Jacobian in reparameterizations

Modeling
8

That target written out is

target += -0.5*(derived_var)^2;

I guess this is meant to be a different way to write

target += normal_lpdf(data_var| parameter_1, parameter_2);

but that’s the same as

target +=  -log(parameter_2) - 0.5*((data_var - parameter_1)/parameter_2)^2;

That extra -log(parameter_2) is the log Jacobian correction for multiplying by parameter_2.

2 Likes