Angela Zoss 2/1/2018
knitr::opts_chunk$set(error = TRUE)# Load required libraries
library(tidyverse)## ── Attaching packages ──────────────────────────────────────────────────────────────────── tidyverse 1.2.1 ──
## ✔ ggplot2 2.2.1 ✔ purrr 0.2.4
## ✔ tibble 1.4.2 ✔ dplyr 0.7.4
## ✔ tidyr 0.7.2 ✔ stringr 1.2.0
## ✔ readr 1.1.1 ✔ forcats 0.2.0
## ── Conflicts ─────────────────────────────────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
# data comes from https://www.kaggle.com/uciml/adult-census-income
# adding a few settings to correct the data types of a couple of the columns
adult <- read_csv("data/adult.csv", na="?")## Parsed with column specification:
## cols(
## age = col_integer(),
## workclass = col_character(),
## fnlwgt = col_integer(),
## education = col_character(),
## education.num = col_integer(),
## marital.status = col_character(),
## occupation = col_character(),
## relationship = col_character(),
## race = col_character(),
## sex = col_character(),
## capital.gain = col_integer(),
## capital.loss = col_integer(),
## hours.per.week = col_integer(),
## native.country = col_character(),
## income = col_character()
## )
# Bar chart, automatically counting number of observations
ggplot(adult) +
geom_bar(aes(sex))
# Bar chart, using another column for the length
ggplot(adult) +
geom_col(aes(x=sex, y=capital.loss))
# or you can use geom_bar and just change the default statistical function ("stat"),
# which is normally "count"
ggplot(adult) +
geom_bar(aes(sex, capital.loss), stat="identity")
# What does it do when you have multiple records for each category?
adult %>% group_by(sex) %>% summarise(total = sum(capital.loss))## # A tibble: 2 x 2
## sex total
## <chr> <int>
## 1 Female 659052
## 2 Male 2183648
# Bar chart, using another column for length and specifying the summary function
ggplot(adult) +
geom_bar(aes(sex, capital.loss), stat="summary", fun.y="mean")
# In this case, geom_col doesn't work
ggplot(adult) +
geom_col(aes(sex, capital.loss), stat="summary", fun.y="mean")## Warning: Ignoring unknown parameters: stat, fun.y
# You can also summarize numerical variables with a bar chart, but consider using histogram or density instead
ggplot(adult) +
geom_bar(aes(age))
ggplot(adult) +
geom_histogram(aes(age))## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
ggplot(adult) +
geom_histogram(aes(age), binwidth = 1)
ggplot(adult) +
geom_density(aes(age))
# Adding another category - do you want to stack or dodge or fill?
# By default, a new category fills the bar (position="stack")
ggplot(adult) +
geom_bar(aes(sex, fill=race))
# Change position to "dodge" for side-by-side bars
ggplot(adult) +
geom_bar(aes(sex, fill=race), position="dodge")
# Change position to "fill" for bars scaled up to 100%
ggplot(adult) +
geom_bar(aes(sex, fill=race), position="fill")
# Pie charts are... weird. In ggplot2, you make a pie chart by taking a single stacked bar and changing the coordinate system to coord_polar.
ggplot(adult) +
geom_bar(aes(x="", fill=sex))
ggplot(adult) +
geom_bar(aes(x="", fill=sex)) +
coord_polar("y")
# Can get rid of hole in the middle by specifying width = 1
ggplot(adult) +
geom_bar(aes(x="", fill=sex), width=1) +
coord_polar("y")
# Scatter plots can show relationships between numerical variables, but be careful of overplotting (dots stacked on top of each other)
ggplot(adult) +
geom_point(aes(age, capital.loss))
ggplot(adult) +
geom_bin2d(aes(age, capital.loss))
# Note: if your counts are heavily skewed, you can apply a transformation on the color scale
ggplot(adult) +
geom_bin2d(aes(age, capital.loss)) +
scale_fill_continuous(trans="log10")
# Line charts don't include any inherent summary, so individual data points get mapped and connected with a line
ggplot(adult) +
geom_line(aes(x=age, y=capital.loss))
ggplot(adult) +
geom_line(aes(x=age, y=capital.loss)) +
geom_point(aes(x=age, y=capital.loss))
# You can add your own function to summarize all of the y values at the same x value
ggplot(adult) +
geom_line(aes(x=age, y=capital.loss), stat="summary", fun.y=mean)
# Alternately, you can use geom_smooth to calculate a variety of summary lines
ggplot(adult) +
geom_smooth(aes(x=age, y=capital.loss))## `geom_smooth()` using method = 'gam'
ggplot(adult) +
geom_smooth(aes(x=age, y=capital.loss), method = "lm")
# Adding main title and axis labels
ggplot(adult) +
geom_bar(aes(sex)) +
labs(title="This sample has about twice as many men as women.",
x="Sex",
y="Number of Respondents")
# Changing the legend title is a little harder;
# you have to modify the "scale" properties for the non-axis variable
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
labs(title="This sample has about twice as many men as women.",
x="Sex",
y="Number of Respondents") +
scale_fill_discrete(name="Race/Ethnicity")
# The structure is: "scale_" plus whatever aesthetic property you're modifying (e.g., x, fill, size)
# plus either a.) the kind of variable it is (e.g., continuous, discrete) or
# b.) a special function that will be applied (e.g., log10, gradient)
# For example, you can't modify the name of the fill aesthetic if you treat it like a continuous variable
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
labs(title="This sample has about twice as many men as women.",
x="Sex",
y="Number of Respondents") +
scale_fill_continuous(name="Race/Ethnicity")## Error: Discrete value supplied to continuous scale
# ggplot2 does a pretty good job guessing what the axes should look like, but you can modify
# individual properties manually
# Changing axis properties requires adding a "scale" layer for one or both axes,
# just like modifying the legend properties
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90))
# Gridlines often show up for both major and minor breaks. To turn off gridlines for minor
# breaks, an easy way is to set the minor breaks to NULL.
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90), minor_breaks = NULL)
# Can also change how the numbers on the axis are spaces out, without doing any mathematical
# transformations to the data
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90), minor_breaks = NULL) +
scale_y_log10()## Warning: Transformation introduced infinite values in continuous y-axis
# Note: log scale spreads out small values, but you can't plot the value "0" on a log scale,
# so we can filter those out first, then do some formatting on the breaks and labels
ggplot(adult %>% filter(capital.gain > 0)) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90), minor_breaks = NULL) +
scale_y_log10(breaks=c(10,100,1000,10000,100000),
labels=function(x){format(x, scientific = FALSE, big.mark=",")})
# Coordinate layers help control the output of the chart
# coord_fixed helps normalize the units across the two axes;
# ratio = 1 means that each unit on the x axis is the same length as each unit on the y axis
# the syntax for ratio is y/x
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
coord_fixed(ratio = 1)
# a 1:1 ratio for this chart is terrible; the y-axis goes from 0 to 100,000, while the
# x-axis goes from 17 to 90. To have the units look similar, each unit on the x-axis
# should be about 1,000 of the y-axis units.
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
coord_fixed(ratio = 1/1000)
# Now the grid on the chart is approximately square# Another useful coord layer is coord_flip. Some charts require certain variables in
# certain slots. For example, geom_bar requires the categorical variable in the x
# position and the numerical variable in the y position
ggplot(adult) +
geom_bar(aes(x=sex))
ggplot(adult) +
geom_bar(aes(y=sex))## Error: stat_count() must not be used with a y aesthetic.
# To get the categories on the y-axis, you start with the category on the x-axis and then
# add the coord_flip
ggplot(adult) +
geom_bar(aes(x=sex)) +
coord_flip()
# Scales can change the spacing of numbers on the axis - essentially, changing the grid
# against which the numbers are plotted
ggplot(adult %>% filter(capital.gain > 0)) +
geom_point(aes(age, capital.gain)) +
scale_y_log10(breaks=c(10,100,1000,10000,100000),
labels=function(x){format(x, scientific = FALSE, big.mark=",")})
# The coord_trans can also change the grid, but it still uses a cartesian approach to the
# major breaks
ggplot(adult %>% filter(capital.gain > 0)) +
geom_point(aes(age, capital.gain)) +
coord_trans(y="log10")
# As we've already seen, the labels on a numerical axis can be changed by manually setting
# the breaks in the axis and/or formatting the labels
ggplot(adult) +
geom_point(aes(age, capital.gain))
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90))
# regardless of the type of axis, you can set the labels to whatever you want
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
scale_x_continuous(breaks=c(20,30,40,50,60,70,80,90),
labels=c("A","B","C","D","E","F","G","H"))
# the same works for a categorical variable, either on an axis or in a legend
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_x_discrete(labels=c("A","B")) +
scale_fill_discrete(labels=c("A","B","C","D","E"))
# The one thing you can't do with scales is change the order of the categories. For that,
# you have to turn the variable into a factor and specify the order in the "levels" setting
# of the factor() function
adult$sex <- factor(adult$sex, levels = c("Male","Female"))
adult$race <- factor(adult$race, levels = c("Other","Amer-Indian-Eskimo","Asian-Pac-Islander","Black","White"))
ggplot(adult) +
geom_bar(aes(sex, fill=race))
# You can label charts with variables from the dataset or other calculated variables within the chart;
# just like other chart layers, geom_text layers need x and y positions, and they also require "label"
ggplot(adult) +
geom_bar(aes(sex)) +
geom_text(aes(sex, label=sex), stat="count")
# Note: geom_bar has stat="count" embedded within it. If we don't add that to the geom_text
# layer, the text layer will try to process each data point individually and will ask
# for a y value
# Instead of using the exact y position calculated by the stat_count function, you can
# nudge the label up or down. Remember, this "nudge" value needs to be in the same units
# as the axis.
ggplot(adult) +
geom_bar(aes(sex)) +
geom_text(aes(sex, label=..count..), stat="count", nudge_y = 1000)
# geom_label is like geom_text, but it formats the label differently
ggplot(adult) +
geom_bar(aes(sex)) +
geom_label(aes(sex, label=..count..), stat="count", nudge_y = 1000)
# Themes control the overall look and feel of the graph. Several themes are built in to ggplot2.
ggplot(adult) +
geom_bar(aes(sex)) +
theme_gray() +
ggtitle("Gray")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_classic() +
ggtitle("Classic")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_bw() +
ggtitle("Black and white")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_dark() +
ggtitle("Dark")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_light() +
ggtitle("Light")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_minimal() +
ggtitle("Minimal")
ggplot(adult) +
geom_bar(aes(sex)) +
theme_void() +
ggtitle("Void")
# If the preset themes are insufficient, individual properties can be redefined with theme()
ggplot(adult) +
geom_bar(aes(sex)) +
theme_bw() +
theme(panel.background = element_rect(fill="pink"))
# you can also use themes to remove grid lines without removing tick marks
ggplot(adult) +
geom_bar(aes(sex)) +
theme(panel.grid.major.y = element_blank())
# Earlier, we used geom_text for data labels. We can also use it for general notes on the graph.
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
geom_text(x=25,y=75000,label="A note goes here.")
# This is suboptimal, though, because a text object that doesn't relate to the data will process
# slowly. In this case, the chart is actually drawing a separate text object for every data point
# in the dataframe.
# Instead, use annotate for unrelated text objects
# Note: you have to specify a "geom" for each annotation - text, rect, segment, pointrange
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
annotate("text",x=25,y=75000,label="A note goes here.")
ggplot(adult) +
geom_point(aes(age, capital.gain)) +
annotate("rect",xmin=12, xmax=95, ymin=25000, ymax=50000, alpha=.2)
# Most charts use x and y axes to display variables, but you can usually also add another variable to
# the chart by mapping it to color.
# Note - some geoms use the "color" aesthetic (e.g., geom_point), while others use "fill" (e.g., geom_bar)
ggplot(adult) +
geom_point(aes(age, capital.gain, fill=race))
ggplot(adult) +
geom_point(aes(age, capital.gain, color=race))
ggplot(adult) +
geom_bar(aes(sex, fill=race))
ggplot(adult) +
geom_bar(aes(sex, color=race))
# You can change the colors of the chart elements using pre-defined palettes or by manually
# selecting colors
# You can apply pre-defined palettes to a fill or color aesthetic using a scale.
# A useful pre-defined palette is the grey palette, which uses shades of grey to keep a chart
# black and white
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_grey()
# The grey palette picks shades of grey between the start and end value, which by default are 0.2 and 0.8.
# The values can be reversed or made larger or smaller.
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_grey(start=0.8, end=0.2)
# Other default palettes include palettes from colorbrewer2.org
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_brewer(palette="Dark2")
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_brewer(palette="Dark2", direction=-1)
# Specifying colors manually is as simple as giving a vector of colors to a "manual" scale
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_manual(values=c("cadetblue3","lightpink2","lightskyblue3","lightsalmon2","olivedrab3"))
# R knows names for some colors, but others will need to be specified with codes
ggplot(adult) +
geom_bar(aes(sex, fill=race)) +
scale_fill_manual(values=c("#ffffd4","#fed98e","#fe9929","#d95f0e","#993404"))
# For continuous numbers, you can also use scale_color_gradient or scale_fill_gradient
ggplot(adult) +
geom_point(aes(age, capital.gain, color=age))
ggplot(adult) +
geom_point(aes(age, capital.gain, color=age)) +
scale_color_gradient(low="gray80",high="gray20")
# Color Brewer palettes can be used with continuous numbers, but you use "distiller"
# instead of "brewer"
ggplot(adult) +
geom_point(aes(age, capital.gain, color=age)) +
scale_color_distiller(palette="Greens")
# Facets are some of the most useful layers when you have a large amount of data. Facets create
# "small multiples", or a series of charts that each has the same specification but that visualizes
# a different subset of the data. You can think of it as splitting the data into chunks and then
# visualizing each chunk the same way.
# facet_wrap creates a series of charts split by one category and then wraps the charts into
# multiple rows as needed
ggplot(adult) +
geom_point(aes(age, hours.per.week)) +
geom_smooth(aes(age, hours.per.week)) +
facet_wrap(~education)## `geom_smooth()` using method = 'gam'
# facet_grid allows you to split the data by one category in the y direction and another in the x
# syntax: facet_grid(y~x)
# Note: you can use a "." to ignore one direction
ggplot(adult) +
geom_point(aes(age, hours.per.week)) +
geom_smooth(aes(age, hours.per.week)) +
facet_grid(.~race)## `geom_smooth()` using method = 'gam'
ggplot(adult) +
geom_point(aes(age, hours.per.week)) +
geom_smooth(aes(age, hours.per.week)) +
facet_grid(sex~race)## `geom_smooth()` using method = 'gam'
