multilevel (version 2.5)

simple.predict: A simple variant of the generic predict command useful for S4 lmer models

Description

Statistical models can be used to predict levels of an outcome variable given specific values of predictors. R has a number of predict functions linked to specific models (e.g., predict.lm, predict.glm). At their core, predict commands multiple the estimated coefficients and the corresponding predictors, and then sum across the intercept and predictors. Given the existing functions, there is little value in using simple.predict in most cases. As of 2013, however, models produced with lme4 functions (lmer and glmer) had no specific predict commands, so the simple.predict function was primarily designed to provide predicted values from lme4 models.

Usage

simple.predict(orig.mod,formula,newdata,dichot=FALSE)

Arguments

orig.mod

The original model providing the coefficient estimates.

formula

A formula that does NOT specify the outcome variable. The formula should mirror the formula in the orig.mod minus the variable to the left of the ~ sign (e.g., ~IV1+IV2).

newdata

A dataframe with names corresponding to each predictor in orig.mod and providing values upon which to estimate the outcome.

dichot

An option used with binomial models. If TRUE, returns estimates scaled to a response vector between 0 and 1. Equivalent to using type="response" in predict.glm for a binomial model.

Value

Returns a vector of predicted values.

See Also

predict predict.lm predict.glm

Examples

data(bh1996)  

# A simple example with lm

tmod<-lm(WBEING~HRS+LEAD,bh1996)
simple.predict(tmod,~HRS+LEAD,data.frame(HRS=c(5,10),LEAD=c(3.0,4.5)))
predict(tmod,data.frame(HRS=c(5,10),LEAD=c(3.0,4.5)))  #Using the predict.lm function

# A simple example with glm

tmod.glm<-glm(ifelse(WBEING>3,1,0)~HRS+LEAD,family="binomial",bh1996)
simple.predict(tmod.glm,~HRS+LEAD,data.frame(HRS=c(5,10),LEAD=c(3.0,4.5)),dichot=TRUE)
predict(tmod.glm,data.frame(HRS=c(5,10),LEAD=c(3.0,4.5)),type="response") #Using the predict.glm function

################################################
# An example including a factor as a predictor #
################################################

#Creating a factor
bh1996$LEAD.F<-ifelse(bh1996$LEAD<2.5,0,ifelse(bh1996$LEAD>3.5,2,1))
bh1996$LEAD.F<-factor(bh1996$LEAD.F,levels=c(0,1,2),labels=c("Lo","Med","Hi"))

#Estimating the original model
tmod<-lm(WBEING~HRS+LEAD.F,bh1996)

#Setting up the predictor dataframe
TDAT<-data.frame(HRS=c(5,10),LEAD.F=c("Lo","Med"))
TDAT$LEAD.F<-factor(TDAT$LEAD.F,levels=c("Lo","Med","Hi"))  #This is key in making this work with factors

#Contrasting the predicted values from approaches
simple.predict(tmod,~HRS+LEAD.F,TDAT)
predict(tmod,data.frame(HRS=c(5,10),LEAD.F=c("Lo","Med")))  #Using the predict.lm function

#################################################
#  An Example using the lme4 package and lmer   #
# Delete the preceding # to execute the example #
#################################################

#library(lme4)
#tmod<-lmer(WBEING~HRS+LEAD+(1|GRP),bh1996)
#simple.predict(tmod,~HRS+LEAD,data.frame(HRS=c(5,10),LEAD=c(3.0,4.5)))