Tag Archives: regression

Best Subset Regression in R

In this post, we will take a look at best subset regression. Best subset regression fits a model for all possible feature or variable combinations and the decision for the most appropriate model is made by the analyst based on judgment or some statistical criteria.

Best subset regression is an alternative to both Forward and Backward stepwise regression. Forward stepwise selection adds one variable at a time based on the lowest residual sum of squares until no more variables continues to lower the residual sum of squares. Backward stepwise regression starts with all variables in the model and removes variables one at a time. The concern with stepwise methods is they can produce biased regression coefficients, conflicting models, and inaccurate confidence intervals.

Best subset regression bypasses these weaknesses of stepwise models by creating all models possible and then allowing you to assess which variables should be including in your final model. The one drawback to best subset is that a large number of variables means a large number of potential models, which can make it difficult to make a decision among several choices.

In this post, we will use the “Fair” dataset from the “Ecdat” package to predict marital satisfaction based on age, Sex, the presence of children, years married, religiosity, education, occupation, and number of affairs in the past year. Below is some initial code.


We begin our analysis by building the initial model with all variables in it. Below is the code

## Call:
## lm(formula = rate ~ ., data = Fair)
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -3.2049 -0.6661  0.2298  0.7705  2.2292 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  3.522875   0.358793   9.819  < 2e-16 ***
## sexmale     -0.062281   0.099952  -0.623  0.53346    
## age         -0.009683   0.007548  -1.283  0.20005    
## ym          -0.019978   0.013887  -1.439  0.15079    
## childyes    -0.206976   0.116227  -1.781  0.07546 .  
## religious    0.042142   0.037705   1.118  0.26416    
## education    0.068874   0.021153   3.256  0.00119 ** 
## occupation  -0.015606   0.029602  -0.527  0.59825    
## nbaffairs   -0.078812   0.013286  -5.932 5.09e-09 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## Residual standard error: 1.03 on 592 degrees of freedom
## Multiple R-squared:  0.1405, Adjusted R-squared:  0.1289 
## F-statistic:  12.1 on 8 and 592 DF,  p-value: 4.487e-16

The initial results are already interesting even though the r-square is low. When couples have children the have less martial satisfaction than couples without children when controlling for the other factors and this is the strongest regression weight. In addition, the more education a person has there is an increase in marital satisfaction. Lastly, as the number of affairs increases there is also a decrease in martial satisfaction. Keep in mind that the “rate” variable goes from 1 to 5 with one meaning a terrible marriage to five being a great one. The mean marital satisfaction was 3.52 when controlling for the other variables.

We will now create our subset models. Below is the code.


In the code above we create the sub models using the “regsubsets” function from the “leaps” package and saved it in the variable called “sub.fit”. We then saved the summary of “sub.fit” in the variable “best.summary”. We will use the “best.summary” “sub.fit variables several times to determine which model to use.

There are many different ways to assess the model. We will use the following statistical methods that come with the results from the “regsubset” function.

  • Mallow’ Cp
  • Bayesian Information Criteria

We will make two charts for each of the criteria above. The plot to the left will explain how many features to include in the model. The plot to the right will tell you which variables to include. It is important to note that for both of these methods, the lower the score the better the model. Below is the code for Mallow’s Cp.

plot(sub.fit,scale = "Cp")


The plot on the left suggest that a four feature model is the most appropriate. However, this chart does not tell me which four features. The chart on the right is read in reverse order. The high numbers are at the bottom and the low numbers are at the top when looking at the y-axis. Knowing this, we can conclude that the most appropriate variables to include in the model are age, children presence, education, and number of affairs. Below are the results using the Bayesian Information Criterion

plot(sub.fit,scale = "bic")


These results indicate that a three feature model is appropriate. The variables or features are years married, education, and number of affairs. Presence of children was not considered beneficial. Since our original model and Mallow’s Cp indicated that presence of children was significant we will include it for now.

Below is the code for the model based on the subset regression.

## Call:
## lm(formula = rate ~ age + child + education + nbaffairs, data = Fair)
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -3.2172 -0.7256  0.1675  0.7856  2.2713 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  3.861154   0.307280  12.566  < 2e-16 ***
## age         -0.017440   0.005057  -3.449 0.000603 ***
## childyes    -0.261398   0.103155  -2.534 0.011531 *  
## education    0.058637   0.017697   3.313 0.000978 ***
## nbaffairs   -0.084973   0.012830  -6.623 7.87e-11 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## Residual standard error: 1.029 on 596 degrees of freedom
## Multiple R-squared:  0.1352, Adjusted R-squared:  0.1294 
## F-statistic: 23.29 on 4 and 596 DF,  p-value: < 2.2e-16

The results look ok. The older a person is the less satisfied they are with their marriage. If children are presence the marriage is less satisfying. The more educated the more satisfied they are. Lastly, the higher the number of affairs indicate less marital satisfaction. However, before we get excited we need to check for collinearity and homoscedasticity. Below is the code

##       age     child education nbaffairs 
##  1.249430  1.228733  1.023722  1.014338

No issues with collinearity.For vif values above 5 or 10 indicate a problem. Let’s check for homoscedasticity



The normal qqplot and residuals vs leverage plot can be used for locating outliers. The residual vs fitted and the scale-location plot do not look good as there appears to be a pattern in the dispersion which indicates homoscedasticity. To confirm this we will use Breusch-Pagan test from the “lmtest” package. Below is the code

##  studentized Breusch-Pagan test
## data:  fit2
## BP = 16.238, df = 4, p-value = 0.002716

There you have it. Our model violates the assumption of homoscedasticity. However, this model was developed for demonstration purpose to provide an example of subset regression.


Logistic Regression in R

Logistic regression is used when the dependent variable is categorical with two choices. For example, if we want to predict whether someone will default of their loan. The dependent variable is categorical with two choices yes they default and no they do not.

Interpreting the output of a logistic regression analysis can be tricky. Basically, you need to interpret the odds ratio. For example, if the results of a study say the odds of default are 40% higher when someone is unemployed it is an increase in the likelihood of something happening. This is different from the probability which is what we normally use. Odds can go from any value from negative infinity to positive infinity. Probability is constrained to be anywhere from 0-100%.

We will now take a look at a simple example of logistic regression in R. We want to calculate the odds of defaulting on a loan. The dependent variable is “default” which can be either yes or no. The independent variables are “student” which can be yes or no, “income” which how much the person made, and “balance” which is the amount remaining on their credit card.

Below is the coding for developing this model.

The first step is to load the “Default” dataseat. This dataseat is a part of the “ISLR” package. Below is the code to get started


It is always good to examine the data first before developing a model. We do this by using the ‘summary’ function as shown below.

##  default    student       balance           income     
##  No :9667   No :7056   Min.   :   0.0   Min.   :  772  
##  Yes: 333   Yes:2944   1st Qu.: 481.7   1st Qu.:21340  
##                        Median : 823.6   Median :34553  
##                        Mean   : 835.4   Mean   :33517  
##                        3rd Qu.:1166.3   3rd Qu.:43808  
##                        Max.   :2654.3   Max.   :73554

We now need to check our two continous variables “balance” and “income” to see if they are normally distributed. Below is the code followed by the histograms.





The ‘income’ variable looks fine but there appears to be some problems with ‘balance’ to deal with this we will perform a square root transformation on the ‘balance’ variable and then examine it again by looking at a histogram. Below is the code.



As you can see this is much better looking.

We are now ready to make our model and examine the results. Below is the code.

Credit_Model<-glm(default~student+sqrt_balance+income, family=binomial, Default)
## Call:
## glm(formula = default ~ student + sqrt_balance + income, family = binomial, 
##     data = Default)
## Deviance Residuals: 
##     Min       1Q   Median       3Q      Max  
## -2.2656  -0.1367  -0.0418  -0.0085   3.9730  
## Coefficients:
##                Estimate Std. Error z value Pr(>|z|)    
## (Intercept)  -1.938e+01  8.116e-01 -23.883  < 2e-16 ***
## studentYes   -6.045e-01  2.336e-01  -2.587  0.00967 ** 
## sqrt_balance  4.438e-01  1.883e-02  23.567  < 2e-16 ***
## income        3.412e-06  8.147e-06   0.419  0.67538    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Dispersion parameter for binomial family taken to be 1)
##     Null deviance: 2920.6  on 9999  degrees of freedom
## Residual deviance: 1574.8  on 9996  degrees of freedom
## AIC: 1582.8
## Number of Fisher Scoring iterations: 9

The results indicate that the variable ‘student’ and ‘sqrt_balance’ are significant. However, ‘income’ is not significant. What all this means in simple terms is that being a student and having a balance on your credit card influence the odds of going into default while your income makes no difference. Unlike, multiple regression coefficients, the logistic coefficients require a transformation in order to interpret them The statistical reason for this is somewhat complicated. As such, below is the code to interpret the logistic regressuin coeffiecients.

##  (Intercept)   studentYes sqrt_balance       income 
## 3.814998e-09 5.463400e-01 1.558568e+00 1.000003e+00

To explain this as simply as possible. You subtract 1 from each coefficent to determine the actually odds. For example, if a person is a student the odds of them defaulting are 45% lower than when somebody is not a student when controlling for balance and income. Furthermore, for every 1 unit increase in the square root of the balance the odds of default go up by 55% when controlling for being a student and income. Naturally, speaking in terms of a 1 unit inrease in the square root of anything is confusing. However, we had to transform the variable in order to improve normality.


Logistic regression is one approach for predicting and modeling that involves a categorical dependent variable. Although the details are little confusing this approach is valuable at times when doing an analysis.

Assumption Check for Multiple Regression

The goal of the post is to attempt to explain the salary of a baseball based on several variables. We will see how to test various assumptions of multiple regression as well as deal with missing data. The first thing we need to do is load our data. Our data will come from the “ISLR” package and we will use the data set “Hitters”. There are 20 variables in the dataset as shown by the “str” function

#Load data 
## 'data.frame':    322 obs. of  20 variables:
##  $ AtBat    : int  293 315 479 496 321 594 185 298 323 401 ...
##  $ Hits     : int  66 81 130 141 87 169 37 73 81 92 ...
##  $ HmRun    : int  1 7 18 20 10 4 1 0 6 17 ...
##  $ Runs     : int  30 24 66 65 39 74 23 24 26 49 ...
##  $ RBI      : int  29 38 72 78 42 51 8 24 32 66 ...
##  $ Walks    : int  14 39 76 37 30 35 21 7 8 65 ...
##  $ Years    : int  1 14 3 11 2 11 2 3 2 13 ...
##  $ CAtBat   : int  293 3449 1624 5628 396 4408 214 509 341 5206 ...
##  $ CHits    : int  66 835 457 1575 101 1133 42 108 86 1332 ...
##  $ CHmRun   : int  1 69 63 225 12 19 1 0 6 253 ...
##  $ CRuns    : int  30 321 224 828 48 501 30 41 32 784 ...
##  $ CRBI     : int  29 414 266 838 46 336 9 37 34 890 ...
##  $ CWalks   : int  14 375 263 354 33 194 24 12 8 866 ...
##  $ League   : Factor w/ 2 levels "A","N": 1 2 1 2 2 1 2 1 2 1 ...
##  $ Division : Factor w/ 2 levels "E","W": 1 2 2 1 1 2 1 2 2 1 ...
##  $ PutOuts  : int  446 632 880 200 805 282 76 121 143 0 ...
##  $ Assists  : int  33 43 82 11 40 421 127 283 290 0 ...
##  $ Errors   : int  20 10 14 3 4 25 7 9 19 0 ...
##  $ Salary   : num  NA 475 480 500 91.5 750 70 100 75 1100 ...
##  $ NewLeague: Factor w/ 2 levels "A","N": 1 2 1 2 2 1 1 1 2 1 ...

We now need to assess the amount of missing data. This is important because missing data can cause major problems with different analysis. We are going to create a simple function that well explain to us the amount of missing data for each variable in the “Hitters” dataset. After using the function we need to use the “apply” function to display the results according to the amount of data missing by column and row.

Missing_Data <- function(x){sum(is.na(x))/length(x)*100}
##     AtBat      Hits     HmRun      Runs       RBI     Walks     Years 
##   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000 
##    CAtBat     CHits    CHmRun     CRuns      CRBI    CWalks    League 
##   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000 
##  Division   PutOuts   Assists    Errors    Salary NewLeague 
##   0.00000   0.00000   0.00000   0.00000  18.32298   0.00000

For column we can see that the missing data is all in the salary variable, which is missing 18% of its data. By row (not displayed here) you can see that a row might be missing anywhere from 0-5% of its data. The 5% is from the fact that there are 20 variables and there is only missing data in the salary variable. Therefore 1/20 = 5% missing data for a row. To deal with the missing data, we will us the ‘mice’ package. You can install it yourself and run the following code


##     AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun CRuns CRBI
## 263     1    1     1    1   1     1     1      1     1      1     1    1
##  59     1    1     1    1   1     1     1      1     1      1     1    1
##         0    0     0    0   0     0     0      0     0      0     0    0
##     CWalks League Division PutOuts Assists Errors NewLeague Salary   
## 263      1      1        1       1       1      1         1      1  0
##  59      1      1        1       1       1      1         1      0  1
##          0      0        0       0       0      0         0     59 59
Hitters1 <- mice(Hitters,m=5,maxit=50,meth='pmm',seed=500)


## Multiply imputed data set
## Call:
## mice(data = Hitters, m = 5, method = "pmm", maxit = 50, seed = 500)

In the code above we did the following

  1. loaded the ‘mice’ package Run the ‘md.pattern’ function Made a new variable called ‘Hitters’ and ran the ‘mice’ function on it.
  2. This function made 5 datasets  (m = 5) and used predictive meaning matching to guess the missing data point for each row (method = ‘pmm’).
  3. The seed is set for the purpose of reproducing the results The md.pattern function indicates that

There are 263 complete cases and 59 incomplete ones (not displayed). All the missing data is in the ‘Salary’ variable. The ‘mice’ function shares various information of how the missing data was dealt with. The ‘mice’ function makes five guesses for each missing data point. You can view the guesses for each row by the name of the baseball player. We will then select the first dataset as are new dataset to continue the analysis using the ‘complete’ function from the ‘mice’ package.

#View Imputed data


#Make Complete Dataset
completedData <- complete(Hitters1,1)

Now we need to deal with the normality of each variable which is the first assumption we will deal with. To save time, I will only explain how I dealt with the non-normal variables. The two variables that were non-normal were “salary” and “Years”. To fix these two variables I did a log transformation of the data. The new variables are called ‘log_Salary’ and “log_Years”. Below is the code for this with the before and after histograms

#Histogram of Salary


#log transformation of Salary
#Histogram of transformed salary


#Histogram of years
#Log transformation of Years completedData$log_Years<-log(completedData$Years) hist(completedData$log_Years)


We can now do are regression analysis and produce the residual plot in order to deal with the assumpotion of homoscedestacity and lineraity. Below is the code

Salary_Model<-lm(log_Salary~Hits+HmRun+Walks+log_Years+League, data=completedData)
#Residual Plot checks Linearity 

When using the ‘plot’ function you will get several plots. The first is the residual vs fitted which assesses linearity. The next is the qq plot which explains if are data is normally distributed. The scale location plot explains if there is equal variance. The residual vs leverage plot is used for finding outliers. All plots look good.


## Call:
## lm(formula = log_Salary ~ Hits + HmRun + Walks + log_Years + 
##     League, data = completedData)
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -2.1052 -0.3649  0.0171  0.3429  3.2139 
## Coefficients:
##              Estimate Std. Error t value Pr(>|t|)    
## (Intercept) 3.8790683  0.1098027  35.328  < 2e-16 ***
## Hits        0.0049427  0.0009928   4.979 1.05e-06 ***
## HmRun       0.0081890  0.0046938   1.745  0.08202 .  
## Walks       0.0063070  0.0020284   3.109  0.00205 ** 
## log_Years   0.6390014  0.0429482  14.878  < 2e-16 ***
## League2     0.1217445  0.0668753   1.820  0.06963 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## Residual standard error: 0.5869 on 316 degrees of freedom
## Multiple R-squared:  0.5704, Adjusted R-squared:  0.5636 
## F-statistic: 83.91 on 5 and 316 DF,  p-value: < 2.2e-16

Furthermore, the model explains 57% of the variance in salary. All varibles (Hits, HmRun, Walks, Years, and League) are significant at 0.1. Are last step is to find the correlations among the variables. To do this, we need to make a correlational matrix. We need to remove variables that are not a part of our study to do this. We also need to load the “Hmisc” package and use the ‘rcorr’ function to produce the matrix along with the p values. Below is the code

#find correlation
completedData1<-completedData;completedData1$Chits<-NULL;completedData1$CAtBat<-NULL;completedData1$CHmRun<-NULL;completedData1$CRuns<-NULL;completedData1$CRBI<-NULL;completedData1$CWalks<-NULL;completedData1$League<-NULL;completedData1$Division<-NULL;completedData1$PutOuts<-NULL;completedData1$Assists<-NULL; completedData1$NewLeague<-NULL;completedData1$AtBat<-NULL;completedData1$Runs<-NULL;completedData1$RBI<-NULL;completedData1$Errors<-NULL; completedData1$CHits<-NULL;completedData1$Years<-NULL; completedData1$Salary<-NULL


##            Hits HmRun Walks log_Salary log_Years
## Hits       1.00  0.56  0.64       0.47      0.13
## HmRun      0.56  1.00  0.48       0.36      0.14
## Walks      0.64  0.48  1.00       0.46      0.18
## log_Salary 0.47  0.36  0.46       1.00      0.63
## log_Years  0.13  0.14  0.18       0.63      1.00
## n= 322 
## P
##            Hits   HmRun  Walks  log_Salary log_Years
## Hits              0.0000 0.0000 0.0000     0.0227   
## HmRun      0.0000        0.0000 0.0000     0.0153   
## Walks      0.0000 0.0000        0.0000     0.0009   
## log_Salary 0.0000 0.0000 0.0000            0.0000   
## log_Years  0.0227 0.0153 0.0009 0.0000

There are no high correlations among our variables so multicolinearity is not an issue


This post provided an example dealing with missing data, checking the assumptions of a regression model, and displaying plots. All this was done using R.

Multiple Regression Prediction in R

In this post, we will learn how to predict using multiple regression in R. In a previous post, we learn how to predict with simple regression. This post will be a large repeat of this other post with the addition of using more than one predictor variable. We will use the “College” dataset and we will try to predict Graduation rate with the following variables

  • Student to faculty ratio
  • Percentage of faculty with PhD
  • Expenditures per student

Preparing the Data

First we need to load several packages and divide the dataset int training and testing sets. This is not new for this blog. Below is the code for this.

library(ISLR); library(ggplot2); library(caret)
 p=0.7, list=FALSE)
trainingset <- College[inTrain, ]
testingset <- College[-inTrain, ]
dim(trainingset); dim(testingset)

Visualizing the Data

We now need to get a visual idea of the data. Since we are use several variables the code for this is slightly different so we can look at several charts at the same time. Below is the code followed by the plots

> featurePlot(x=trainingset[,c("S.F.Ratio","PhD","Expend")],y=trainingset$Grad.Rate, plot="pairs")

To make these plots we did the following

  1. We used the ‘featureplot’ function told R to use the ‘trainingset’ data set and subsetted the data to use the three independent variables.
  2. Next, we told R what the y= variable was and told R to plot the data in pairs

Developing the Model

We will now develop the model. Below is the code for creating the model. How to interpret this information is in another post.

> TrainingModel <-lm(Grad.Rate ~ S.F.Ratio+PhD+Expend, data=trainingset)
> summary(TrainingModel)

As you look at the summary, you can see that all of our variables are significant and that the current model explains 18% of the variance of graduation rate.

Visualizing the Multiple Regression Model

We cannot use a regular plot because are model involves more than two dimensions.  To get around this problem to see are modeling, we will graph fitted values against the residual values. Fitted values are the predict values while residual values are the acutally values from the data. Below is the code followed by the plot.

> CheckModel<-train(Grad.Rate~S.F.Ratio+PhD+Expend, method="lm", data=trainingset)
> DoubleCheckModel<-CheckModel$finalModel
> plot(DoubleCheckModel, 1, pch=19, cex=0.5)

Here is what happen

  1. We created the variable ‘CheckModel’.  In this variable, we used the ‘train’ function to create a linear model with all of our variables
  2. We then created the variable ‘DoubleCheckModel’ which includes the information from ‘CheckModel’ plus the new column of’finalModel’
  3. Lastly, we plot ‘DoubleCheckModel’

The regression line was automatically added for us. As you can see, the model does not predict much but shows some linearity.

Predict with Model

We will now do one prediction. We want to know the graduation rate when we have the following information

  • Student-to-faculty ratio = 33
  • Phd percent = 76
  • Expenditures per Student = 11000

Here is the code with the answer

> newdata<-data.frame(S.F.Ratio=33, PhD=76, Expend=11000)
> predict(TrainingModel, newdata)

To put it simply, if the student-to-faculty ratio is 33, the percentage of PhD faculty is 76%, and the expenditures per student is 11,000, we can expect 57% of the students to graduate.


We will now test our model with the testing dataset. We will calculate the RMSE. Below is the code for creating the testing model followed by the codes for calculating each RMSE.

> TestingModel<-lm(Grad.Rate~S.F.Ratio+PhD+Expend, data=testingset)
> sqrt(sum((TrainingModel$fitted-trainingset$Grad.Rate)^2))
[1] 369.4451
> sqrt(sum((TestingModel$fitted-testingset$Grad.Rate)^2))
[1] 219.4796

Here is what happened

  1. We created the ‘TestingModel’ by using the same model as before but using the ‘testingset’ instead of the ‘trainingset’.
  2. The next to line of codes should look familiar.
  3. From this output the performance of the model improvement on the testing set since the RMSE is lower than compared to the training results.


This post attempted to explain how to predict and assess models with multiple variables. Although complex for some, prediction is a valuable statistical tool in many situations.

Using Regression for Prediction in R

In the last post about R, we looked at plotting information to make predictions. We will now look at an example of making predictions using regression.

We will used the same data as last time with the help of the ‘caret’ package as well. The code below sets up the seed and the training and testing sets we need.

> library(caret); library(ISLR); library(ggplot2)
> data("College");set.seed(1)
> PracticeSet<-createDataPartition(y=College$Grad.Rate, 
+                                  p=0.5, list=FALSE)
> TrainingSet<-College[PracticeSet, ]; TestingSet<-
+         College[-PracticeSet, ]
> head(TrainingSet)

The code above should look familiar from previous post.

Make the Scatterplot

We will now create the scatterplot showing the relationship between “S.F. Ratio” and “Grad.Rate” with the code below and the scatterplot.

> plot(TrainingSet$S.F.Ratio, TrainingSet$Grad.Rate, pch=5, col="green", 
xlab="Student Faculty Ratio", ylab="Graduation Rate")


Here is what we did

  1. We used the ‘plot’ function to make this scatterplot. The x variable was ‘S.F.Ratio’ of the ‘TrainingSet’ the y variable was ‘Grad.Rate’.
  2. We picked the type of dot to use using the ‘pch’ argument and choosing ’19’
  3. Next we chose a color and labeled each axis

Fitting the Model

We will now develop the linear model. This model will help us to predict future models. Furthermore, we will compare the model of the Training Set with the Test Set. Below is the code for developing the model.

> TrainingModel<-lm(Grad.Rate~S.F.Ratio, data=TrainingSet)
> summary(TrainingModel)

How to interpret this information was presented in a previous post. However, to summarize, we can say that when the student to faculty ratio increases one the graduation rate decreases 1.29. In other words, an increase in the student to faculty ratio leads to decrease in the graduation rate.

Adding the Regression Line to the Plot

Below is the code for adding the regression line followed by the scatterplot

> plot(TrainingSet$S.F.Ratio, TrainingSet$Grad.Rate, pch=19, col="green", xlab="Student Faculty Ratio", ylab="Graduation Rate")
> lines(TrainingSet$S.F.Ratio, TrainingModel$fitted, lwd=3)


Predicting New Values

With are model complete we can now predict values. For our example, we will only predict one value. We want to know what the graduation rate would be if we have a student to faculty ratio of 33. Below is the code for this with the answer

> newdata<-data.frame(S.F.Ratio=33)
> predict(TrainingModel, newdata)

Here is what we did

  1. We made a variable called ‘newdata’ and stored a data frame in it with a variable called ‘S.F.Ratio’ with a value of 33. This is x value
  2. Next, we used the ‘predict’ function from the ‘caret’ package to determine what the graduation rate would be if the student to faculty ratio is 33. To do this we told caret to use the ‘TrainingModel’ we developed using regression and to run this model with the information in the ‘newdata’ dataframe
  3. The answer was 40.68. This means that if the student to faculty ratio is 33 at a university then the graduation rate would be about 41%.

Testing the Model

We will now test the model we made with the training set with the testing set. First, we will make a visual of both models by using the “plot” function. Below is the code follow by the plots.

TrainingSet$Grad.Rate, pch=19, col=’green’,  xlab=”Student Faculty Ratio”, ylab=’Graduation Rate’)
lines(TrainingSet$S.F.Ratio,  predict(TrainingModel), lwd=3)
plot(TestingSet$S.F.Ratio,  TestingSet$Grad.Rate, pch=19, col=’purple’,
xlab=”Student Faculty Ratio”, ylab=’Graduation Rate’)
lines(TestingSet$S.F.Ratio,  predict(TrainingModel, newdata = TestingSet),lwd=3)


In the code, all that is new is the “par” function which allows us to see to plots at the same time. We also used the ‘predict’ function to set the plots. As you can see, the two plots are somewhat differ based on a visual inspection. To determine how much so, we need to calculate the error. This is done through computing the root mean square error as shown below.

> sqrt(sum((TrainingModel$fitted-TrainingSet$Grad.Rate)^2))
[1] 328.9992
> sqrt(sum((predict(TrainingModel, newdata=TestingSet)-TestingSet$Grad.Rate)^2))
[1] 315.0409

The main take away from this complicated calculation is the number 328.9992 and 315.0409. These numbers tell you the amount of error in the training model and testing model. The lower the number the better the model. Since the error number in the testing set is lower than the training set we know that our model actually improves when using the testing set. This means that our model is beneficial in assessing graduation rates. If there were problems we may consider using other variables in the model.


This post shared ways to develop a regression model for the purpose of prediction and for model testing.