Vectorized `ordered_logistic` compiles but then does not sample correctly

Modeling
4

Hey all. I have a somewhat related issue with these data. I’m fitting a simplified version of a model from the 2nd edition of Statistical Rethinking. In Section 12.4, McElreath introduced an ordered categorical predictor, education, which has 8 levels ranging from "Elementary School" to "Graduate Degree". In McElreath’s style of notation, the full version of the model was

\begin{align} R_i & \sim \operatorname{Ordered-logit} (\phi, \kappa) \\ \phi_i & = \beta_E \sum_{j=0}^{E_i - 1} \delta_j + \beta_A A_i + \beta_I I_i + \beta_C C_i \\ \kappa_k & \sim \operatorname{Normal}(0, 1.5) \\ \beta_A, \beta_I, \beta_C, \beta_E & \sim \operatorname{Normal}(0, 1) \\ \delta & \sim \operatorname{Dirichlet}(\{ 2, 2, 2, 2, 2, 2, 2 \}), \end{align}

where R is the ordinal response variable, E is our new ordered categorical predictor, and A through C are other predictors. I’m dropping those other predictors here to streamline the model and the code, which simplifies my model for \phi_i to

\phi_i = \beta_E \sum_{j=0}^{E_i - 1} \delta_j.

I’m currently working through the rstan code, with this version running successfully:

library(tidyverse)
library(tidybayes)
library(rstan)

data(Trolley, package = "rethinking")

edu_levels_vec <- c("Elementary School", "Middle School", "Some High School", "High School Graduate",
                    "Some College", "Bachelor's Degree", "Master's Degree", "Graduate Degree")

stan_data <- Trolley |> 
  mutate(edu_new = factor(edu, levels = edu_levels_vec)) |> 
  select(response, edu_new) |> 
  compose_data(alpha = rep(2, times = 7))

model_code <- '
data {
  int n;
  array[n] int response;
  array[n] int edu_new;
  vector[7] alpha;  // For the `delta` prior
}
parameters {
  ordered[6] kappa;
  real bE;
  simplex[7] delta;
}
model {
  vector[n] phi;
  vector[8] delta_j;
  delta_j = append_row(0, delta);
  
  for (i in 1:n) phi[i] = bE * sum(delta_j[1:edu_new[i]]);
  response ~ ordered_logistic(phi, kappa);
  
  kappa ~ normal(0, 1.5);
  bE ~ normal(0, 1);
  delta ~ dirichlet(alpha);
}
'

fit1 <- stan(
  model_code = model_code, 
  data = stan_data,
  cores = 3, chains = 3, seed = 12)

Now we get to the heart of my problem. I’m trying to vectorize the code for the phi, and can’t seem to get it right. Here’s my current attempt:

'
model {
  vector[n] phi;
  vector[8] delta_j;
  delta_j = append_row(0, delta);
  
  phi = bE * sum(delta_j[1:edu_new]);  // Only this line has chaged.
  response ~ ordered_logistic(phi, kappa);
  
  kappa ~ normal(0, 1.5);
  bE ~ normal(0, 1);
  delta ~ dirichlet(alpha);
}
'

When I try running the model this way, I get this error message:

Error in stanc(file = file, model_code = model_code, model_name = model_name,  : 
  0
Semantic error in 'string', line 17, column 27 to column 34:
   -------------------------------------------------
    15:    delta_j = append_row(0, delta);
    16:  
    17:    phi = bE * sum(delta_j[1:edu_new]);  // Only this line has chaged.
                                    ^
    18:    response ~ ordered_logistic(phi, kappa);
    19:  
   -------------------------------------------------

Range bound must be of type int. Instead found type array[] int.

I don’t understand what this is prompting me to do. Is there a way to vectorize phi?