Scribd
Upload
181 views

Simple Linear Regression

This document describes using simple linear regression to predict median house values from the percentage of low-income households using the Boston housing dataset. The author splits the data into training and test sets, transforms the predictor variable to improve linearity, fits a linear regression model on the training data, assesses the model fit, makes predictions on new data, and calculates the mean squared error to evaluate model performance on the test set.

Uploaded by

ramesh158
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
181 views

Simple Linear Regression

This document describes using simple linear regression to predict median house values from the percentage of low-income households using the Boston housing dataset. The author splits the data into training and test sets, transforms the predictor variable to improve linearity, fits a linear regression model on the training data, assesses the model fit, makes predictions on new data, and calculates the mean squared error to evaluate model performance on the test set.

Uploaded by

ramesh158
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

DSO 530: Simple Linear Regression

Abbass Al Sharif
Predicting House Value from Percent of Low Income Household
We are going to use a dataset called Boston which is part of the MASS package. It recordes the median value
of houses for 506 neighborhoods around Boston. Our task is to predict the median house value (medv) using
only one predictor (lstat: percent of households with low socioeconomic status).

# load MASS package


library(MASS)

#split the data by using the first 400 observations as the training data and the remaining as the testin
train = 1:400
test = -train
# speficy that we are going to use only two variables (lstat and medv)
variables = which(names(Boston) ==c("lstat", "medv"))
training_data = Boston[train, variables]
testing_data = Boston[test, variables]
# check the dimensions of the new data sets
dim(training_data)
STEP 1: Split the dataset
## [1] 400

dim(testing_data)
## [1] 106

STEP 2: Check for Linearity In order to perfom linear regression in R, we will use a function called
lm() which stands for linear models. We will use it to fit a simple linear regression with medv as the response
(dependent variable) and lstat as the predictor or independent variable, and then save it in a place called
model. Notice that the formula for lm in R is in this format y~x, where y is the dependent variable, and x is
the indepenent variable. But before we run our model, lets visually check if the relationship between x and y
is linear.
#attach training data
attach(training_data)
#scatterplot of lstat vs. medv
plot(lstat, medv)

50
40
30
10

20

medv

10

20

30

lstat
According to the plot, we see that the relationship is a not linear. Lets transorm our explanatory variable
lstat.
#scatterplot of log(lstat) vs. medv

30
10

20

medv

40

50

plot(log(lstat), medv)

0.5

1.0

1.5

2.0

2.5

3.0

3.5

log(lstat)
STEP 3: Run the linear regression model Look at the plot, it is more linear, so we can proceed and
perform lm():

model = lm(medv ~ log(lstat), data = training_data)


model
##
##
##
##
##
##
##

Call:
lm(formula = medv ~ log(lstat), data = training_data)
Coefficients:
(Intercept)
log(lstat)
51.8
-12.2

Notice that some basic information is shown to us when we print model. This only gave us the slope (-12.2)
and the intercept (51.8) of the linear model. Note that here we are looking at log(lstat) and not lstat
anymore. So for every one unit increase in lstat, the median value of the house will decrease by e(12.2).
For more detailed information, we can use the summary() function:
summary(model)
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##
##

Call:
lm(formula = medv ~ log(lstat), data = training_data)
Residuals:
Min
1Q
-11.385 -3.908

Median
-0.779

3Q
2.245

Max
25.728

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept)
51.783
1.097
47.2
<2e-16 ***
log(lstat)
-12.203
0.472
-25.9
<2e-16 ***
--Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 5.6 on 398 degrees of freedom
Multiple R-squared: 0.627, Adjusted R-squared: 0.626
F-statistic: 669 on 1 and 398 DF, p-value: <2e-16

Now, we have access to p-values and standard errors for the coefficients, as well as the R2. The output states
that the slope is statistically significant and different from 0 and with a t-value= -25.9 (p-value <0.05), which
means that there is a significant relationship between the percentage of households with low socioeconomic
income and the median house value. This relationship is negative. That is as the percantage of household
with low socioeconomic income increases, the median house value decreases.
Looking at R2, we can deduce that 62.7% of the model variation is being explained by the predictor
log(lstat). This is probably low, but indeed it would increase if we had more independent (explanatory)
variables.
We can use the names() function to see what other peices of information are stored in our linear model
(model).
names(model)

##
##
##

[1] "coefficients" "residuals"


[5] "fitted.values" "assign"
[9] "xlevels"
"call"

"effects"
"qr"
"terms"

"rank"
"df.residual"
"model"

model$coefficients
## (Intercept)
##
51.78

log(lstat)
-12.20

To obtain the confidence intervel for the linear model (model), we can use the confint() function:
confint(model, level = 0.95)
##
2.5 % 97.5 %
## (Intercept) 49.63 53.94
## log(lstat) -13.13 -11.28
So, a 95% confidence interval for the slope of log(lstat) is (-13.13, -11.28). Notice that this confidence
interval gives us the same result as the hypothesis test performed earlier, by stating that we are 95% confident
that the slope of lstat is not zero (in fact it is less than zero, which means that the relationship is negative.)
STEP 4: Plot the regression model

Now, lets plot our regression line on top of our data.

#scatterplot of lstat vs. medv


plot(log(lstat), medv)

30
10

20

medv

40

50

# Add the regression line to the existing scatterplot


abline(model)

0.5

1.0

1.5

2.0
log(lstat)

Lets play with the look of the plot, and makes it perttier!
4

2.5

3.0

3.5

#scatterplot of lstat vs. medv


plot(log(lstat), medv,
xlab = "Log Transform of % of Houshold with Low Socioeconomic Income",
ylab = "Median House Value",
col = "red",
pch = 20)

40
30
20
10

Median House Value

50

# Make the line color blue, and the line's width =3 (play with the width!)
abline(model, col = "blue", lwd =3)

0.5

1.0

1.5

2.0

2.5

3.0

3.5

Log Transform of % of Houshold with Low Socioeconomic Income


Do you want to see what point character (pch) you want to choose? Try this:
plot(1:20, 1:20, pch = 1:20)

20
15
10
5

1:20

10

15

20

1:20
STEP 5: Assess your model! Final thing to do would be to predict using our fitted model. We can use
the predict() function for this purpose:
#predict what is the median value of the house with lstat= 5%
predict(model, data.frame(lstat = c(5)))
##
1
## 32.14
#predict what is the median values of houses with lstat= 5%, 10%, and 15%
predict(model, data.frame(lstat = c(5,10,15), interval = "prediction"))
##
1
2
3
## 32.14 23.68 18.74
Now lets assess our model, by computing th mean squared error (MSE). To assess the model we created,
then we will be using the test data!
#save the testing median values for houses (testing y) in y
y = testing_data$medv
#compute the predicted value for this y (y hat)
y_hat = predict(model, data.frame(lstat = testing_data$lstat))
#Now we have both y and y_hat for our testing data. let's find the mean square error
error = y-y_hat
error_squared = error^2
MSE = mean(error_squared)
MSE
6

## [1] 17.7
The mean squared error for our model is 17.7. This number will be useful later on, when we create another
model using the same data but multiple variables. Do you think this MSE would go lower or higher for this
dataset? That would be our next tast!

You might also like