Petr Keil

January 2016

- Likelihood, maximum likelihood, deviance.
- The basic prob. distributions.
- Distinguish the deterministic and stoch. model parts.
- Posterior, prior, likelihood and their connection.
- MCMC
- Credible and prediction intervals.
- Elementary model specification in JAGS.
- GLM, occupancy models, autoregression, random effects.

- ALWAYS start with simple models (and small data).
- Make your models cool and complex only AFTER your simple models run.

Learn your probability distributions. The useful ones are:

- Normal, Poisson, Binomial, Uniform
- Beta, Gamma, Exponential, Negative Binomial, Lognormal
- Categoriacal, Multinomial, Double exponential
- Truncated and censored distributions

- Copy other people's codes and models.
- Copy your own codes and models.

- Your priors may be too wide.
- You may need to provide better initial values manually.
- You have latent variabels - they need good inits.
- You have mistaken
`~`

for`<-`

- You provide negative \( \lambda \) to Poisson -> log link.
- Same with Bernoulli -> logit link.
- Standardize and center your variables,
**especially for log link**.

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