In this post, we will look at how to add a regression line to a plot using the “ggplot2” package. This is mostly a review of what we learned in the post on adding a LOESS line to a plot. The main difference is that a regression line is a straight line that represents the relationship between the x and y variable while a LOESS line is used mostly to identify trends in the data.

One new wrinkle we will add to this discussion is the use of faceting when developing plots. Faceting is the development of multiple plots simultaneously with each sharing different information about the data.

The data we will use is the “Housing” dataset from the “Ecdat” package. We will examine how lotsize affects housing price when also considering whether the house has central air conditioning or not. Below is the initial code in order to be prepared for analysis

library(ggplot2);library(Ecdat)

## Loading required package: Ecfun
## 
## Attaching package: 'Ecdat'
## 
## The following object is masked from 'package:datasets':
## 
##     Orange

data("Housing")

The first plot we will make is the basic plot of lotsize and price with the data being distinguished by having central air or not, without a regression line. The code is as follows

ggplot(data=Housing, aes(x=lotsize, y=price, col=airco))+geom_point()

We will now add the regression line to the plot. We will make a new plot with an additional piece of code.

ggplot(data=Housing, aes(x=lotsize, y=price, col=airco))+geom_point()+stat_smooth(method='lm')

As you can see we get two lines for the two conditions of the data. If we want to see the overall regression line we use the code that is shown below.

ggplot()+geom_point(data=Housing, aes(x=lotsize, y=price, col=airco))+stat_smooth(data=Housing, aes(x=lotsize, y=price ),method='lm')

We will now experiment with a technique called faceting. Faceting allows you to split the data by various subgroups and display the result via plot simultaneously. For example, below is the code for splitting the data by central air for examining the relationship between lot size and price.

ggplot(data=Housing, aes(lotsize, price, col=airco))+geom_point()+stat_smooth(method='lm')+facet_grid(.~airco)

By adding the “facet_grid” function we can subset the data by the categorical variable “airco”.

In the code below we have three plots. The first two show the relationship between lotsize and price based on central air and the last plot shows the overall relationship.

ggplot(data=Housing, aes(lotsize, price, col=airco))+geom_point()+stat_smooth(method="lm")+facet_grid(.~airco, margins = TRUE)

By adding the argument “margins” and setting it to true we are able to add the third plot that shows the overall results.

So far all of are facetted plots have had the same statistical transformation of the use of a regression. However, we can actually mix the type of transformations that happen when facetting the results. This is shown below.

ggplot(data=Housing, aes(lotsize, price, col=airco))+geom_point()+stat_smooth(data=subset(Housing, airco=="yes"))+stat_smooth(data=subset(Housing, airco=="no"), method="lm")+facet_grid(.~airco)

In the code we needed to use two functions of “stat_smooth” and indicate the information to transform inside the function. The plot to the left is a regression line with houses without central air and the plot to the right is a LOESS line with houses that have central air.

Conclusion

In this post, we explored the use of regression lines and advance faceting techniques. Communicating data with ggplot2 is one of many ways in which a data analyst can portray valuable information.

A common goal of statistics is to try and identify trends in the data as well as to predict what may happen. Both of these goals can be partially achieved through the development of graphs and or charts.

In this post, we will look at adding a smooth line to a scatterplot using the “ggplot2” package.

To accomplish this, we will use the “Carseats” dataset from the “ISLR” package. We will explore the relationship between the price of carseats with actual sales along with whether the carseat was purchase in an urban location or not. Below is some initial code to prepare for the analysis.

library(ggplot2);library(ISLR)
data("Carseats")

We are going to use a layering approach in this example. This means we will add one piece of code at a time until we have the complete plot.We are now going to plot the initial scatterplot. We simply want a scatterplot depicting the relationship between Price and Sales of carseats.

ggplot(data=Carseats, aes(x=Price, y=Sales, col=Urban))+geom_point()

The general trend appears to be negative. As price increases sales decrease regardless if carseat was purchase in an urban setting or not.

We will now add ar LOESS line to the graph. LOESS stands for “Locally weighted smoothing” this is a commonly used tool in regression analysis. The addition of a LOESS line allows in identifying trends visually much easily. Below is the code

ggplot(data=Carseats, aes(x=Price, y=Sales, col=Urban))+geom_point()+ 
        stat_smooth()

Unlike a regression line which is strictly straight, a LOESS line curves with the data. As you look at the graph the LOESS line is mostly straight with curves at the extremes and for a small rise in fall in the middle for carseats purchased in urban areas.

So far we have created LOESS lines by the categorical variable Urban. We can actually make a graph with three LOESS lines. One for Yes urban, another for No Urban, and the last one that is an overall line that does not take into account the Urban variable. Below is the code.

ggplot()+ geom_point(data=Carseats, aes(x=Price, y=Sales, col=Urban))+ stat_smooth(data=Carseats, aes(x=Price, y=Sales))+stat_smooth(data=Carseats, aes(x=Price, y=Sales, col=Urban))

Notice that the code is slightly different with the information being mostly outside of the “ggplot” function. You can barely see the third line in the graph but if you look closely you will see a new blue line that was not there previously. This is the overall trend line. If you want you can see the overall trend line with the code below.

ggplot()+ geom_point(data=Carseats, aes(x=Price, y=Sales, col=Urban))+ stat_smooth(data=Carseats, aes(x=Price, y=Sales))

The very first graph we generated in this post only contained points. This is because we used the “geom_point” function. Any of the graphs we created could be generated with points by removing the “geom_point” function and only using the “stat_smooth” function as shown below.

ggplot(data=Carseats, aes(x=Price, y=Sales, col=Urban))+ 
        stat_smooth()

Conclusion

This post provided an introduction to adding LOESS lines to a graph using ggplot2. For presenting data in a visually appealing way, adding lines can help in identifying key characteristics in the data.

In this post, we will explore the use of the “qplot” function from the “ggplot2” package. One of the major advantages of “ggplot” when compared to the base graphics package in R is that the “ggplot2” plots are much more visually appealing. This will make more sense when we explore the grammar of graphics. for now, we will just make plots to get used to using the “qplot” function.

We are going to use the “Carseats” dataset from the “ISLR” package in the examples. This dataset has data about the purchase of carseats for babies. Below is the initial code you will need to make the various plots.

library(ggplot2);library(ISLR)
data("Carseats")

In the first scatterplot, we are going to compare the price of a carseat with the volume of sales. Below is the code

qplot(Price, Sales,data=Carseats)

Most of this coding format you are familiar. “Price” is the x variable. “Sales” is the y variable and the data used is “Carseats. From the plot, we can see that as the price of the carseat increases there is normally a decline in the number of sales.

For our next plot, we will compare sales based on shelf location. This requires the use of a boxplot. Below is the code

qplot(ShelveLoc, Sales, data=Carseats, geom="boxplot")

The new argument in the code is the “geom” argument. This argument indicates what type of plot is drawn.

The boxplot appears to indicate that a “good” shelf location has the best sales. However, this would need to be confirmed with a statistical test.

Perhaps you are wondering how many of the Carseats were in the bad, medium, and good shelf locations. To find out, we will make a barplot as shown in the code below

qplot(ShelveLoc, data=Carseats, geom="bar")

The most common location was medium with bad and good be almost equal.

Lastly, we will now create a histogram using the “qplot” function. We want to see the distribution of “Sales”. Below is the code

qplot(Sales, data=Carseats, geom="histogram")

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

The distribution appears to be normal but again to know for certain requires a statistical test. For one last, trick we will add the median to the plot by using the following code

qplot(Sales, data=Carseats, geom="histogram") + geom_vline(xintercept = median(Carseats$Sales), colour="blue")

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

To add the median all we needed to do was add an additional argument called “geom_vline” which adds a line to a plot. Inside this argument, we had to indicate what to add by indicating the median of “Sales” from the “Carseats” package.

Conclusion

This post provided an introduction to the use of the “qplot” function in the “ggplot2” package. Understanding the basics of “qplor” is beneficial in providing visually appealing graphics

Data visualization is a critical component of communicate results with an audience. Fortunately, R provides many different ways to present numerical data it a clear way visually. This post will look specifically at making data visualizations with the base r package “graphics”.

Generally, functions available in the “graphics” package can be either high-level functions or low-level functions. High-level functions actually make the plot. Examples of high-level functions includes are the “hist” (histogram), “boxplot” (boxplot), and “barplot” (bar plot).

Low-level functions are used to add additional information to a plot. Some commonly used low-level functions includes “legend” (add legend) and “text” (add text). When coding we allows call high-level functions before low-level functions as the other way is not accepted by R.

We are going to begin with a simple graph. We are going to use the“Caschool” dataset from the “Ecdat” package. For now, we are going to plot the average expenditures per student by the average number of computers per student. Keep in mind that we are only plotting the data so we are only using a high-level function (plot). Below is the code

library(Ecdat)

data("Caschool")
plot(compstu~expnstu, data=Caschool)

The plot is not “beautiful” but it is a start in plotting data. Next, we are going to add a low-level function to our code. In particular, we will add a regression line to try and see the diretion of the relationship between the two variables via a straight line. In addition, we will use the “loess.smooth” function. This function will allow us to see the general shape of the data. The regression line is green and the loess smooth line is blue. The coding is mostly familiy but the “lwd” argument allows us to make the line thicker.

plot(compstu~expnstu, data=Caschool)
abline(lm(compstu~expnstu, data=Caschool), col="green", lwd=5)
lines(loess.smooth(Caschool$expnstu, Caschool$compstu), col="blue", lwd=5)

Boxplots allow you to show data that has been subsetted in some way. This allows for the comparisions of groups. In addition, one or more boxplots can be used to identify outliers.

In the plot below, the student-to-teacher ratio of k-6 and k-8 grades are displayed.

boxplot(str~grspan, data=Caschool)

As you look at the data you can see there is very little difference. However, one major differnce is that the K-8 group has much more extreme values than K-6.

Histograms are an excellent way to display information about one continuous variable. In the plot below, we can see the spread of the expenditure per student.

hist(Caschool$expnstu)

We will now add median to the plot by calling the low-level function “abline”. Below is the code.

hist(Caschool$expnstu)
abline(v=median(Caschool$expnstu), col="green", lwd=5)

Conclusion

In this post, we learned some of the basic structures of creating plots using the “graphics” package. All plots in include both low and high-level functions that work together to draw and provide additional information for communicating data in a visual manner