School Data, from Charnes et al. (1981). The aim is to explain scores on 3
different tests, reading
, mathematics
and selfesteem
from 70 school sites by means of 5 explanatory variables.
data(schooldata)
A data frame with 70 observations on the following 8 variables.
education
education level of mother as measured in terms of percentage of high school graduates among female parents
occupation
highest occupation of a family member according to a pre-arranged rating scale
visit
parental visits index representing the number of visits to the school site
counseling
parent counseling index calculated from data on time spent with child on school-related topics such as reading together, etc.
teacher
number of teachers at a given site
reading
total reading score as measured by the Metropolitan Achievement Test
mathematics
total mathematics score as measured by the Metropolitan Achievement Test
selfesteem
Coopersmith Self-Esteem Inventory, intended as a measure of self-esteem
This dataset was shamelessly borrowed from the FRB
package.
The relationships among these variables are unusual, a fact only revealed by plotting.
# NOT RUN { data(schooldata) # initial screening plot(schooldata) # better plot library(corrgram) corrgram(schooldata, lower.panel=panel.ellipse, upper.panel=panel.pts) #fit the MMreg model school.mod <- lm(cbind(reading, mathematics, selfesteem) ~ education + occupation + visit + counseling + teacher, data=schooldata) # shorthand school.mod <- lm(cbind(reading, mathematics, selfesteem) ~ ., data=schooldata) Anova(school.mod) heplot(school.mod) heplot3d(school.mod) # robust model, using robmlm() school.rmod <- robmlm(cbind(reading, mathematics, selfesteem) ~ ., data=schooldata) # note that counseling is now significant Anova(school.rmod) # compare classical HEplot with robust heplot(school.mod, cex=1.4, lty=1, fill=TRUE, fill.alpha=0.1) heplot(school.rmod, add=TRUE, error.ellipse=TRUE, lwd=c(2,2), lty=c(2,2), term.labels=FALSE, err.label="", fill=TRUE) # }