Bayesian Measurement error survival model

General
1

I’m want to fit a Weibull regression, but I’m considering a covariate to prone measurement error. In particular classical error model. However, I have questions about the priors, because, I have had some problems with the convergence.

This is my code.

data{
  int n;
  vector[n] x_obs; // measurement of x_true
  vector[n] Time;
  vector[n] x; // group
  real<lower=0> tau; // measurement noise
  int nbetaS;
  vector[n] death;
  int K;
  vector[K] xk;
  vector[K] wk; 
}

parameters{
  vector[n] x_true; // unknow true value
  real<lower=0> taux; # prior scale
  real mux;
  vector[nbetaS] betaS;
  real alpha;
  real<lower=0> gamma;
}

model{


   matrix[n,K] h;
   vector[n] cumHaz;
   // LOG-PRIORS
   
   // Measurement error


   target += gamma_lpdf(taux | 2, 5);
   target += normal_lpdf(mux | 0, 1);
   target += normal_lpdf(x_true| mux, taux);
   target += normal_lpdf(x_obs| x_true, tau);
   // Survival fixed effects
   target += normal_lpdf(betaS | 0, 100);
   
   // Association parameter
   target += normal_lpdf(alpha | 0, 100);
   
   // Weibull shape parameter
   target += gamma_lpdf(gamma | 2, 1);
   
   // LOG-LIKELIHOOD FOR SURVIVAL SUBMODEL
   for(i in 1:n){
		for(j in 1:K){
			// hazard function
			h[i,j] = gamma*pow(Time[i]/2*(xk[j]+1), gamma-1) * exp( betaS[1] + betaS[2]*x[i] +   alpha*x_true[i] );
		}
    
		// Cumulative hazard H[t] = int_0^t h[u] du - Gauss-Legendre quadrature
		cumHaz[i] = Time[i]/2*dot_product(wk,h[i,]);
    
		// Survival log-likelihood
		target += death[i]*log(h[i,K]) - cumHaz[i];
	}



   


}

Thanks a lot!

2

Sorry, it seems your question slipped through - did you manage to resolve this? If not, could you provide more information about your problems? In particular the dataset you are trying to fit (or code to simulate new data), the pairs plot (or subset of it) and the exact messages you are getting?

Best of luck with your model!