I have a variable that is Weibull Distributed

 Duration ~ dweib(Shape,Scale)

The Shape and Scale parameters are distributed to log-normal and Weibull distributions, respectively.

 Shape ~ dlnorm(Hyperlogmean,HyperlogPrecision)
 Scale ~ dweib(Hyperweibshape,Hyperweibullscale)

I am having trouble with the standard deviations for Duration from the model.

Here are the parameter estimates from RJAGS

                      MEAN          SD
Duration            4.154e+55    2.275e+58 
Scale               3.809e-01    1.813e+00 
Shape               2.696e+00    2.455e+00 
Hyperlogmean        6.965e-01    7.889e-02
HyperlogPrecision   1.751e+00    2.562e-01
HyperWeibullScale   7.236e-02    2.259e-02
HyperWeibullShape   3.529e-01    3.223e-02

Here are the corresponding quantiles:

                       2.5%       25%        50%      75%      97.5%
Duration             2.058e-01  1.481557   4.05502 2.729e+01 5.091e+05    
Scale                1.758e-06  0.001968   0.02439 0.17805   3.045e+00
Shape                4.420e-01  1.199896   2.00382 3.355e+00 9.083e+00
HyperLogMean         5.413e-01  0.643618   0.69648 7.495e-01 8.515e-01
HyperLogPrecision    1.286e+00  1.571430   1.73832 1.916e+00 2.288e+00
HyperWeibullScale    3.774e-02  0.056396   0.06917 8.478e-02 1.255e-01
HyperWeibullShape    2.924e-01  0.330716   0.35193 3.742e-01 4.184e-01

I am thinking that my problem is coming from the HyperWeibullScale parameter partly because its 97.5% quartile becomes very large when it transformed from RJAGS Weibull parameterization to the R Weibull parameterization. Most of my mixing trace plots and MCMC diagnostics look fine.

I am wondering whether the high standard deviation for Duration has to do with my choice of priors.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.