glmer2stan (version 0.995)

stanpredict: Compute predictions from a model fit with glmer2stan

Description

This function computes predictions and prediction intervals from a model fit by glmer2stan. Predictions use the entire posterior for inference. Uncertainty due to both posterior probability and sampling distribution are included.

Usage

stanpredict( stanfit , data , vary_prefix="vary_" , fixed_prefix="beta_" ,  probs=c(0.025,0.975) , nsims=1e4 )

Arguments

fit

stanfit object, produced with glmer2stan

data

data frame of cases to use for prediction

fixed_prefix

Identifying prefix for fixed effects, in the samples

vary_prefix

Identifying prefix for varying effects, in the samples

probs

Prediction quantiles to report (default is 95 percent interval)

nsims

Number of simulations, for generating sampling distributions

Details

This function uses the posterior samples from a stanfit object, fit with glmer2stan, to simulate model predictions. The returned value is a list with a named entry for each outcome variable in the model. Under each outcome variable is a list with: (1) mu, expectation of mean; (2) mu.ci, quantiles for the mean; and (3) obs.ci, quantiles for the predicted sampling distribution. When passing new data to stanpredict, you can use a list with different length vectors. Short vectors will be grown to match the length of the longest vector, using repetition. For cluster variables, using a value of zero will be interpreted as omitting varying effects from predictions, generating predictions for average cases in the population. Otherwise, cluster variable values should be integers.

See Also

glmer2stan

Examples

# simple linear regression
data(cars)
m <- glmer2stan( dist ~ speed , data=cars )
speed.seq <- seq( from=min(cars$speed) , to=max(cars$speed) , length.out=20 )
pred.dist <- stanpredict( m , data=list( speed=speed.seq ) )$dist
plot( dist ~ speed , cars )
lines( speed.seq , pred.dist$mu )
lines( speed.seq , pred.dist$mu.ci[1,] , lty=2 )
lines( speed.seq , pred.dist$mu.ci[2,] , lty=2 )

# binomial example
data(UCBadmit)
m <- glmer2stan( cbind(admit,reject) ~ (1|dept) + male , 
    data=UCBadmit , family="binomial" )
pred.admit <- stanpredict( m , data=UCBadmit )$admit
prop.admit <- UCBadmit$admit / UCBadmit$applications
plot( prop.admit , ylim=c(0,1) , pch=ifelse(UCBadmit$male==1,2,1) ,
    ylab="probability admit" , xlab="case" )
lines( 1:12 , pred.admit$mu )
lines( 1:12 , pred.admit$mu.ci[1,] , lty=2 )
lines( 1:12 , pred.admit$mu.ci[2,] , lty=2 )
lines( 1:12 , pred.admit$obs.ci[1,] , lty=3 )
lines( 1:12 , pred.admit$obs.ci[2,] , lty=3 )