In the output of estimation, the "se_mean"s of Cholesky_Omega[1,1] and Cholesky_Omega[1,2] are “NaN”.

Also, the "n_eff"s and "Rhat"s are also “NaN”.

I’m wondering what these values imply?

In the output of estimation, the "se_mean"s of Cholesky_Omega[1,1] and Cholesky_Omega[1,2] are “NaN”.

Also, the "n_eff"s and "Rhat"s are also “NaN”.

I’m wondering what these values imply?

1 Like

Those quantities are only defined if there is variability across samples in the value of the parameter. I think the element at `[1,1]`

should be expected to be constant at 1, but I’m not sure why you’re getting a constant value at `[1,2]`

as well.

4 Likes

Oh wait, yeah, `[1,2]`

will always be zero.

So this is reflects that things are not smart enough to either omit the constants from the output or discern that when dealing with a cholesky factor variable those diagnostics shouldn’t be applied to the elements expected to be constant.

So, that is not a problem because the elements in the first row of Cholesky factor(a 2 by 2 matrix) are fixed.

发自我的iPhone

The value of element in position 1,1 is constrained to be 1 because in Cholesky factor the length of each row should be 1. Since the Chokesky factor I defined is a 2 by 2 matrix, the element 1,1 must be 1.(Sine the element 1,2 is constrained to be 0, making it a lower triangular matrix.)

发自我的iPhone

1 Like

Correct.