R Quick Tour

Sawtooth Software Conference
Orlando, FL

These slides: https://goo.gl/ACdG6B
Slide code: https://goo.gl/PjBceN

Key Points

• It's where new methods are developed first
• Over 13000 add on packages and tools
• It is open source and free to use (except a very few packages)
• Don't think of R as a “statistics program.''
• R is primarily a programming language

• For the latest statistics methods
• For maximum power to develop your own analyses
• It excels for iteration and automation
• Large community support for advanced analytics
• It's free
• Includes integrating reporting capabilities (such as these slides)

• It requires programming to do most analyses
• Routine analyses are more difficult at first
• Charting and plotting are not easy, and tend to be rather basic unless they are tweaked substantially

• R base
• R Studio

Switch to RStudio and look at the elements of the R IDE (integrated development environment)

We'll use a data set from Chapman & Feit (2015).

• CSat survey with 4 items
• 2 Satisfaction items: with Product & Sales
• 2 Likelihood to recommend items: Product & Sales
• Each item assessed on 7 point Likert scale

Let's switch to R and walk through some typical code …

satData <- read.csv("http://r-marketing.r-forge.r-project.org/data/rintro-chapter2.csv")

# examine the data
head(satData)
str(satData)
summary(satData)

# convert Segment to a factor variable
satData$Segment <- factor(satData$Segment)
summary(satData)

# Satisfaction by segment
library(ggplot2)
by(satData$iProdSAT, satData$Segment, mean_se, mult=1.96)


###### EXAMPLE OF PLOTTING

### first aggregate the data we want to plot
# ... there are other ways to do this, but this illustrates data frame manipulation

# get the mean and standard errors
prod.sat.seg <- aggregate(satData$iProdSAT, list(satData$Segment), mean_se, mult=1.96)
str(prod.sat.seg)

# coerce those to a nice data frame
prod.sat.seg <- data.frame(prod.sat.seg$Group.1, lapply(data.frame(prod.sat.seg$x), unlist))
prod.sat.seg

# and label the columns to be more readable
names(prod.sat.seg) <- c("Segment", "average", "lowerCI", "upperCI")
prod.sat.seg

### the plot itself ...

# now plot the interquartile range
p <- ggplot(data=prod.sat.seg, 
            aes(x=Segment, y=average, ymax=upperCI, ymin=lowerCI)) +
     geom_point() +
     geom_errorbar()

p

# color the points and make them larger
p <- ggplot(data=prod.sat.seg, 
            aes(x=Segment, y=average, ymax=upperCI, ymin=lowerCI)) +
     geom_point(aes(color=Segment), size=3) +      # <=======
     geom_errorbar() 

p

# color the error bars and make them narrower
p <- ggplot(data=prod.sat.seg, 
            aes(x=Segment, y=average, ymax=upperCI, ymin=lowerCI)) +
     geom_point(aes(color=Segment), size=3) +
     geom_errorbar(aes(color=Segment), width=0.3)      # <=======

p

# adjust the Y axis range
p <- ggplot(data=prod.sat.seg, 
            aes(x=Segment, y=average, ymax=upperCI, ymin=lowerCI)) +
     geom_point(aes(color=Segment), size=3) +
     geom_errorbar(aes(color=Segment), width=0.3) +
     coord_cartesian(ylim=c(1, 5))    # <========

p

# add some titles to be more readable
p <- ggplot(data=prod.sat.seg, 
            aes(x=Segment, y=average, ymax=upperCI, ymin=lowerCI)) +
     geom_point(aes(color=Segment), size=3) +
     geom_errorbar(aes(color=Segment), width=0.3) +
     coord_cartesian(ylim=c(1, 5)) +
     ggtitle("Average Sat and Confidence Interval by Segment") +    # <=====
     ylab("Mean satisfaction and 95% CI")

p


########## CORRELATIONS

# correlation matrix
library(corrplot)
corrplot(cor(satData[ , -3]))

# tinker with the plot
corrplot.mixed(cor(satData[ , -3]))


#### ADVANCED MODELING. EXAMPLE: SAT/REC STRUCTURAL MODEL

# define a structural model for Satisfaction and Recommendation
satModel <- "SAT =~ iProdSAT + iSalesSAT
             REC =~ iProdREC + iSalesREC
             REC ~  SAT "

# fit the structural model
library(lavaan)
sat.fit <- cfa(satModel, data=satData)

# look at the fit
summary(sat.fit, fit.measures=TRUE)

# plot the structural model
library(semPlot)
semPaths(sat.fit, what="est", nCharNodes=9, residuals=FALSE)

Pro
1. Extreme power and precision. It does exactly what you want.
2. Complete flexibility at every step
3. Once script is done it is reusable and re-runnable

Con
1. It only does exactly what you tell it. Defaults are missing or ugly.
2. You have to write code.
3. Everything takes longer … the first time.

To learn R, you will need:

• Skill or genuine interest in learning to program
• A real data problem that you need to solve
• Routine methods that you understand well (e.g., ANOVA, lm)
• Time to work through the steep learning curve
• Discipline to force yourself to finish the project!
• Ability to iterate on successive projects, as it gets easier and easier

Only try advanced/new methods after you are fluent in the basics.

Thank you!

Chris Chapman cchapman@google.com

Kenneth Fairchild kenneth@sawtoothsoftware.com