Principles of Data Visualization and Introduction to ggplot2
I have provided you with data about the 5,000 fastest growing companies in the US, as compiled by Inc. magazine. lets read this in:
inc <- read.csv("https://raw.githubusercontent.com/charleyferrari/CUNY_DATA_608/master/module1/Data/inc5000_data.csv", header= TRUE)
And lets preview this data:
head(inc)
## Rank Name Growth_Rate Revenue
## 1 1 Fuhu 421.48 1.179e+08
## 2 2 FederalConference.com 248.31 4.960e+07
## 3 3 The HCI Group 245.45 2.550e+07
## 4 4 Bridger 233.08 1.900e+09
## 5 5 DataXu 213.37 8.700e+07
## 6 6 MileStone Community Builders 179.38 4.570e+07
## Industry Employees City State
## 1 Consumer Products & Services 104 El Segundo CA
## 2 Government Services 51 Dumfries VA
## 3 Health 132 Jacksonville FL
## 4 Energy 50 Addison TX
## 5 Advertising & Marketing 220 Boston MA
## 6 Real Estate 63 Austin TX
summary(inc)
## Rank Name Growth_Rate
## Min. : 1 (Add)ventures : 1 Min. : 0.340
## 1st Qu.:1252 @Properties : 1 1st Qu.: 0.770
## Median :2502 1-Stop Translation USA: 1 Median : 1.420
## Mean :2502 110 Consulting : 1 Mean : 4.612
## 3rd Qu.:3751 11thStreetCoffee.com : 1 3rd Qu.: 3.290
## Max. :5000 123 Exteriors : 1 Max. :421.480
## (Other) :4995
## Revenue Industry Employees
## Min. :2.000e+06 IT Services : 733 Min. : 1.0
## 1st Qu.:5.100e+06 Business Products & Services: 482 1st Qu.: 25.0
## Median :1.090e+07 Advertising & Marketing : 471 Median : 53.0
## Mean :4.822e+07 Health : 355 Mean : 232.7
## 3rd Qu.:2.860e+07 Software : 342 3rd Qu.: 132.0
## Max. :1.010e+10 Financial Services : 260 Max. :66803.0
## (Other) :2358 NA's :12
## City State
## New York : 160 CA : 701
## Chicago : 90 TX : 387
## Austin : 88 NY : 311
## Houston : 76 VA : 283
## San Francisco: 75 FL : 282
## Atlanta : 74 IL : 273
## (Other) :4438 (Other):2764
Think a bit on what these summaries mean. Use the space below to add some more relevant non-visual exploratory information you think helps you understand this data:
We’ll start with this:
library(psych)
describe(inc) # A nice little function that gives you a bit more than summary()
## vars n mean sd median trimmed
## Rank 1 5001 2501.64 1443.51 2.502e+03 2501.73
## Name* 2 5001 2501.00 1443.81 2.501e+03 2501.00
## Growth_Rate 3 5001 4.61 14.12 1.420e+00 2.14
## Revenue 4 5001 48222535.49 240542281.14 1.090e+07 17334966.26
## Industry* 5 5001 12.10 7.33 1.300e+01 12.05
## Employees 6 4989 232.72 1353.13 5.300e+01 81.78
## City* 7 5001 732.00 441.12 7.610e+02 731.74
## State* 8 5001 24.80 15.64 2.300e+01 24.44
## mad min max range skew kurtosis se
## Rank 1853.25 1.0e+00 5.0000e+03 4.9990e+03 0.00 -1.20 20.41
## Name* 1853.25 1.0e+00 5.0010e+03 5.0000e+03 0.00 -1.20 20.42
## Growth_Rate 1.22 3.4e-01 4.2148e+02 4.2114e+02 12.55 242.34 0.20
## Revenue 10674720.00 2.0e+06 1.0100e+10 1.0098e+10 22.17 722.66 3401441.44
## Industry* 8.90 1.0e+00 2.5000e+01 2.4000e+01 -0.10 -1.18 0.10
## Employees 53.37 1.0e+00 6.6803e+04 6.6802e+04 29.81 1268.67 19.16
## City* 604.90 1.0e+00 1.5190e+03 1.5180e+03 -0.04 -1.26 6.24
## State* 19.27 1.0e+00 5.2000e+01 5.1000e+01 0.12 -1.46 0.22
Lot’s of information here, not all of which is relevant (categorical data isn’t going to have a mean for example), but includes some basics like n and standard deviations, and some more niche stuff like skew and kurtosis.
Another pretty basic but helpful thing to do:
print(dim(inc))
## [1] 5001 8
print(sapply(inc, class))
## Rank Name Growth_Rate Revenue Industry Employees
## "integer" "factor" "numeric" "numeric" "factor" "integer"
## City State
## "factor" "factor"
You technically get this stuff by printing the head and looking at the data-viewer (top right), but it’s helpful to know what the data ‘looks like’
Create a graph that shows the distribution of companies in the data-set by State (ie how many are in each state). There are a lot of States, so consider which axis you should use. This visualization is ultimately going to be consumed on a ‘portrait’ oriented screen (ie taller than wide), which should further guide your layout choices.
library(ggplot2)
##
## Attaching package: 'ggplot2'
## The following objects are masked from 'package:psych':
##
## %+%, alpha
library(dplyr)
##
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
##
## filter, lag
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
State_counts <- inc %>% #Counts the number of states for each state
group_by(State) %>%
count()
ggplot(State_counts, aes(x = reorder(State, n), y = n)) + #tells it to look at the 'State' variable in the 'inc dataset, orders it from highest to lowest
geom_bar(stat = 'identity', width=0.8, fill = 'blue', color = 'black') + #Makes a barplot, also adds color so it's less dull
coord_flip() + #Inverts it so the states are listed on the y axis, which I like to do when there are lots of categories
ggtitle('Number of Fast Growing Companies by State') + #A title
xlab("State") + #Things got flipped, so this will show on the actual y axis
ylab("Number of Companies") +
scale_y_continuous(breaks = seq(0, 750, by = 50)) #Adds a few more tickmarks
A pretty basic bar-plot. Counting n on the x axis makes categorical labeling easier when there are lots of categories like there are here. An alternative to that would be to rotate the labels, but when they’re short abbreviations like this it doesn’t make as much sense.
Lets dig in on the state with the 3rd most companies in the data set. Imagine you work for the state and are interested in how many people are employed by companies in different industries. Create a plot that shows the average and/or median employment by industry for companies in this state (only use cases with full data, use R’s complete.cases()
function.) In addition to this, your graph should show how variable the ranges are, and you should deal with outliers.
#First we'll sepeate NY, the one with the third most companies in the dataset
NY <- inc[complete.cases(inc),] %>% #Removes incomplete observations
filter(State == 'NY') %>% #Only NY. Were I doing this with a non-static dataset, I would select this programatticaly
select(Industry, Employees) #We just need the industry and employees for each company
#Deals with outliers by using data within a certain multiple of the IQR from the third quartile
#Note: this is taking the whole dataset into account, and not doing it per industry (which would or would not be optimal)
Outlier_coef = 3 #Choose 3 for this because it looks good
NY <- NY[NY$Employees < quantile(NY$Employees, 0.75) + IQR(NY$Employees)*Outlier_coef,]
#Makes box and whisker plot
ggplot(NY, aes(x = Industry, y = Employees)) +
geom_boxplot()+
scale_x_discrete(guide = guide_axis(angle = 50)) + #Founds this line here: https://stackoverflow.com/questions/1330989/rotating-and-spacing-axis-labels-in-ggplot2
scale_y_continuous(breaks = seq(0, 400, by = 25))
Above is a basic box and whisker plot, with an outlier cutoff at 3*IQR (for the data-set as a whole). I like box and whisker for cases like this because it gives a decent and quick visualization of where the data is and how variable it is. The outlier cutoff here is a bit trickier: I went with about twice the usual to preserve some of the data. Including all outliers obscures the plot.
Now imagine you work for an investor and want to see which industries generate the most revenue per employee. Create a chart that makes this information clear. Once again, the distribution per industry should be shown.
(Assuming this is for all states again)
IRE <- inc[complete.cases(inc),] %>% #Gets rid of those pesky Na's
mutate(rev_per_emp = Revenue/Employees) %>% #Makes a new variable of revenue per employee
select(Industry, rev_per_emp) %>% #Subsets the variables we want
group_by(Industry) %>% #Groups by indsutry
summarize(Median = median(rev_per_emp), n = n(), Q1 = quantile(rev_per_emp, 0.25), Q3 = quantile(rev_per_emp, 0.75)) # gives us stats we want
ggplot(IRE, aes(x = reorder(Industry, -Median), y = Median)) + #Iniates plot. Orders it by decending order
geom_bar(stat = 'identity', col = "black", fill ="lightblue") + #Makes it a barplot
geom_errorbar(aes(ymin = Median - Q1, ymax = Median + Q3), width = 0.5) + #Techically not 'error'; this adds the IQR
scale_x_discrete(guide = guide_axis(angle = 50)) + #Angles X axis text so it's readable
scale_y_continuous(labels=scales::dollar_format(), breaks = seq(0, 1e7, by = 2e5)) + #Adds scale formatting. https://datavizpyr.com/dollar-format-for-axis-labels-with-ggplot2/
ylab("Revenue per Employee") + #Labels
xlab('Industry') +
ggtitle("Median Revenue per Employee by Industry")
Above is the median revenue per employee by industry. The blue bars are medians, and the thin blacks bars represent variation as the inter-quartile range.
I went with a bar-plot here because while they may not get quite as much information across as a box-plot, it’s a bit easier to read and still gives you a good sense of variability if you use the ‘error’ bars .
A note on statistics: if I were trying to present information to an investor, I would try to present ‘typical’ cases instead of average ones, especially considering how skewed this data-set is. Thus, I have used medians and IQRs to represent the data here instead of means and standard deviations. This is also much more resilient when it comes to outliers and variability, and as you can see, there’s quite a lot of both, and all outliers are taken into account this time.