R for Marketing Research and Analytics
Chris Chapman and Elea McDonnell Feit
January 2016
Load CRM data
Converting data to factors
Basic scatterplot
Let's look at scatterplots. How does age relate to credit score?
plot(x=cust.df$age, y=cust.df$credit.score)
A better plot
Add color, labels, and adjust the axis limits:
plot(cust.df$age, cust.df$credit.score,
col="blue",
xlim=c(15, 55), ylim=c(500, 900),
main="Active Customers as of June 2014",
xlab="Customer Age (years)", ylab="Credit Score ")
Add a regression line
abline() adds a regression line from a linear model (we'll discuss regression in depth later)
plot(cust.df$age, cust.df$credit.score,
col="blue", xlim=c(15, 55), ylim=c(500, 900),
xlab="Customer Age (years)", ylab="Credit Score ")
abline(lm(cust.df$credit.score ~ cust.df$age))
Zero-inflated and skewed data
Scatterplot with skew
How does in-store spending relate to online spending? It's hard to tell because both are skewed! The apparent negative slope is very misleading.
plot(cust.df$store.spend, cust.df$online.spend,
xlab="Prior 12 months in-store sales ($)",
ylab="Prior 12 months online sales ($)")
Looking at the skew
A histogram reveals the skew, e.g., for in-store spending:
hist(cust.df$store.spend,
breaks=(0:ceiling(max(cust.df$store.spend)/10))*10,
xlab="Prior 12 months online sales ($)" )
Using logarithmic axes
Use the log= argument to set one or both axes to logarithmic. Caution: log(x \( \le \) 0) is not defined, so make sure values are positive. Add a constant such as +1 if needed. Now we see little or no relationship:
plot(cust.df$store.spend + 1, cust.df$online.spend + 1,
log="xy")
Multi-panel plots
How does distance relate to in-store and online spending? Use par(mfrow=c(ROWS, COLUMNS)) to put multiple plots together:
par(mfrow=c(2, 2))
with(cust.df, plot(distance.to.store, store.spend))
with(cust.df, plot(distance.to.store, online.spend))
with(cust.df, plot(distance.to.store, store.spend+1, log="xy"))
with(cust.df, plot(distance.to.store, online.spend+1, log="xy"))
Scatterplot matrix: Quick 2-way Visualization
pairs(formula = ~ age + credit.score + distance.to.store +
online.spend + store.trans + store.spend,
data=cust.df)
Fancy alternative: scatterplotMatrix in "car"
library(car) # install if needed
scatterplotMatrix(formula = ~ age + credit.score +
distance.to.store + online.spend +
store.trans + store.spend,
data=cust.df, diagonal="histogram")
Correlation
Correlation matrix
Redoing that with complete cases
Visualize correlation matrix
Use the “corrplot” package. See book (and help) for more options.
library(corrplot) # install if needed
corrplot(corr=cor(cust.df[ , c(2, 3, 5:12)],
use="complete.obs"),
method ="ellipse")
Data Transformation
Polychoric correlation
With ordinal variables such as satisfaction ratings, consider polychoric correlation instead of Pearson's r. We need it because the ratings are not fine grained, and plot on top of one another:
plot(cust.df$sat.service, cust.df$sat.selection,
xlab="Sat, Service", ylab="Sat, Selection")
Polychoric correlation test
Exercise!
Answers (1)
Plot salary vs. years since PhD.
with(Salaries, plot(yrs.since.phd, salary))
Answers (2)
Answers (3)
Draw a visualization of all bivariate relationships
library(car)
scatterplotMatrix(Salaries) # could use pairs() instead
