Using R and Python to Improve a Marketing Promotion
An exercise using R and Python to improve a marketing promotion in Major League Baseball.
Introduction
This is an exploration of improving the attendance of a Major League Baseball game by using data science. Using Major League Baseball fan attendance data from 2012, R and Python will be used to determine what night would be the best to run a marketing promotion to increase attendance.
Import Data
Let’s import the data, shall we?
# Load the readr package:
library(readr)
# Read the baseball data in from the csv file:
baseball <- read.csv("dodgers.csv", stringsAsFactors = FALSE)
Summarize
Now let’s take a quick look using the summary and str commands to make sure our import worked as expected and we’re seeing the data we expect:
# Show a summary to get an understanding of the results:
summary(baseball)
## month day attend day_of_week
## Length:81 Min. : 1.00 Min. :24312 Length:81
## Class :character 1st Qu.: 8.00 1st Qu.:34493 Class :character
## Mode :character Median :15.00 Median :40284 Mode :character
## Mean :16.14 Mean :41040
## 3rd Qu.:25.00 3rd Qu.:46588
## Max. :31.00 Max. :56000
## opponent temp skies day_night
## Length:81 Min. :54.00 Length:81 Length:81
## Class :character 1st Qu.:67.00 Class :character Class :character
## Mode :character Median :73.00 Mode :character Mode :character
## Mean :73.15
## 3rd Qu.:79.00
## Max. :95.00
## cap shirt fireworks bobblehead
## Length:81 Length:81 Length:81 Length:81
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
str(baseball)
## 'data.frame': 81 obs. of 12 variables:
## $ month : chr "APR" "APR" "APR" "APR" ...
## $ day : int 10 11 12 13 14 15 23 24 25 27 ...
## $ attend : int 56000 29729 28328 31601 46549 38359 26376 44014 26345 44807 ...
## $ day_of_week: chr "Tuesday" "Wednesday" "Thursday" "Friday" ...
## $ opponent : chr "Pirates" "Pirates" "Pirates" "Padres" ...
## $ temp : int 67 58 57 54 57 65 60 63 64 66 ...
## $ skies : chr "Clear " "Cloudy" "Cloudy" "Cloudy" ...
## $ day_night : chr "Day" "Night" "Night" "Night" ...
## $ cap : chr "NO" "NO" "NO" "NO" ...
## $ shirt : chr "NO" "NO" "NO" "NO" ...
## $ fireworks : chr "NO" "NO" "NO" "YES" ...
## $ bobblehead : chr "NO" "NO" "NO" "NO" ...
It looks like our data has twelve different variables, with the last four cap, shirt, fireworks, and bobblehead as promotions at those home games.
Variables
Now let’s create a few variables from this data set:
# First let's get a look at the names of our different columns of data that will become the variables:
for (i in 1:length(baseball)) {
column <- (names(baseball[i]))
print(column)
}
## [1] "month"
## [1] "day"
## [1] "attend"
## [1] "day_of_week"
## [1] "opponent"
## [1] "temp"
## [1] "skies"
## [1] "day_night"
## [1] "cap"
## [1] "shirt"
## [1] "fireworks"
## [1] "bobblehead"
Freebies!
Everybody loves free stuff. Let’s create variables for the promotional items, showing only the data for days when there was a promotional event:
promoCap <- subset(baseball, baseball$cap == "YES")
promoShirt <- subset(baseball, baseball$shirt == "YES")
promoFireworks <- subset(baseball, baseball$fireworks == "YES")
promoBobblehead <- subset(baseball, baseball$bobblehead == "YES")
Now let’s get a look at the output of the promotional variables:
promoCap
## month day attend day_of_week opponent temp skies day_night cap shirt
## 42 JUL 3 33884 Tuesday Reds 70 Cloudy Night YES NO
## 55 AUG 5 42495 Sunday Cubs 83 Clear Day YES NO
## fireworks bobblehead
## 42 NO NO
## 55 NO NO
promoShirt
## month day attend day_of_week opponent temp skies day_night cap shirt
## 12 APR 29 48753 Sunday Nationals 74 Clear Day NO YES
## 31 JUN 11 50559 Monday Angels 68 Clear Night NO YES
## 70 SEP 4 40619 Tuesday Padres 78 Clear Night NO YES
## fireworks bobblehead
## 12 NO NO
## 31 NO NO
## 70 NO NO
promoFireworks
## month day attend day_of_week opponent temp skies day_night cap shirt
## 4 APR 13 31601 Friday Padres 54 Cloudy Night NO NO
## 10 APR 27 44807 Friday Nationals 66 Clear Night NO NO
## 16 MAY 11 35591 Friday Rockies 65 Clear Night NO NO
## 21 MAY 18 40906 Friday Cardinals 64 Clear Night NO NO
## 24 MAY 25 36283 Friday Astros 59 Cloudy Night NO NO
## 34 JUN 15 40432 Friday White Sox 67 Clear Night NO NO
## 38 JUN 29 49763 Friday Mets 72 Clear Night NO NO
## 43 JUL 4 53570 Wednesday Reds 70 Clear Night NO NO
## 44 JUL 13 43873 Friday Padres 76 Cloudy Night NO NO
## 53 AUG 3 43537 Friday Cubs 73 Clear Night NO NO
## 62 AUG 24 39805 Friday Marlins 71 Clear Night NO NO
## 66 AUG 31 37622 Friday Snakes 77 Clear Night NO NO
## 73 SEP 14 40167 Friday Cardinals 85 Clear Night NO NO
## 76 SEP 28 37133 Friday Rockies 77 Clear Night NO NO
## fireworks bobblehead
## 4 YES NO
## 10 YES NO
## 16 YES NO
## 21 YES NO
## 24 YES NO
## 34 YES NO
## 38 YES NO
## 43 YES NO
## 44 YES NO
## 53 YES NO
## 62 YES NO
## 66 YES NO
## 73 YES NO
## 76 YES NO
promoBobblehead
## month day attend day_of_week opponent temp skies day_night cap shirt
## 11 APR 28 54242 Saturday Nationals 71 Clear Night NO NO
## 20 MAY 15 47077 Tuesday Snakes 70 Clear Night NO NO
## 28 MAY 29 51137 Tuesday Brewers 74 Clear Night NO NO
## 32 JUN 12 55279 Tuesday Angels 66 Cloudy Night NO NO
## 37 JUN 28 49006 Thursday Mets 75 Clear Night NO NO
## 40 JUL 1 55359 Sunday Mets 75 Clear Night NO NO
## 45 JUL 14 54014 Saturday Padres 75 Clear Night NO NO
## 51 JUL 31 52832 Tuesday Snakes 75 Cloudy Night NO NO
## 57 AUG 7 55024 Tuesday Rockies 80 Clear Night NO NO
## 60 AUG 21 56000 Tuesday Giants 75 Clear Night NO NO
## 65 AUG 30 54621 Thursday Snakes 80 Clear Night NO NO
## fireworks bobblehead
## 11 NO YES
## 20 NO YES
## 28 NO YES
## 32 NO YES
## 37 NO YES
## 40 NO YES
## 45 NO YES
## 51 NO YES
## 57 NO YES
## 60 NO YES
## 65 NO YES
Now let’s get a look at the summary of the promotional variables:
summary(promoCap)
## month day attend day_of_week
## Length:2 Min. :3.0 Min. :33884 Length:2
## Class :character 1st Qu.:3.5 1st Qu.:36037 Class :character
## Mode :character Median :4.0 Median :38190 Mode :character
## Mean :4.0 Mean :38190
## 3rd Qu.:4.5 3rd Qu.:40342
## Max. :5.0 Max. :42495
## opponent temp skies day_night
## Length:2 Min. :70.00 Length:2 Length:2
## Class :character 1st Qu.:73.25 Class :character Class :character
## Mode :character Median :76.50 Mode :character Mode :character
## Mean :76.50
## 3rd Qu.:79.75
## Max. :83.00
## cap shirt fireworks bobblehead
## Length:2 Length:2 Length:2 Length:2
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
summary(promoShirt)
## month day attend day_of_week
## Length:3 Min. : 4.00 Min. :40619 Length:3
## Class :character 1st Qu.: 7.50 1st Qu.:44686 Class :character
## Mode :character Median :11.00 Median :48753 Mode :character
## Mean :14.67 Mean :46644
## 3rd Qu.:20.00 3rd Qu.:49656
## Max. :29.00 Max. :50559
## opponent temp skies day_night
## Length:3 Min. :68.00 Length:3 Length:3
## Class :character 1st Qu.:71.00 Class :character Class :character
## Mode :character Median :74.00 Mode :character Mode :character
## Mean :73.33
## 3rd Qu.:76.00
## Max. :78.00
## cap shirt fireworks bobblehead
## Length:3 Length:3 Length:3 Length:3
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
summary(promoFireworks)
## month day attend day_of_week
## Length:14 Min. : 3.00 Min. :31601 Length:14
## Class :character 1st Qu.:13.00 1st Qu.:37255 Class :character
## Mode :character Median :16.50 Median :40300 Mode :character
## Mean :18.21 Mean :41078
## 3rd Qu.:26.50 3rd Qu.:43789
## Max. :31.00 Max. :53570
## opponent temp skies day_night
## Length:14 Min. :54.00 Length:14 Length:14
## Class :character 1st Qu.:65.25 Class :character Class :character
## Mode :character Median :70.50 Mode :character Mode :character
## Mean :69.71
## 3rd Qu.:75.25
## Max. :85.00
## cap shirt fireworks bobblehead
## Length:14 Length:14 Length:14 Length:14
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
summary(promoBobblehead)
## month day attend day_of_week
## Length:11 Min. : 1.00 Min. :47077 Length:11
## Class :character 1st Qu.:13.00 1st Qu.:51984 Class :character
## Mode :character Median :21.00 Median :54242 Mode :character
## Mean :19.64 Mean :53145
## 3rd Qu.:28.50 3rd Qu.:55152
## Max. :31.00 Max. :56000
## opponent temp skies day_night
## Length:11 Min. :66.00 Length:11 Length:11
## Class :character 1st Qu.:72.50 Class :character Class :character
## Mode :character Median :75.00 Mode :character Mode :character
## Mean :74.18
## 3rd Qu.:75.00
## Max. :80.00
## cap shirt fireworks bobblehead
## Length:11 Length:11 Length:11 Length:11
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
Exploration of Promotions
Looking at the promotions, we see that of the four different types of promotions the Dodgers ran during home games this season, they broke down as follows:
| Promotion Type | Promotion Frequency | Mean Attendance |
|---|---|---|
| Cap | 2 | 38,190 |
| Shirt | 3 | 46,644 |
| Fireworks | 14 | 41,078 |
| Bobblehead | 11 | 53,145 |
As a little synopsis of that information, it’s probably fair to say that fireworks at the conclusion of a game are pretty common. Lots of teams do this, and they probably aren’t a huge driver in the decision to take in a baseball game. The mean for attendance on games when fireworks for a promotion was 41,078. Not too shabby.
The cap and shirt promotions were pretty few and far between, with only five games offering a fan either of those giveaways. The mean attendance on cap giveaway games was 38,190 and on shirt giveaways it was 46,644.
But look at the bobbleheads. There were a total of eleven games when the Dodgers gave away a bobblehead, and the mean attendance during these games was 53,145. Dude. People love free stuff, and a bobblehead is just a really cool, quirky, and fun thing to have on your desk at work or home, eh?
Box Plots
Now let’s create a few box plots to get a look at our data. We’ll start with one that looks at attendance on the days of the week:
# Boxplot of day of the week attendance:
boxplot(attend~day_of_week,data=baseball, main="Day of the Week Attendance",
xlab="Day", ylab="Attendance")
That’s a bit messy so let’s order the days like a regular weekly calendar rather than using the default alphabetical order. Let’s also give the boxplot’s color something more applicable to our example, the use of dodgerblue:1
baseball$day_of_week <- factor(baseball$day_of_week , levels=c("Sunday", "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday"))
# Boxplot of day of the week attendance:
boxplot(attend~day_of_week,data=baseball, main="Day of the Week Attendance",
xlab="Day", ylab="Attendance", las=2, col="dodgerblue")
A few things stand out here. It looks like Mondays and Wednesdays are the days of the week where fans aren’t coming out for baseball. That makes sense because most people are either working or in school during the day, and at night might not come out due to getting an early start the next day for work or school.
Higher attendance on a Friday, Saturday, or Sunday makes sense.
Terrific Tuesday
What’s up with Tuesdays? Pulling in an average of 50,000 people on a Tuesday seems rather remarkable. Let’s keep pulling this thread. Let’s go in for a deeper dive on trying to see what drove that Tuesday attendance.
tuesday <- subset(baseball, baseball$day_of_week == "Tuesday")
print(tuesday)
## month day attend day_of_week opponent temp skies day_night cap shirt
## 1 APR 10 56000 Tuesday Pirates 67 Clear Day NO NO
## 8 APR 24 44014 Tuesday Braves 63 Cloudy Night NO NO
## 14 MAY 8 32799 Tuesday Giants 75 Clear Night NO NO
## 20 MAY 15 47077 Tuesday Snakes 70 Clear Night NO NO
## 28 MAY 29 51137 Tuesday Brewers 74 Clear Night NO NO
## 32 JUN 12 55279 Tuesday Angels 66 Cloudy Night NO NO
## 42 JUL 3 33884 Tuesday Reds 70 Cloudy Night YES NO
## 48 JUL 17 53498 Tuesday Phillies 70 Clear Night NO NO
## 51 JUL 31 52832 Tuesday Snakes 75 Cloudy Night NO NO
## 57 AUG 7 55024 Tuesday Rockies 80 Clear Night NO NO
## 60 AUG 21 56000 Tuesday Giants 75 Clear Night NO NO
## 70 SEP 4 40619 Tuesday Padres 78 Clear Night NO YES
## 80 OCT 2 42473 Tuesday Giants 83 Clear Night NO NO
## fireworks bobblehead
## 1 NO NO
## 8 NO NO
## 14 NO NO
## 20 NO YES
## 28 NO YES
## 32 NO YES
## 42 NO NO
## 48 NO NO
## 51 NO YES
## 57 NO YES
## 60 NO YES
## 70 NO NO
## 80 NO NO
summary(tuesday)
## month day attend day_of_week
## Length:13 Min. : 2.00 Min. :32799 Sunday : 0
## Class :character 1st Qu.: 7.00 1st Qu.:42473 Monday : 0
## Mode :character Median :12.00 Median :51137 Tuesday :13
## Mean :14.08 Mean :47741 Wednesday: 0
## 3rd Qu.:21.00 3rd Qu.:55024 Thursday : 0
## Max. :31.00 Max. :56000 Friday : 0
## Saturday : 0
## opponent temp skies day_night
## Length:13 Min. :63.00 Length:13 Length:13
## Class :character 1st Qu.:70.00 Class :character Class :character
## Mode :character Median :74.00 Mode :character Mode :character
## Mean :72.77
## 3rd Qu.:75.00
## Max. :83.00
##
## cap shirt fireworks bobblehead
## Length:13 Length:13 Length:13 Length:13
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
Looking things over here, it seems the Dodgers are heavily using promotional giveaways on Tuesday to bring fans out. In fact, of the 13 home games on Tuesdays in their 2012 season, the Dodgers gave fans a free cap (1 time), shirt (1 time), and bobbleheads (6 times) just for showing up (and I’m assuming that the promo items are limited to the first x amount of fans to arrive as this encourages people to get there early/on time to receive the promotional item).
Just like in our Exploration of Promotions area earlier, we see that promotions are important for briging out crowds, and fans especially are fans of bobbleheads.
Monday, Monday
Now let’s isolate Mondays and see if we can tell why the draw on that day is so low:
monday <- subset(baseball, baseball$day_of_week == "Monday")
print(monday)
## month day attend day_of_week opponent temp skies day_night cap shirt
## 7 APR 23 26376 Monday Braves 60 Cloudy Night NO NO
## 13 MAY 7 43713 Monday Giants 67 Clear Night NO NO
## 19 MAY 14 24312 Monday Snakes 67 Clear Night NO NO
## 27 MAY 28 38016 Monday Brewers 73 Clear Night NO NO
## 31 JUN 11 50559 Monday Angels 68 Clear Night NO YES
## 41 JUL 2 34493 Monday Reds 70 Clear Night NO NO
## 47 JUL 16 32238 Monday Phillies 67 Clear Night NO NO
## 50 JUL 30 33180 Monday Snakes 73 Clear Night NO NO
## 56 AUG 6 32659 Monday Rockies 79 Clear Night NO NO
## 59 AUG 20 36878 Monday Giants 80 Clear Night NO NO
## 69 SEP 3 33540 Monday Padres 84 Cloudy Night NO NO
## 79 OCT 1 33624 Monday Giants 86 Clear Night NO NO
## fireworks bobblehead
## 7 NO NO
## 13 NO NO
## 19 NO NO
## 27 NO NO
## 31 NO NO
## 41 NO NO
## 47 NO NO
## 50 NO NO
## 56 NO NO
## 59 NO NO
## 69 NO NO
## 79 NO NO
summary(monday)
## month day attend day_of_week
## Length:12 Min. : 1.00 Min. :24312 Sunday : 0
## Class :character 1st Qu.: 5.25 1st Qu.:32554 Monday :12
## Mode :character Median :12.50 Median :33582 Tuesday : 0
## Mean :13.42 Mean :34966 Wednesday: 0
## 3rd Qu.:20.75 3rd Qu.:37162 Thursday : 0
## Max. :30.00 Max. :50559 Friday : 0
## Saturday : 0
## opponent temp skies day_night
## Length:12 Min. :60.00 Length:12 Length:12
## Class :character 1st Qu.:67.00 Class :character Class :character
## Mode :character Median :71.50 Mode :character Mode :character
## Mean :72.83
## 3rd Qu.:79.25
## Max. :86.00
##
## cap shirt fireworks bobblehead
## Length:12 Length:12 Length:12 Length:12
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
With a mean attendance of 34,966, Mondays were the lowest attendance draw at Dodger Stadium in 2012. One thing really stand out to me as a good reason why, that being how many promotions were done on Mondays: of the 12 home games on Mondays in 2012, the Dodgers only had 1 promotional giveaway, a shirt on June 11th against the Angels. The other thing that stands out? The night they gave away that shirt to their fans they had the largest Monday night crowd of the season bringing in 50,559 people. Swag fills the seats!
Scatter Plots
Now let’s create a few scatter plots to get a look at our data. We’ll start with one that looks at attendance and the game time temperature to see if there’s any correlation there:
# Scatter plot of attendance and weather:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
baseball = pd.read_csv (r'dodgers.csv')
x = baseball.attend
y = baseball.temp
plt.scatter(x, y, c='c', alpha=0.5, s=150.00)
plt.title("Game Time Temps and Attendance")
plt.xlabel("Attendance")
plt.ylabel("Game Temperature")
plt.show()
Linear Regression Model
I’d like to do linear regression analysis using game attendance and game time temperature to see if there is any impact.
# Run our linear regression model:
simple.fit = lm(attend~temp, data=baseball)
summary(simple.fit)
##
## Call:
## lm(formula = attend ~ temp, data = baseball)
##
## Residuals:
## Min 1Q Median 3Q Max
## -16121 -6555 -1023 6348 15567
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 33819.22 8221.94 4.113 9.49e-05 ***
## temp 98.72 111.69 0.884 0.379
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 8309 on 79 degrees of freedom
## Multiple R-squared: 0.009791, Adjusted R-squared: -0.002743
## F-statistic: 0.7812 on 1 and 79 DF, p-value: 0.3795
A cursory glance at the regression output:
- Residuals: The section summarizes the residuals, the error between the prediction of the model and the actual results. Smaller residuals are better.
- Coefficients: For each variable and the intercept, a weight is produced and that weight has other attributes like the standard error, a t-test value and significance.
- Residual Standard Error: This is the standard deviation of the residuals. Smaller is better.
Based upon a quick look at things here, the game time temperature doesn’t really show that it impacts people coming out for games at Dodger Stadium.
Train/Test and Regression Testing
Let’s do some train/test splits on our data, again I will isolate weather:
import pandas as pd
from sklearn import linear_model
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as pltd
# Import the baseball file into a data frame:
baseballP = pd.read_csv (r'dodgers.csv')
print(baseballP)
## month day attend day_of_week ... cap shirt fireworks bobblehead
## 0 APR 10 56000 Tuesday ... NO NO NO NO
## 1 APR 11 29729 Wednesday ... NO NO NO NO
## 2 APR 12 28328 Thursday ... NO NO NO NO
## 3 APR 13 31601 Friday ... NO NO YES NO
## 4 APR 14 46549 Saturday ... NO NO NO NO
## .. ... ... ... ... ... .. ... ... ...
## 76 SEP 29 40724 Saturday ... NO NO NO NO
## 77 SEP 30 35607 Sunday ... NO NO NO NO
## 78 OCT 1 33624 Monday ... NO NO NO NO
## 79 OCT 2 42473 Tuesday ... NO NO NO NO
## 80 OCT 3 34014 Wednesday ... NO NO NO NO
##
## [81 rows x 12 columns]
columns = "temp".split()
print(columns)
## ['temp']
df = pd.DataFrame(baseballP, columns=columns)
print(df)
# define the attendand variable 'attend' (dependent variable) as y:
## temp
## 0 67
## 1 58
## 2 57
## 3 54
## 4 57
## .. ...
## 76 84
## 77 95
## 78 86
## 79 83
## 80 82
##
## [81 rows x 1 columns]
y = baseballP.attend
# Now we can use the train_test_split function in order to make the split. The test_size=0.2 inside the function indicates the percentage of the data that should be held over for testing. It’s usually around 80/20 or 70/30:
# Create the training and testing vars
X_train, X_test, y_train, y_test = train_test_split(df, y, test_size=0.2)
print(X_train.shape, y_train.shape)
## (64, 1) (64,)
print(X_test.shape, y_test.shape)
## (17, 1) (17,)
Now we’ll fit the model on the training data:
# fit a model
lm = linear_model.LinearRegression()
model = lm.fit(X_train, y_train)
predictions = lm.predict(X_test)
print(predictions)
## [42572.38727899 43091.38714324 42157.18738758 41845.78746903
## 41949.58744188 42053.38741473 41119.18765908 40288.78787627
## 44025.58689889 41119.18765908 40807.78774053 42572.38727899
## 40911.58771338 41845.78746903 42883.78719754 41119.18765908
## 41430.58757763]
Now let’s plot that training model:
## The line / model
plt.scatter(y_test, predictions)
plt.xlabel("True Values")
plt.ylabel("Predictions")
And our accuracy score:
print("Score:", model.score(X_test, y_test))
## Score: -0.17279789036241633
All of the training/test split info also yields points to temperature data not having a big impact on crowd size.
Final Recommendation
After going through the data in different ways, including scatter plots, box plots, and regression analysis, my main recommendation is for the Dodgers marketing department to plan to offer more giveaway items, in particular, bobbleheads. Of the 11 times the Dodgers had a bobblehead giveaway, they put large numbers in the stadium. Looking through the data, they had a large turnout on Tuesdays, primarily due to bobblehead giveaways.
To answer the original problem — “What night would be the best to run a marketing promotion to increase attendance?” — my recommendation is to start offering bobbleheads on Mondays to help bring more people out. Mondays were the lowest draw during the week, so giving the fans a reason to show up by giving them a freebie would be a good way to get attendance boosted on that day.
-
Ugh, as a Cubs fan, this hurts, but I’ll stay professional for the sake of this assignment. ↩