Plotting relationships between variables

Lecture 12

Aidan Combs

Duke University
SOCIOL 333 - Summer Term 1 2023

2023-06-07

Logistics

Project component 2, descriptive statistics, due tomorrow 11:59pm
I recommend trying all parts today and coming tomorrow with questions
Tomorrow we will start on hypothesis testing

Today

Single variable plots exercise solutions
Factors
Multivariate plots

Exercise solutions: single variable plots

Factors

Remember our bar plot of helmet wearing habits from the exercise:

ggplot(data = yrbss, aes(x = helmet_12m)) + 
  geom_bar()

Factors

Categorical variables (those where the values are words/letters/numbers in quotation marks) default to being ordered alphabetically

glimpse(yrbss)

Rows: 13,583
Columns: 13
$ age                      <int> 14, 14, 15, 15, 15, 15, 15, 14, 15, 15, 15, 1…
$ gender                   <chr> "female", "female", "female", "female", "fema…
$ grade                    <chr> "9", "9", "9", "9", "9", "9", "9", "9", "9", …
$ hispanic                 <chr> "not", "not", "hispanic", "not", "not", "not"…
$ race                     <chr> "Black or African American", "Black or Africa…
$ height                   <dbl> NA, NA, 1.73, 1.60, 1.50, 1.57, 1.65, 1.88, 1…
$ weight                   <dbl> NA, NA, 84.37, 55.79, 46.72, 67.13, 131.54, 7…
$ helmet_12m               <chr> "never", "never", "never", "never", "did not …
$ text_while_driving_30d   <chr> "0", NA, "30", "0", "did not drive", "did not…
$ physically_active_7d     <int> 4, 2, 7, 0, 2, 1, 4, 4, 5, 0, 0, 0, 4, 7, 7, …
$ hours_tv_per_school_day  <chr> "5+", "5+", "5+", "2", "3", "5+", "5+", "5+",…
$ strength_training_7d     <int> 0, 0, 0, 0, 1, 0, 2, 0, 3, 0, 3, 0, 0, 7, 7, …
$ school_night_hours_sleep <chr> "8", "6", "<5", "6", "9", "8", "9", "6", "<5"…

Factors

But we can change that!
Factors are a specific type of R object. They are categorical with response options that are represented by words/letters/numbers in quotation marks, but they allow you to assign those options any order you’d like.
Useful for ordinal variables
To change the order of the bars, we change the order of the response options in the data set before we plot.

Factors

We can use mutate() with a new function, factor(), for this.

table(yrbss$helmet_12m, useNA = "always")


      always did not ride most of time        never       rarely    sometimes 
         399         4549          293         6977          713          341 
        <NA> 
         311

yrbss_reordered <- mutate(yrbss,
                          # we are turning our helmet variable into a factor
                          # its name will remain the same (the new variable 
                          # we create will overwrite the old one)
                          helmet_12m = factor(
                            helmet_12m, 
                            # and we want its response options to be in this order
                            # least to greatest frequency
                            levels = c("did not ride", "never", "rarely", 
                                       "sometimes", "most of time", "always")))

table(yrbss_reordered$helmet_12m)


did not ride        never       rarely    sometimes most of time       always 
        4549         6977          713          341          293          399

Factors

Our new plot! I’ve added better labels as well.
What is easier to see in this plot than the old one?
Does the pattern surprise you?

ggplot(data = yrbss_reordered, aes(x = helmet_12m)) + 
  geom_bar() + 
  labs(x = "How frequently did you wear a helmet while riding your bike?",
       y = "Number of students",
       title = "Distribution of reported helmet wearing in the last 12 months")

Multivariate plots

Plots for relationships between variables

Plots are an excellent tool for looking at the relationships between variables.
The same structure applies for plotting two/three/four/etc variables as one variable.
What’s different? Some new function options, some new geoms.

Types of plots for different combinations of variables: overview

One numeric variable
- Histogram, geom_histogram()
One categorical variable
- Bar chart, geom_bar()

Two numeric variables
- Scatterplots, geom_point()
One numeric and one categorical variable
- Boxplots, geom_boxplot()
Two categorical variables
- Bar chart with color or facets

This is not the be-all end-all list; there are more options! But these are common and useful places to start.

What about three or more variables?

Start with your main explanatory variable and main response variable and make a two variable plot

Adding an additional categorical variable
- Can you add color (one per category)?
- Can you break a big plot into several small plots (facets)?

Adding an additional numeric variable
- Can you add color (along a gradient)?
- Can you add size?

Adding variables in ggplot

aes() (aesthetics) is where we tell R what variables we want represented by what plot elements
For example:
- ggplot(acs12, aes(x = income))
- ggplot(acs12, aes(x = income, y = married))
- ggplot(acs12, aes(x = income, y = married, color = gender))
- ggplot(acs12, aes(x = time_to_work, y = lang, size = income))
- ggplot(acs12, aes(x = time_to_work, y = lang, color = income))

Bar charts for two categorical variables

RQ: How do reported rates of texting while driving vary by gender? (in yrbss)
We can use color!
- color (for lines and points) and fill (for filling in areas like bars) are the arguments we use inside aes() to add color.

# if I don't know what variables to look at, I can use glimpse()
glimpse(yrbss)

Rows: 13,583
Columns: 13
$ age                      <int> 14, 14, 15, 15, 15, 15, 15, 14, 15, 15, 15, 1…
$ gender                   <chr> "female", "female", "female", "female", "fema…
$ grade                    <chr> "9", "9", "9", "9", "9", "9", "9", "9", "9", …
$ hispanic                 <chr> "not", "not", "hispanic", "not", "not", "not"…
$ race                     <chr> "Black or African American", "Black or Africa…
$ height                   <dbl> NA, NA, 1.73, 1.60, 1.50, 1.57, 1.65, 1.88, 1…
$ weight                   <dbl> NA, NA, 84.37, 55.79, 46.72, 67.13, 131.54, 7…
$ helmet_12m               <chr> "never", "never", "never", "never", "did not …
$ text_while_driving_30d   <chr> "0", NA, "30", "0", "did not drive", "did not…
$ physically_active_7d     <int> 4, 2, 7, 0, 2, 1, 4, 4, 5, 0, 0, 0, 4, 7, 7, …
$ hours_tv_per_school_day  <chr> "5+", "5+", "5+", "2", "3", "5+", "5+", "5+",…
$ strength_training_7d     <int> 0, 0, 0, 0, 1, 0, 2, 0, 3, 0, 3, 0, 0, 7, 7, …
$ school_night_hours_sleep <chr> "8", "6", "<5", "6", "9", "8", "9", "6", "<5"…

# this will give me the response options
table(yrbss$text_while_driving_30d, 
      yrbss$gender, 
      useNA = "always")

               
                female male <NA>
  0               2254 2534    4
  1-2              438  487    0
  10-19            154  219    0
  20-29            124  174    0
  3-5              208  282    3
  30               358  467    2
  6-9              140  171    0
  did not drive   2485 2159    2
  <NA>             460  457    1

Bar charts for two categorical variables

fill = gender is the argument we want for bar color (an area rather than a point or line)
This is an okay start, but it’s not super obvious yet what the story is

yrbss |> 
  filter(!is.na(gender) & !is.na(text_while_driving_30d)) |> # remove missing data
  mutate(text_while_driving_30d = factor(text_while_driving_30d, # change the order of the bars
                                         levels = c("did not drive", "0", "1-2", "3-5", 
                                                    "6-9", "10-19", "20-29", "30"))) |> 
  # we add the fill argument here with the variable we want the fill color to represent.
  ggplot(aes(x = text_while_driving_30d, fill = gender)) + 
  geom_bar()

Bar charts for two categorical variables

We can set the bars to be next to one another rather than stacked with the position = "dodge" argument to geom_bar()

yrbss |> 
  filter(!is.na(gender) & !is.na(text_while_driving_30d)) |>
  mutate(text_while_driving_30d = factor(text_while_driving_30d,
                                         levels = c("did not drive", "0", "1-2", "3-5", 
                                                    "6-9", "10-19", "20-29", "30"))) |> 
  ggplot(aes(x = text_while_driving_30d, fill = gender)) + 
  geom_bar(position = "dodge") + 
  labs(x = "Number of days last month respondent texted while driving",
       y = "",
       # I can change the label on other aes features just like on axes--this goes above the legend
       fill = "Sex")

Bar charts for two categorical variables

Here the story is clearer!

New plots for two variables: scatter plots and boxplots

Scatter plots show the relationship between two numeric variables
Explanatory (independent) variable on the x axis, response (dependent) variable on the y axis

Scatter plot

ggplot(acs12, aes(x = time_to_work, y = hrs_work)) + # we specify x and y axis variables here
  geom_point() + # and this tells R we want a scatter plot
  labs(x = "Commute time (minutes)",
       y = "Hours worked per week")

Scatter plot with color

What if we wanted to add a third variable? How about gender?
One option is to use color

Scatter plot with color

# This code is nearly identical to the last plot--we just add one argument to aes!
ggplot(acs12, 
       aes(x = time_to_work, 
           y = hrs_work,
           color = gender)) + 
  geom_point() +
  labs(x = "Commute time (minutes)",
       y = "Hours worked per week")

Box plots

Box plots are good for showing the median, quartiles, outliers, and range of a variable.
You can plot a single numeric variable using a boxplot, but they’re most useful for comparing the distribution of a numeric variable across different groups represented by a categorical variable.

Box plots and distributions: exercise

Box plots

Example: How does the number of hours worked per week vary by education level?

ggplot(acs12, aes(x = hrs_work, y = edu)) + 
  geom_boxplot() # this gives us our boxplot

Three variables: Adding color to our boxplot

RQ: Among people who report being employed, how does the relationship between education level and number of hours worked per week vary by ability status?

acs12 |> # we start with the acs
  filter(employment == "employed") |> # first we filter out people who are not employed
  ggplot(aes(x = hrs_work, y = edu, color = disability)) + # this part is the same; we just add the color argument!
  geom_boxplot()

My expectations for your plots

Make plots that are interpretable
- Choose plot types that are appropriate to your variables
- Use useful labels

Do they have to be pretty?
- No. Ugly but interpretable is fine.
- BUT, R is really good at making beautiful graphics–and if you’d like to learn more about how to do that, I’m happy to help!

Additional resources for plotting

Check out the R Graph Gallery
- This walks through different kinds of plots and their strengths and weaknesses. It includes lots of example code.
- There are lots of resources here for making plots both more interpretable and more beautiful!
If this interests you, I recommend:
- Soc 232: Visualizing Social Data, taught by Dr. Kieran Healy
  - or, Dr. Healy’s book, Data Visualization: A Practical Introduction
- Stat 313: Advanced Data Visualization, taught by Dr. Mine Çetinkaya-Rundel
- Soc 223: Data Analytics and Visualization for Business, taught by Dr. Stephen Vaisey

Steps to make a plot for your project

Step 1: Identify the variables you want to plot and figure out if they’re numeric or categorical
- Watch out for variables that are theoretically numeric but treated in R as categorical (like the texting while driving variable above)
Step 2: Choose a plot type that fits your variables and question
- Sketch it out! Figure out what it should look like in your head before you jump into R.
- What should each axis represent? Do you need color or something else in order to include all your variables?
Step 3: Look for resources. Find the relevant code in the slides and read the R Graph Gallery’s information about your plot type.
Step 4: Make the super simple version
Step 5: Refine it to make it more interpretable
- My expectation: add labels and drop NA values if necessary

For tomorrow

No exercise on multivariate plots
Try out part 4 of your project instead! Come tomorrow with questions.