The difference in means are not the focus. There are 36 studies and each will have a different mean. The study mean is used in each arm of the study. We are really interested in the 7 treatments and in comparing them with each other.
Another version of the model that is labeled a “random effects” has separate treatment effects for each treatment by study. I have coded this, and it’s a little more challenging to get the model to converge, but it did converge, and the mixing looks good, however, it seems that the sampler gets stuck in certain places for a not huge but not insignificant number of samples. I got this from looking at the plot along the axis below the distribution (you get 36 means, and 36 random treatment effects).
So basically the d-vector is the ball ball game and the other parameters are considered
nuisance parameter. Also, I might calculate things like odd ratios and some other measures either in the model or with the samples. It’s probably easier to have them in the model since they are named and tracked and ready without any additional data wrangling.
Thank you for taking the time to look at the model.