Bayesian Biostatistics - Conclusions

Petr Keil
Feb 2014

You should now be able to use

  • likelihood, maximum likelihood, deviance
  • basic prob. distributions
  • deterministic vs stoch. model parts
  • posterior, prior, likelihood, their connection
  • MCMC
  • credible and prediction intervals
  • model specification in JAGS
  • GLM, occupancy models, autoregression, random effects

Advice

  • ALWAYS start with simple models.
  • Make your models complex only AFTER your simple models run.

Advice

  • Copy other people's codes and models.

Advice

  • Learn the difference between data and prior.
  • Learn to make formal definition of models.

Advice

  • Give your data and the code to the readers.

%#&$*! IT STILL DOES NOT RUN

  1. Your priors are too wide.
  2. Provide better initial values manually.
  3. Latent variabels need good inits.
  4. You have mistaken ~ for <-
  5. You provide negative \( \lambda \) to Poisson -> log link.
  6. Same with Bernoulli -> logit link.
  7. Standardize and center your variables, especially for log link.

And finally:

  • If you have a hammer, every problem appears to be a nail.
  • Do not forget the biology for all the stats.

THANK YOU!