Statistics 4240 - Data Mining

I mentioned in class that I had written a cleaner function in R to plot error bars.  It is below along with some simulated data to show how to use it.

-Yair

###################################################################################################

# basic function to plot a model's coefficients and error bars

# by Yair Ghitza (yg2173@columbia.edu)

# based on Jonathan Kastellec and Eduardo Leoni's code (www.tables2graphs.com)

library(arm)

plot.errorbars <- function(model, ...)

{

v_coef <- coef(model)

v_se <- se.coef(model)

v_low <- v_coef-qnorm(.975)*v_se

v_high <- v_coef+qnorm(.975)*v_se

v_labels <- labels(coef(model))

v_yaxis <- c(length(v_coef):1)

v_xlimits <- c(floor(min(v_low)), ceiling(max(v_high)))

par(mar=c(3, 15, 4, 2)) # make room for the labels on the left

plot(v_coef, v_yaxis, pch=19, axes=F, xlab="", ylab="", xlim=v_xlimits, ...)

segments(v_coef-qnorm(.975)*v_se, v_yaxis, v_coef+qnorm(.975)*v_se, v_yaxis, lwd=1)

segments(v_coef-qnorm(.750)*v_se, v_yaxis, v_coef+qnorm(.750)*v_se, v_yaxis, lwd=2)

axis(1, at=seq(v_xlimits[1], las=1, v_xlimits[2], 

by=ifelse(v_xlimits[2] - v_xlimits[1] > 10, ceiling((v_xlimits[2] - v_xlimits[1])/10), 1)))

axis(2, at=v_yaxis, label=v_labels, las=1, tick=F)

segments(0, -1, 0, length(v_coef)+1, lty=2)

par(mar=c(5, 4, 4, 2) + 0.1) # reset margins

}

# simulate fake data and create a model

n <- 1000

x1 <- rnorm(n)

x2 <- rnorm(n)

x3 <- rnorm(n)

x4 <- rnorm(n)

x5 <- rnorm(n)

error <- rnorm(n,0,15) # large error term so we get visible standard errors

y <- 2 - 1*x1 - 1.5*x2 + 1*x3 - 0.01*x4 + 1.5*x5 + error

M1 <- lm(y ~ x1 + x2 + x3 + x4 + x5)

summary(M1)

# plot it (note the "main=..." argument is optional to create the title")

plot.errorbars(M1, main=paste("Linear model 1,\nN=", length(M1$fitted.values), sep=""))