Transitions that exceeded treedepth for an intercept model with many Missing Data

Modeling
, , ,
1

Dear Stan community,
d.Rdata (10.2 KB)
Intercept_model.stan (1.9 KB)
model_fitting.rmd (1.4 KB)

I am fitting an intercept model to my data, but I encountered confusing fitting problems, and I would like to ask for help.

In my data, the outcome/dependent variables are participants’ skin conductance responses (SCR). Please see attached files, where I attached the Stan model, an example dataset with 10 participants, and a RMarkdown file in which using the rstan to do the model fitting.

There are three things that should be noted:
(1) Due to the experimental protocol, half of the SCR are missing (i.e., NaN), and I don’t want to delete them, because in my experiment the trial sequences are meaningful.
(2) Since SCR (i.e., the DV) can only take positive values, so in my model, I employed a truncated Gaussian likelihood.
(3) My model includes varying/random intercepts at the participant level.

Following is my Stan model, which is syntactically correct:

data {
  int<lower=0> N_obs; // number of observed data points
  int<lower=0> N_mis; // number of missing data points 
  int<lower=1> N_all; // number of total data points (i.e., observed + missing)
  int<lower=1, upper=N_all> ii_obs[N_obs]; // indexes of the observed data
  int<lower=1, upper=N_all> ii_mis[N_mis]; // indexes of the missing data
  
  int<lower=1> S; // number of participants
  
  int<lower=1, upper=S> ID[N_all]; // participant index
  array[N_obs] real<lower=0> SCR_obs; // observed SCR data
  
}

parameters {
  
  // missing data
  array[N_mis] real<lower=0> SCR_mis; // missing SCR data
  
  // Hierarchical mu parameters (at the population-level)
  real mu_b0; // intercept
  
  // sigma parameters
  real<lower=0> sigma; // for the truncated Gaussian likelihood
  real<lower=0> sigma_b0; // from the hierarchical distribution of intercept

  
  // Normal distributions for non-center parameterization, i.e., Matt trick
  real z_b0[S];
 
  
}


transformed parameters {
  
  // data
  array[N_all] real<lower=0> SCR; 
  SCR[ii_obs] = SCR_obs;
  SCR[ii_mis] = SCR_mis; 
  
   // participant-level parameters
  real b0[S];
  
  for (s in 1:S) {
    b0[s] = mu_b0 + z_b0[s]*sigma_b0;
  }

}
  

model {
  
  mu_b0 ~ normal(0, 10);
  
  sigma ~ gamma(2, 2);
  sigma_b0 ~ gamma(2, 2);

  z_b0 ~ normal(0, 1);
  
  for (n in 1:N_all) {
     target += normal_lpdf(SCR[n] | b0[ID[n]], sigma) -
        normal_lccdf(0 | b0[ID[n]], sigma);
  }
 
  
}
 
 
generated quantities {
  vector[N_all] log_lik;
  
  {for (n in 1:N_all) {
    log_lik[n] = normal_lpdf(SCR[n] | b0[ID[n]], sigma) -
        normal_lccdf(0 | b0[ID[n]], sigma);
  }
    
  }
  
}

When I fitted this model, I first tried one chain with 5,00 warmup and 1,000 iteration, and adapt_delta = 0.95. But the model fitting failed and I got the following warnings:

image
image1020×197 11.8 KB

I have no idea what’s going wrong. Might anyone know what’s going wrong? I am immeasurably grateful for any suggestions/recommendations/comments.

I hope I included all relevant information and this is the right place to ask a question like this.

Thank you very much!!!

Operating System: Windows 10
rstan Version: 2.26.13
Stan Version: 2.26.1
All the best,