arm (version 1.0-3)

coefplot: Generic Function for Making Coefficient Plot

Description

Functions that plot the coefficients $\pm$ 1 and 2 sd from a lm, glm, bugs, and polr fits.

Usage

coefplot(object,...)

## S3 method for class 'default':
coefplot(coefs, sds, 
                varnames=NULL, CI=2, vertical=TRUE, 
                cex.var=0.8, cex.pts=0.9, 
                col.pts=1, var.las=2,\dots)
                
## S3 method for class 'lm':
coefplot(object, varnames=NULL, intercept=FALSE, \dots)
## S3 method for class 'glm':
coefplot(object, varnames=NULL, intercept=FALSE, \dots)
## S3 method for class 'polr':
coefplot(object, varnames=NULL, \dots)
## S3 method for class 'bugs':
coefplot(object, varnames=NULL, CI=2, \dots)

Arguments

object

fitted objects-lm, glm, bugs and polr, or a vector of coefficients.

...

further arguments passed to or from other methods.

coefs

a vector of coefficients.

sds

a vector of sds of coefficients.

varnames

a vector of variable names, default is NULL, which will use the names of variables.

CI

confidence interval, default is 2, which will plot $\pm2$ sds or 95% CI. If CI=1, plot $\pm1$ sds or 50% CI instead.

vertical

orientation of the plot, default is TRUE which will plot variable names in the 2nd axis. If FALSE, plot variable names in the first axis instead.

cex.var

The fontsize of the varible names, default=0.8.

cex.pts

The size of data points, default=0.9.

col.pts

color of points and segments, default is black.

var.las

the orientation of variable names against the axis, default is 2. see the usage of las in par.

intercept

If TRUE will plot intercept, default=FALSE to get better presentation.

Value

• Plot of the coefficients from a lm or glm fit. You can add the intercept, the variable names and the display the result of the fitted model.

Details

This function plots coefficients from lm, glm and polr with 1 sd and 2 sd interval bars.

References

Andrew Gelman and Jennifer Hill, Data Analysis Using Regression and Multilevel/Hierarchical Models, Cambridge University Press, 2006.

See Also

display, par, lm, glm, bayesglm, plot

Examples

y1 <- rnorm(1000,50,23)
 y2 <- rbinom(1000,1,prob=0.72)
 x1 <- rnorm(1000,50,2) 
 x2 <- rbinom(1000,1,prob=0.63) 
 x3 <- rpois(1000, 2) 
 x4 <- runif(1000,40,100) 
 x5 <- rbeta(1000,2,2) 
 
 longnames <- c("a long name01","a long name02","a long name03",
                "a long name04","a long name05")

 fit1 <- lm(y1 ~ x1 + x2 + x3 + x4 + x5)
 fit2 <- glm(y2 ~ x1 + x2 + x3 + x4 + x5, 
            family=binomial(link="logit"))
 
 # plot 1
 par (mfrow=c(2,2), mar=c(3,3,4,1), mgp=c(2,0.25,0), tcl=-0.2)
 coefplot(fit1, xlab="", ylab="", main="Regression Estimates")
 coefplot(fit2, col.pts="blue",
    xlab="", ylab="", main="Regression Estimates")
 
 
 # plot 2
 par (mar=c(2,8,2,0.5))
 coefplot(fit1, longnames, intercept=TRUE, CI=1,
     xlab="", ylab="", main="Regression Estimates")
 
 # plot 3
 par (mar=c(2,2,2,2))
 coefplot(fit2, vertical=FALSE, var.las=1,
     xlab="", ylab="", main="Regression Estimates")
 
 # plot 4: comparison to show bayesglm works better than glm
 n <- 100
 x1 <- rnorm (n)
 x2 <- rbinom (n, 1, .5)
 b0 <- 1
 b1 <- 1.5
 b2 <- 2
 y <- rbinom (n, 1, invlogit(b0+b1*x1+b2*x2))
 y <- ifelse (x2==1, 1, y)
 x1 <- rescale(x1)
 x2 <- rescale(x2, "center")
 
 M1 <- glm (y ~ x1 + x2, family=binomial(link="logit"))
       display (M1)
 M2 <- bayesglm (y ~ x1 + x2, family=binomial(link="logit"))
       display (M2)
 
    ## stacked plot
    par(mar=c(2,5,3,1), mgp=c(2,0.25,0), oma=c(0,0,2,0), tcl=-0.2)
 
    coefplot(M2, xlim=c(-1,5), intercept=TRUE, xlab="", ylab="")
    points(coef(M1), c(3:1)-0.1, col="red", pch=19)
    segments(coef(M1) + se.coef(M1), c(3:1)-0.1,
        coef(M1) - se.coef(M1), c(3:1)-0.1, lwd=2, col="red")
    segments(coef(M1) + 2*se.coef(M1), c(3:1)-0.1,
        coef(M1) - 2*se.coef(M1), c(3:1)-0.1, col="red")    
    mtext("Coefficients", side=3, at=0.1, outer=TRUE)
    mtext("Estimate", side=3, at=0.6, outer=TRUE)
 
    ## arrayed plot
    par(mfrow=c(1,2), mar=c(2,5,5,1), mgp=c(2,0.25,0), tcl=-0.2)
    x.scale <- c(0, 7.5) # fix x.scale for comparison
 
    coefplot(M1, xlim=x.scale, main="glm", intercept=TRUE,
         xlab="", ylab="")
    coefplot(M2, xlim=x.scale, main="bayesglm", intercept=TRUE,
         xlab="", ylab="")

# plot 5: the ordered logit model from polr
 par (mar=c(3,8,4,1), mgp=c(2,0.25,0), tcl=-0.2)
 
 M3 <- polr(Sat ~ Infl + Type + Cont, weights = Freq, data = housing)
 coefplot(M3, xlab="", ylab="", main="polr")
   
 M4 <- bayespolr(Sat ~ Infl + Type + Cont, weights = Freq, data = housing)
 coefplot(M4, xlab="", ylab="", main="bayespolr")

# plot 6: plot bugs 
 par (mar=c(3,8,4,1), mgp=c(2,0.25,0), tcl=-0.2)
 M5 <- lmer(Reaction ~ Days + (1|Subject), sleepstudy)
 M5.sim <- mcsamp(M5)
 coefplot(M5.sim, xlab="", ylab="", main="BUGS model")
 
# plot 7: plot coefficients & sds vectors
 par (mar=c(3,4,4,4), mgp=c(2,0.25,0), tcl=-0.2)
 coef.vect <- c(0.2, 1.4, 2.3, 0.5)
 sd.vect <- c(0.12, 0.24, 0.23, 0.15)
 longnames <- c("var1", "var2", "var3", "var4")
 coefplot (coef.vect, sd.vect, longnames,
    xlab="", ylab="", main="Regression Estimates")
 coefplot (coef.vect, sd.vect, longnames, 
    vertical=FALSE, var.las=1, las=2,
    xlab="", ylab="", main="Regression Estimates")