I have successfully created a right-censored mixed model with brms. I would now like to test the hypotheses that different specific combinations of three binary predictors do not equal each other.

The basic model is “outcome ~ P1 * P2 * P3 + (1 | subject)”

P1, P2, and P3 each are binomial factors. What is the syntax to test, for example, the hypotheses that P1.a:P2.a:P3.a <> P1.a:P2.a:P3.b and P1.a:P2.a:P3.b <> P1.a:P2.b:P3.b

So you have three binary factors (P1, P2, P3) each with the levels “a” and “b”.

How you have to phrase this in the hypothesis function depends on how these categorical variables were coded into the regression. Did you use the defauly dummy coding?

I hate to be “that guy”–ok, I guess it’s my job to be “that guy”–but let me just say that I don’t recommend that you test the hypotheses that various coefficients do not equal each other. In real life none of these parameters will equal each other, and what I recommend is that you use the simulations from Stan (or its wrapper such as brms) to summarize your uncertainties about these parameters.

I mean, sure, do what you want, and if you do want to do a hypothesis test, sure, use Stan. But I think you can do so much more if you move away from the hypothesis testing framework. I recommend the 8 schools example in chapter 5 of Bayesian Data Analysis as an example of what can be done from the Bayesian perspective.

Sadly, I agree with you. The study isn’t my own. I was given data that was both clustered and right-censored. I thought to try a censored mixed-level model, but the only option in R is through brms. I will look up that example and see if I can phrase things in a way that will be palatable to senior faculty who have spent their entire lives in a purely frequentist world.