Projective variable selection for a Ordered Logistic model

I have a straightforward Ordered Logistic model coded as:

data {
    int<lower = 2> K;
    int<lower = 0> N;
    int<lower = 1> D;
    int<lower = 1, upper = K> y[N];
    row_vector[D] x[N];

parameters {
    vector[D] beta;
    ordered[K - 1] c;

model {
    beta ~ std_normal();
    for (n in 1:N) {
        y[n] ~ ordered_logistic(x[n] * beta, c);

with N \approx~100, D \approx 30. I would like to use Projective Variable Selection as in Piironen et al., using the package projpred.
Sadly, the ordered logistic model has not a reference model implemented (no applicable method for 'get_refmodel' applied to an object of class "stanfit"). Any advice on how to implement variable selection for an OL model?

Have you looked at the init_refmodel() function of the projpred package?

@erognli. How about the family parameter in init_refmodel function? Only gaussian, binomial and poisson are supported.

Ah, you’re quite right. Didn’t see that - sorry for the then not-helpful comment.

1 Like

I think you may need @avehtari for help with projpred.

1 Like

Thanks @Bob_Carpenter , I look forward to getting directions on how to perform the analysis