ROPE range and functions of y

[emareynaud]

hello

if I fit a brm model of a DV y as a function of predictors
brm(y~predictors)
then rope_range will be based on the sd of y.

If I scale this variable y before taking it into the model as in :
z=f(y)
brm(z~predictors)
then rope_range will be based on the sd of z, which is fine.

BUT : if I use the function into the brm formula as in :
brm(f(x)~predictors)
then the rope_range will not be based on the values of f(x) and remains computed on the basis of the sd of x

not sure if that's correct, as the function f is often "scale"
thought it could be misleading, but it could also be a stupid remark :)

thanks

[mattansb]

What would you suggest?

I think, generally, the default ROPE range should almost always be overridden.

[emareynaud]

I would suggest that the default rope could adapt to the functions of of the DV, as the default is [-0.1sd(DV); + 0.1sd(DV)]

I would change it to [-0.1 sd (f(DV)); +0.1 sd(f(DV))], so that it can also work for f = Identity, and f= scale or else

Thanks again for this great package, and as you mentioned the default ROPE is to be overridden, just if you leave the default options on, they are gonna be used :-)

emanuelle

[strengejacke]

What do you mean by "function"? If we detect some way of transformation of the response variable?

We could try to get the transformation function and apply the inverse before calculating the SD/rope.

[emareynaud]

why can't the sd of the transformed variable be used ?

I might be wrong but the default option does not seem to take into account the transformed variable, but only the variable for the computation of sd

[strengejacke]

Ah, ok, understand. I'm not sure which data is used, must check.