itle: “Introduction to linear regression” |
utput: |
html_document: |
css: ./lab.css |
highlight: pygments |
theme: cerulean |
pdf_document: default |
The movie Moneyball focuses on the “quest for the secret of success in baseball”. It follows a low-budget team, the Oakland Athletics, who believed that underused statistics, such as a player’s ability to get on base, betterpredict the ability to score runs than typical statistics like home runs, RBIs (runs batted in), and batting average. Obtaining players who excelled in these underused statistics turned out to be much more affordable for the team.
In this lab we’ll be looking at data from all 30 Major League Baseball teams and examining the linear relationship between runs scored in a season and a number of other player statistics. Our aim will be to summarize these relationships both graphically and numerically in order to find which variable, if any, helps us best predict a team’s runs scored in a season.
Let’s load up the data for the 2011 season.
download.file("http://www.openintro.org/stat/data/mlb11.RData", destfile = "mlb11.RData")
load("mlb11.RData")
head(mlb11)
## team runs at_bats hits homeruns bat_avg strikeouts
## 1 Texas Rangers 855 5659 1599 210 0.283 930
## 2 Boston Red Sox 875 5710 1600 203 0.280 1108
## 3 Detroit Tigers 787 5563 1540 169 0.277 1143
## 4 Kansas City Royals 730 5672 1560 129 0.275 1006
## 5 St. Louis Cardinals 762 5532 1513 162 0.273 978
## 6 New York Mets 718 5600 1477 108 0.264 1085
## stolen_bases wins new_onbase new_slug new_obs
## 1 143 96 0.340 0.460 0.800
## 2 102 90 0.349 0.461 0.810
## 3 49 95 0.340 0.434 0.773
## 4 153 71 0.329 0.415 0.744
## 5 57 90 0.341 0.425 0.766
## 6 130 77 0.335 0.391 0.725
In addition to runs scored, there are seven traditionally used variables in the data set: at-bats, hits, home runs, batting average, strikeouts, stolen bases, and wins. There are also three newer variables: on-base percentage, slugging percentage, and on-base plus slugging. For the first portion of the analysis we’ll consider the seven traditional variables. At the end of the lab, you’ll work with the newer variables on your own.
runs
and one of the other numerical variables? Plot this relationship using the variable at_bats
as the predictor. Does the relationship look linear? If you knew a team’s at_bats
, would you be comfortable using a linear model to predict the number of runs? -A scatter plot would be the most useful type of plot in this scenerioplot(mlb11$runs ~ mlb11$at_bats, main = "Relationship between Runs and atBats", xlab = "At Bats", ylab = "Runs")
If the relationship looks linear, we can quantify the strength of the relationship with the correlation coefficient.
cor(mlb11$runs, mlb11$at_bats)
## [1] 0.610627
Think back to the way that we described the distribution of a single variable. Recall that we discussed characteristics such as center, spread, and shape. It’s also useful to be able to describe the relationship of two numerical variables, such as runs
and at_bats
above.
Just as we used the mean and standard deviation to summarize a single variable, we can summarize the relationship between these two variables by finding the line that best follows their association. Use the following interactive function to select the line that you think does the best job of going through the cloud of points.
plot_ss(x = mlb11$at_bats, y = mlb11$runs)
## Click two points to make a line.
## Call:
## lm(formula = y ~ x, data = pts)
##
## Coefficients:
## (Intercept) x
## -2789.2429 0.6305
##
## Sum of Squares: 123721.9
After running this command, you’ll be prompted to click two points on the plot to define a line. Once you’ve done that, the line you specified will be shown in black and the residuals in blue. Note that there are 30 residuals, one for each of the 30 observations. Recall that the residuals are the difference between the observed values and the values predicted by the line:
\[ e_i = y_i - \hat{y}_i \]
The most common way to do linear regression is to select the line that minimizes the sum of squared residuals. To visualize the squared residuals, you can rerun the plot command and add the argument showSquares = TRUE
.
plot_ss(x = mlb11$at_bats, y = mlb11$runs, showSquares = TRUE)
## Click two points to make a line.
## Call:
## lm(formula = y ~ x, data = pts)
##
## Coefficients:
## (Intercept) x
## -2789.2429 0.6305
##
## Sum of Squares: 123721.9
Note that the output from the plot_ss
function provides you with the slope and intercept of your line as well as the sum of squares.
plot_ss
, choose a line that does a good job of minimizing the sum of squares. Run the function several times. What was the smallest sum of squares that you got? How does it compare to your neighbors?plot_ss(x = mlb11$at_bats, y = mlb11$runs, showSquares = TRUE)
## Click two points to make a line.
## Call:
## lm(formula = y ~ x, data = pts)
##
## Coefficients:
## (Intercept) x
## -2789.2429 0.6305
##
## Sum of Squares: 123721.9
The smallest sum of squares that I got was 123721.9 with intercept -2789.2429 and coeff 0.6305
It is rather cumbersome to try to get the correct least squares line, i.e. the line that minimizes the sum of squared residuals, through trial and error. Instead we can use the lm
function in R to fit the linear model (a.k.a. regression line).
m1 <- lm(runs ~ at_bats, data = mlb11)
The first argument in the function lm
is a formula that takes the form y ~ x
. Here it can be read that we want to make a linear model of runs
as a function of at_bats
. The second argument specifies that R should look in the mlb11
data frame to find the runs
and at_bats
variables.
The output of lm
is an object that contains all of the information we need about the linear model that was just fit. We can access this information using the summary function.
summary(m1)
##
## Call:
## lm(formula = runs ~ at_bats, data = mlb11)
##
## Residuals:
## Min 1Q Median 3Q Max
## -125.58 -47.05 -16.59 54.40 176.87
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2789.2429 853.6957 -3.267 0.002871 **
## at_bats 0.6305 0.1545 4.080 0.000339 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 66.47 on 28 degrees of freedom
## Multiple R-squared: 0.3729, Adjusted R-squared: 0.3505
## F-statistic: 16.65 on 1 and 28 DF, p-value: 0.0003388
Let’s consider this output piece by piece. First, the formula used to describe the model is shown at the top. After the formula you find the five-number summary of the residuals. The “Coefficients” table shown next is key; its first column displays the linear model’s y-intercept and the coefficient of at_bats
. With this table, we can write down the least squares regression line for the linear model:
\[ \hat{y} = -2789.2429 + 0.6305 * atbats \]
One last piece of information we will discuss from the summary output is the Multiple R-squared, or more simply, \(R^2\). The \(R^2\) value represents the proportion of variability in the response variable that is explained by the explanatory variable. For this model, 37.3% of the variability in runs is explained by at-bats.
homeruns
to predict runs
. Using the estimates from the R output, write the equation of the regression line. What does the slope tell us in the context of the relationship between success of a team and its home runs? Lets follow the steps as the example above. -Plot a scatterplot -Find the correlation -Build a linear model m2plot(mlb11$runs ~ mlb11$at_bats, main = "Relationship between Runs and Home runs", xlab = "Home Runs", ylab = "Runs")
cor(mlb11$runs, mlb11$homeruns)
## [1] 0.7915577
There is a stronger positive reltionship considering that the correlation is .80 rounded up.
m2 <- lm(runs ~ homeruns, data = mlb11)
summary(m2)
##
## Call:
## lm(formula = runs ~ homeruns, data = mlb11)
##
## Residuals:
## Min 1Q Median 3Q Max
## -91.615 -33.410 3.231 24.292 104.631
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 415.2389 41.6779 9.963 1.04e-10 ***
## homeruns 1.8345 0.2677 6.854 1.90e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 51.29 on 28 degrees of freedom
## Multiple R-squared: 0.6266, Adjusted R-squared: 0.6132
## F-statistic: 46.98 on 1 and 28 DF, p-value: 1.9e-07
\[ \hat{y} = -415.2389 + 1.8345*homeruns \]
Let’s create a scatterplot with the least squares line laid on top.
plot(mlb11$runs ~ mlb11$at_bats)
abline(m1)
The function abline
plots a line based on its slope and intercept. Here, we used a shortcut by providing the model m1
, which contains both parameter estimates. This line can be used to predict \(y\) at any value of \(x\). When predictions are made for values of \(x\) that are beyond the range of the observed data, it is referred to as extrapolation and is not usually recommended. However, predictions made within the range of the data are more reliable. They’re also used to compute the residuals.
y<--2789.2429+0.6305*(5578)
y
## [1] 727.6861
We take the closest data point to 5,578 which is 5578 with 713. We simply take the difference to compute residuals.
x<-y-713
x
## [1] 14.6861
To assess whether the linear model is reliable, we need to check for (1) linearity, (2) nearly normal residuals, and (3) constant variability.
Linearity: You already checked if the relationship between runs and at-bats is linear using a scatterplot. We should also verify this condition with a plot of the residuals vs. at-bats. Recall that any code following a # is intended to be a comment that helps understand the code but is ignored by R.
plot(m1$residuals ~ mlb11$at_bats)
abline(h = 0, lty = 3) # adds a horizontal dashed line at y = 0
plot(m1$residuals ~ mlb11$at_bats)
abline(h = 0, lty = 3)
The plot indicates symmetry about the solid line. This implies some degree of linearity. On a deeper level, this can be verified with a test for non constant variance.
Nearly normal residuals: To check this condition, we can look at a histogram
hist(m1$residuals)
or a normal probability plot of the residuals.
qqnorm(m1$residuals)
qqline(m1$residuals) # adds diagonal line to the normal prob plot
Constant variability:
mlb11
that you think might be a good predictor of runs
. Produce a scatterplot of the two variables and fit a linear model. At a glance, does there seem to be a linear relationship? -Lets choose the batting average as a predictor variable. I suspect there to be a correlation between runs and batting average. Lets seecor(mlb11$runs, mlb11$bat_avg)
## [1] 0.8099859
There is avery high correlation coeff. Lets build a model
m3 <- lm(runs ~ bat_avg, data = mlb11)
plot(mlb11$runs ~ mlb11$bat_avg, main = "Relationship3")
abline(m3)
summary(m3)
##
## Call:
## lm(formula = runs ~ bat_avg, data = mlb11)
##
## Residuals:
## Min 1Q Median 3Q Max
## -94.676 -26.303 -5.496 28.482 131.113
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -642.8 183.1 -3.511 0.00153 **
## bat_avg 5242.2 717.3 7.308 5.88e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 49.23 on 28 degrees of freedom
## Multiple R-squared: 0.6561, Adjusted R-squared: 0.6438
## F-statistic: 53.41 on 1 and 28 DF, p-value: 5.877e-08
How does this relationship compare to the relationship between runs
and at_bats
? Use the R\(^2\) values from the two model summaries to compare. Does your variable seem to predict runs
better than at_bats
? How can you tell? When using bat average as a predictor,the r squared variable is 0.6561. When using at_bats as the predictor,the r square variable comes out to 0.3729. Cleary batting average is a better fit based on just the r square values. There is much more room for model validation. For example, you can partition the data into a test and training set and train a regression model on the training data. Test the trained model and its predictive power on the test data.
Now that you can summarize the linear relationship between two variables, investigate the relationships between runs
and each of the other five traditional variables. Which variable best predicts runs
? Support your conclusion using the graphical and numerical methods we’ve discussed (for the sake of conciseness, only include output for the best variable, not all five). We repeat the analysis for batting average, runs, homeruns, hits, and strikeouts. Based on the r square values collected from each of these, batting average still has the best predictive power and better model fit.
m3 <- lm(runs ~ bat_avg, data = mlb11)
hist(m3$residuals)
qqnorm(m3$residuals)
qqline(m3$residuals)
Now examine the three newer variables. These are the statistics used by the author of Moneyball to predict a teams success. In general, are they more or less effective at predicting runs that the old variables? Explain using appropriate graphical and numerical evidence. Of all ten variables we’ve analyzed, which seems to be the best predictor of runs
? Using the limited (or not so limited) information you know about these baseball statistics, does your result make sense?
Lets collect the names of the variables in question.
names(mlb11)
## [1] "team" "runs" "at_bats" "hits"
## [5] "homeruns" "bat_avg" "strikeouts" "stolen_bases"
## [9] "wins" "new_onbase" "new_slug" "new_obs"
This porblem presents us with the unique chance to do some more advance testing. If the goal is to find the best predictor of runs, then we should consider using variable selection process.
Lets examine the three new variables and any subset of the three new variables
library(olsrr)
## Warning: package 'olsrr' was built under R version 3.4.4
##
## Attaching package: 'olsrr'
## The following object is masked from 'package:datasets':
##
## rivers
model <- lm(runs ~ new_onbase+new_slug+new_obs, data = mlb11)
ols_step_all_possible(model)
## # A tibble: 7 x 6
## Index N Predictors `R-Square` `Adj. R-Square` `Mallow's Cp`
## * <int> <int> <chr> <dbl> <dbl> <dbl>
## 1 1 1 new_obs 0.935 0.933 2.92
## 2 2 1 new_slug 0.897 0.893 19.8
## 3 3 1 new_onbase 0.849 0.844 41.1
## 4 4 2 new_slug new_obs 0.938 0.933 3.74
## 5 5 2 new_onbase new_obs 0.937 0.932 4.08
## 6 6 2 new_onbase new_slug 0.935 0.930 4.83
## 7 7 3 new_onbase new_slu~ 0.941 0.935 4.00
We can visualize the best performing subset of predictors by plotting
model <- lm(runs ~ new_onbase+new_slug+new_obs, data = mlb11)
k <- ols_step_all_possible(model)
plot(k)
## Warning: package 'bindrcpp' was built under R version 3.4.4
Model number 7 indicates that it has the higher r square and one of the lowest CP values. A low CP value indicates better precision while a high r square indicates a good fit. The predictor new_obs by its self is just as comparable when used in the model. It only has a marginal lower r square variable than the model using all three new variables. Lets diagnose our models
modelA <- lm(runs ~ new_onbase+new_slug+new_obs, data = mlb11)
modelB <- lm(runs ~ new_obs, data = mlb11)
QQ line
qqnorm(modelA$residuals)
qqline(modelA$residuals)
qqnorm(modelB$residuals)
qqline(modelB$residuals)
This is a product of OpenIntro that is released under a Creative Commons Attribution-ShareAlike 3.0 Unported. This lab was adapted for OpenIntro by Andrew Bray and Mine Çetinkaya-Rundel from a lab written by the faculty and TAs of UCLA Statistics.