bic: Bayesian information criterion (BIC)

In manschmi/blm: Bayesian Linear Regression

Description

Bayesian information criterion (BIC) for the fit of a blm object.

Usage

bic(object, ...)

Arguments

object	a blm object.
...	other arguments (currently ignored).

Details

Bayesian information criterion (BIC) of the blm object. BIC = n * ln(RSS/n) + k * ln(n). Where n is the number of observations, k is the number of free parameters to be estimated and RSS is the residuals sum of squares (ie the deviance). When comparing 2 models the strength of the evidence against the model with the higher BIC value can be summarized as follows according to Wikipedia (https://en.wikipedia.org/wiki/Bayesian_information_criterion):

ΔBIC	Evidence against higher BIC
0-2	Not worth more than a bare mention
2-6	Positive
6-10	Strong
>10	Very Strong

Examples

set.seed(1) 
x <- seq(-10,10,.1) 
b <- 0.3

w0 <- 0.2 ; w1 <- 3 ; w2 <- 10

y <- rnorm(201, mean = w0 + w1 * x + w2 *sin(x), sd = sqrt(1/b)) 
mod1 <- blm(y ~ x + sin(x))

## Not run: 
  plot(mod1, xlim=c(-10,10)) 

## End(Not run)
bic(mod1) ## 224.351

#another mod removing the sinus term, clearly less well fitting
mod2 <- blm(y ~ x)

bic(mod2) ## 805.5729 -> huge difference in bic

##much less distinguished model
b <- 0.003
y <- rnorm(201, mean = w0 + w1 * x + w2 *sin(x), sd = sqrt(1/b)) 
mod1 <- blm(y ~ x + sin(x))
## Not run: 
  plot(mod1, xlim=c(-10,10)) 

## End(Not run)
mod2 <- blm(y ~ x)
bic(mod1) #1209.717
bic(mod2) #1215.66 ... still positive support but not strong
  

manschmi/blm documentation built on May 21, 2019, 11:25 a.m.