In this exercise, you will further analyze the Wage data set considered throughout this chapter
(a) Perform polynomial regression to predict wage using age. Use cross-validation to select the optimal degree d for the polynomial. What degree was chosen, and how does this compare to the results of hypothesis testing using ANOVA? Make a plot of the resulting polynomial fit to the data.
data(Wage)
data1=Wage
set.seed(100)
deltas <- rep(NA, 10)
for (i in 1:10) {
fit <- glm(wage ~ poly(age, i), data = data1)
deltas[i] <- cv.glm(data1, fit, K = 10)$delta[1]
}
plot(1:10, deltas, xlab = "Degree", ylab = "Test MSE", type = "l")
d.min <- which.min(deltas)
points(which.min(deltas), deltas[which.min(deltas)], col = "red", cex = 2, pch = 20)
We note that using a degree equal to 9 will give us the lowest test MSE based on our K-fold cross validation (K=10).
fit1 <- lm(wage ~ age, data = data1)
fit2 <- lm(wage ~ poly(age, 2), data = data1)
fit3 <- lm(wage ~ poly(age, 3), data = data1)
fit4 <- lm(wage ~ poly(age, 4), data = data1)
fit5 <- lm(wage ~ poly(age, 5), data = data1)
anova(fit1, fit2, fit3, fit4, fit5)
## Analysis of Variance Table
##
## Model 1: wage ~ age
## Model 2: wage ~ poly(age, 2)
## Model 3: wage ~ poly(age, 3)
## Model 4: wage ~ poly(age, 4)
## Model 5: wage ~ poly(age, 5)
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 2998 5022216
## 2 2997 4793430 1 228786 143.5931 < 2.2e-16 ***
## 3 2996 4777674 1 15756 9.8888 0.001679 **
## 4 2995 4771604 1 6070 3.8098 0.051046 .
## 5 2994 4770322 1 1283 0.8050 0.369682
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Based on our ANOVA, we see that models with polynomials higher than cubic are not significant, which is contradictory to what our cv showed. Based on the ANOVA and the K-fold cv, I would recommend using a polynomial of 3 for the predictor age when predicting on wage.
plot(wage ~ age, data = data1, col = "darkgrey")
agelims <- range(data1$age)
age.grid <- seq(from = agelims[1], to = agelims[2])
fit <- lm(wage ~ poly(age, 3), data = data1)
preds <- predict(fit, newdata = list(age = age.grid))
lines(age.grid, preds, col = "green", lwd = 2)
Above is the plot for the polynomial fit to the data.
(b) Fit a step function to predict wage using age, and perform crossvalidation to choose the optimal number of cuts. Make a plot of the fit obtained.
cvs <- rep(NA, 10)
for (i in 2:10) {
data1$age.cut <- cut(data1$age, i)
fit <- glm(wage ~ age.cut, data = data1)
cvs[i] <- cv.glm(data1, fit, K = 10)$delta[1]
}
plot(2:10, cvs[-1], xlab = "Cuts", ylab = "Test MSE", type = "l")
d.min <- which.min(cvs)
points(which.min(cvs), cvs[which.min(cvs)], col = "red", cex = 2, pch = 20)
Based on the above plot, we can see that the lowest Test MSE value occurs when using 8 cuts.
plot(wage ~ age, data = data1, col = "darkgrey")
agelims <- range(data1$age)
age.grid <- seq(from = agelims[1], to = agelims[2])
fit <- glm(wage ~ cut(age, 8), data = data1)
preds <- predict(fit, data.frame(age = age.grid))
lines(age.grid, preds, col = "green", lwd = 2)
Above is the plot for the step function fit to the data.
This question relates to the College data set.
(a) Split the data into a training set and a test set. Using out-of-state tuition as the response and the other variables as the predictors, perform forward stepwise selection on the training set in order to identify a satisfactory model that uses just a subset of the predictors.
data2 =College
set.seed(100)
index = sample(nrow(data2), nrow(data2)*0.8)
train = data2[index,]
test = data2[-index,]
fit <- regsubsets(Outstate ~ ., data = train, nvmax = 17, method = "forward")
fit.summary <- summary(fit)
par(mfrow = c(1, 3))
plot(fit.summary$cp, xlab = "Number of variables", ylab = "Cp", type = "l")
min.cp <- min(fit.summary$cp)
std.cp <- sd(fit.summary$cp)
abline(h = min.cp + 0.2 * std.cp, col = "red", lty = 2)
abline(h = min.cp - 0.2 * std.cp, col = "red", lty = 2)
plot(fit.summary$bic, xlab = "Number of variables", ylab = "BIC", type='l')
min.bic <- min(fit.summary$bic)
std.bic <- sd(fit.summary$bic)
abline(h = min.bic + 0.2 * std.bic, col = "red", lty = 2)
abline(h = min.bic - 0.2 * std.bic, col = "red", lty = 2)
plot(fit.summary$adjr2, xlab = "Number of variables", ylab = "Adjusted R2", type = "l", ylim = c(0.4, 0.84))
max.adjr2 <- max(fit.summary$adjr2)
std.adjr2 <- sd(fit.summary$adjr2)
abline(h = max.adjr2 + 0.2 * std.adjr2, col = "red", lty = 2)
abline(h = max.adjr2 - 0.2 * std.adjr2, col = "red", lty = 2)
Based on the Cp, BIC, and adj \(R^2\) we can see that 6 is the optimal number of variables to include in our model.
fit <- regsubsets(Outstate ~ ., data = data2, method = "forward")
coeffs <- coef(fit, id = 6)
names(coeffs)
## [1] "(Intercept)" "PrivateYes" "Room.Board" "PhD" "perc.alumni"
## [6] "Expend" "Grad.Rate"
(b) Fit a GAM on the training data, using out-of-state tuition as the response and the features selected in the previous step as the predictors. Plot the results, and explain your findings.
fit <- gam(Outstate ~ Private + s(Room.Board, df=2) + s(perc.alumni, df=2) +
s(PhD, df=2) + s(Expend, df=2) + s(Grad.Rate, df=2), data=train)
par(mfrow=c(2, 3))
plot(fit, se=TRUE, col="blue")
(c) Evaluate the model obtained on the test set, and explain the results obtained.
preds <- predict(fit, test)
MSE <- mean((test$Outstate - preds)^2)
MSE
## [1] 3503217
tss <- mean((test$Outstate - mean(test$Outstate))^2)
rss <- 1 - MSE / tss
rss
## [1] 0.77415
We obtain a test \(R^2\) of 0.77415 using GAM with 6 predictors.
(d) For which variables, if any, is there evidence of a non-linear relationship with the response?
summary(fit)
##
## Call: gam(formula = Outstate ~ Private + s(Room.Board, df = 2) + s(perc.alumni,
## df = 2) + s(PhD, df = 2) + s(Expend, df = 2) + s(Grad.Rate,
## df = 2), data = train)
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -7577.35 -1158.57 10.38 1280.72 8896.24
##
## (Dispersion Parameter for gaussian family taken to be 3673569)
##
## Null Deviance: 10130673051 on 620 degrees of freedom
## Residual Deviance: 2237202639 on 608.9998 degrees of freedom
## AIC: 11163.66
##
## Number of Local Scoring Iterations: NA
##
## Anova for Parametric Effects
## Df Sum Sq Mean Sq F value Pr(>F)
## Private 1 2967958313 2967958313 807.922 < 2.2e-16 ***
## s(Room.Board, df = 2) 1 2181592923 2181592923 593.862 < 2.2e-16 ***
## s(perc.alumni, df = 2) 1 787193426 787193426 214.286 < 2.2e-16 ***
## s(PhD, df = 2) 1 380388969 380388969 103.547 < 2.2e-16 ***
## s(Expend, df = 2) 1 667015965 667015965 181.572 < 2.2e-16 ***
## s(Grad.Rate, df = 2) 1 73074653 73074653 19.892 9.759e-06 ***
## Residuals 609 2237202639 3673569
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Anova for Nonparametric Effects
## Npar Df Npar F Pr(F)
## (Intercept)
## Private
## s(Room.Board, df = 2) 1 2.239 0.13506
## s(perc.alumni, df = 2) 1 0.432 0.51148
## s(PhD, df = 2) 1 3.726 0.05405 .
## s(Expend, df = 2) 1 60.290 3.464e-14 ***
## s(Grad.Rate, df = 2) 1 3.776 0.05245 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Based on our ANOVA analysis, we can see that Expend, Phd, and Grad.Rate have evidence of a non-linear relationship if we assume a p-value of 0.06.