summary.poisreg {bpr}

Summarizing Bayesian Poisson Regression Fit

Description

This function is a method for class poisreg. It prints summary statistics and returns posterior estimates of regression quantities.

Usage

## S3 method for class 'poisreg'
summary(object, ...)

## S3 method for class 'poisreg'
print(x, ...)

Arguments

object

object of class "poisreg" (usually, the result of a call to sample_bpr).

...

further arguments passed to or from other methods.

x

object of class "poisreg" (usually, the result of a call to sample_bpr).

Details

The printed output of summary.poisreg summarizes the main quantities of the fit. The first component Call recaps the type of prior and algorithm used.

Coefficients is a table of estimated quantities for the regression parameters. The first three columns report the estimated posterior mean, standard errors and medians. The last two columns correspond to the lower and upper bounds of the 0.95 credible intervals. If the credible interval does not include zero, a star is printed in correspondence of each parameter (similarly to the 'significance stars' of summary.lm). All the estimates are computed discarding the first part of the chain as burn-in (more details are printed in the Algorithm section).

Algorithm briefly summarizes the main diagnostics of convergence and efficiency of the algorithm. It prints the number of iterations actually used to obtain the estimates, after removing the burn-in and thinning. If a Metropolis-Hastings algorithm is used, the summary reports the acceptance rate, which is the most commonly used indicator to tune the performance of the algorithm, along with the mean effective sample size (averaged over all parameters). If the importance sampler is used, the summary only reports the effective sample size, which is computed as \sum_{t} w_t^2 / (\sum_{t} w_t)^2 (where w_t is the sequence of weights) and is a measure of the efficiency of the sampler.

Value

summary.poisreg returns a list with elements:

formula : the component from object.

data : list with elements the matrix of covariates X and response variable y.

prior : prior$type from object.

prior_pars : prior parameters from object.

coefficients : the matrix of coefficient estimantes, standard errors and 95% credible intervals.

psi2 : if a horseshoe prior is selected, the estimate of the local shrinkage parameter.

len_burnin : the length of the burn-in used to compute the estimates.

effSize : the mean effective sample size of the chains used to compute the estimates.

Examples

# For examples see example(sample_bpr)