  • they involve something positive or negative (valence)
  • they can take place in participants' own lives, or in someone else's (perspective)

So the within-participants factors are Valence and Perspective, yielding four types of trial: Pos-Self, Pos-Other, Neg-Self, Neg-Other.

Participants will make a choice: how many of such events do they prefer? Their choice is between events that take place in different circumstances. First, they will choose whether they prefer 1 event in Circumstance 1, or 2 events in Circumstance 2. If they pick one event in C1, we will offer a further choice: 1 event in C1 vs 4 events in C2. If they again choose one event, we will offer 1 event in C1 vs 6 in C2, and so forth with 8, 10, or 'more than 10' events.

The DV is therefore 'tradeoff point' with a value of 2, 4, 6, 8, 10, or more than 10, and represents the number of events at which participants trade off their initial preference. We will drop participants whose initial preference is not in the expected direction.

There will also be a covariate measured on a 7-point Likert scale.

Improve this question
$\endgroup$
6
  • $\begingroup$ Here is some of my ideas: stats.stackexchange.com/questions/370734/… $\endgroup$
    – user158565
    Commented Nov 28, 2018 at 17:24
  • $\begingroup$ I'd suggest staring this this paper. It also includes binary outcome. If you go to their online version, in the supplement you'll also find the SAS code for the simulation study they used for demonstration. $\endgroup$
    – Penguin_Knight
    Commented Nov 28, 2018 at 17:36
  • $\begingroup$ @user158565, thank you - a very useful practical approach at that link. I will go there to ask you a question about it. $\endgroup$
    – rumnraisin
    Commented Nov 29, 2018 at 11:13
  • $\begingroup$ @Penguin_Knight, thanks - I am not sure I have easy access to SAS, but will find out, and might ask the authors if they happen to have the code also in R (you never know). $\endgroup$
    – rumnraisin
    Commented Nov 29, 2018 at 11:14
  • $\begingroup$ @user158565, I couldn't add a comment to your answer on the question you linked to as I don't have enough reputation. May I ask the question here? At your Step 4, you repeat steps 2/3 5000 or so times (generate the sample according to the model with sample size N, fit the model, perform the test, and record the rejection or acceptance of null hypothesis). How do you go about repeating these steps 5000 times? Do you use code in R or some other program, and do you happen to know of any open source code I could use for this? $\endgroup$
    – rumnraisin
    Commented Nov 29, 2018 at 11:20

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.