These functions are used to specify prior distributions for the model parameters.
normal(location = 0, scale)half_normal(scale)
log_normal(location, scale)
cauchy(location = 0, scale)
half_cauchy(scale)
student_t(location = 0, scale, df)
half_student_t(scale, df)
exponential(scale = 1/rate, rate = 1/scale)
Prior location. Typically prior mean (see details).
Prior scale. Typically prior standard deviation (see details).
Prior degrees of freedom.
Prior rate.
Object of class nma_prior.
The location and scale parameters are typically the prior mean
and standard deviation, with the following exceptions:
For the Cauchy distribution location is the prior median and
scale is the prior scale.
For the log-Normal distribution, location and scale are the prior
mean and standard deviation of the logarithm.
The following table summarises which prior distributions may be used with which model parameters. Essentially, priors that take only non-negative values (e.g. half-Normal) may only be used for non-negative parameters (heterogeneity SD/variance/precision, and any auxiliary parameter). If a real-valued prior distribution is specified for a non-negative parameter, it will be truncated at 0 to be non-negative.
Intercept prior_intercept |
Treatment effects prior_trt |
Heterogeneity prior_het |
Regression coefficients prior_reg |
Auxiliary parameter prior_aux |
|
Normal normal() |
Yes | Yes | Yes | Yes | Yes |
half-Normal half_normal() |
- | - | Yes | - | Yes |
log-Normal log_normal() |
- | - | Yes | - | Yes |
Cauchy cauchy() |
Yes | Yes | Yes | Yes | Yes |
half-Cauchy half_cauchy() |
- | - | Yes | - | Yes |
Student t student_t() |
Yes | Yes | Yes | Yes | Yes |
half-Student t half_student_t() |
- | - | Yes | - | Yes |
Exponential exponential() |
- | - | Yes | - | Yes |
summary.nma_prior() for summarising details of prior
distributions. plot_prior_posterior() for plots comparing the prior and
posterior distributions of model parameters.