Data visualization with nflfastR

We will now apply our newly acquired plotting skills to make some visualizations working with data from nflfastR. Let’s load in our usual libraries, as well as nflfastR.

library(tidyverse)
library(nflfastR)
library(ggplot2)

Let’s load in the data for the 2024 season. This dataset contains play-by-play data for all games in the 2024 NFL season.

pbp_2024 <- load_pbp(2024)

We can see that there are more than 100 columns. Let’s tackle specific questions one at a time. First, lets look at quarterback completions by filtering for where play_type == pass and then selecting the following columns in a new table caled pbp_pass.

pbp_pass <- pbp_2024 %>% 
  filter(play_type == "pass") %>% 
  select(play_type, pass_location, yards_gained, air_yards, yards_after_catch, 
         passer_player_name, complete_pass, incomplete_pass, cpoe)

head(pbp_pass)

## ── nflverse play by play data ──────────────────────────────────────────────────

## ℹ Data updated: 2025-04-30 02:39:06 EDT

## # A tibble: 6 × 9
##   play_type pass_location yards_gained air_yards yards_after_catch
##   <chr>     <chr>                <dbl>     <dbl>             <dbl>
## 1 pass      left                    22        -3                25
## 2 pass      middle                   9         2                 7
## 3 pass      middle                   8         6                 2
## 4 pass      right                    0        12                NA
## 5 pass      <NA>                     0        NA                NA
## 6 pass      left                     5         5                 0
## # ℹ 4 more variables: passer_player_name <chr>, complete_pass <dbl>,
## #   incomplete_pass <dbl>, cpoe <dbl>

nflfastR unfortunately doesn’t include a position field to identify quarterbacks. To workaround, this, groupby player and filter for players with >150 pass attempts to remove players without sufficient play time. Hint: pass attempts will be the sum of entries in complete_pass and incomplete_pass
Let’s take a preliminary look at how location might affect pass outcomes. First, use !na with filter() to filter out pass attempts without a location label. Then, create a bar plot to visualize the frequency of attempts in each pass location. Feed pass_location as a fill argument to make each bar a different color and make sure to use labs() to set axis and plot titles. You should see that ggplot automatically adds a legend for you!
Let’s look at the distribution of yards gained based on pass location with some box plots. Use geom_boxplot() to create a box plot of yards_gained by pass_location. Continue using fill and labs to format the plots nicely.
Completion yards over expected is a metric that adjusts for the difficulty of a QB’s throw by calculating the yards completed above the expected amount based on throw timing, coverage, and location. Take a look at how cpoe may differ based on the pass location alone with geom_violin() Discuss your plots for 4 and 5 with other students. What conclusions can you draw about the pass location and how it may affect pass outcomes?
Let’s now turn to aggregated pass statistics per quarterback. Group by passer_player_name and create a table called pbp_grouped with columns passer_player_name, attempts, avg_yards_gained, and avg_cpoe using reframe().
We can first visualize how attempts and avg_yards_gained are related by using a scatterplot. Let’s feed a color and size argument to geom_point() to visually change the appearance of your plot and make the points take up an appropriate amount of space. Feel free to change the color argument to a color of your choice!

ggplot(pbp_grouped, aes(x = attempts, y = avg_yards_gained)) +
  geom_point(color = 'darkseagreen3', size = 3) +
  labs(
    title = "Quarterback Attempts vs Average Yards Gained",
    x = "Attempts",
    y = "Average Yards Gained"
  ) +
  theme_minimal()

Next, add an abline to the plot using geom_abline() to see how linear the data may be. Use the slope and intercept arguments to set the slope to 0.005 and the intercept to 5. Set the linetype to dashed, and the color to black. Note: This line was arbitrarily made with a slope and intercept that “looks best”, but in lecture 6, you will learn how to fit regression models to the data mathematically to make a line of best fit.
Now, let’s add a third variable. Set the color to equal avg_cpoe and use scale_color_viridis_c() to set the color scale.
Finally, visualize the average yards gained and completion yards over expected by quarterback. Choose to arrange either by avg_yards_gained or avg_cpoe and obtain the top 10 quarterbacks for this metric. Pivot long to create a table similar to the one shown in lecutre 4, where each QB has a row for their average yards gained and average cpoe, then generate a bar plot with geom_col(). Make sure to use dodge and to either use 45 degree axis labels or flip the coordinates with coord_flip() so that the QB names are visible.

What have you learned from these graphs? How do quarterbacks differ in their average yards gained vs average completion yards over expected? Is Lamar the GOAT??

Let’s do one more small case study on explosive plays.

First, start by filtering the data as below to get the appropriate columns.

pbp_explosive <- pbp_2024 %>%
  filter(play_type %in% c("pass", "run"),
         !is.na(yards_gained),
         !is.na(defteam),
         !is.na(down),
         !is.na(yardline_100)) %>%
  mutate(
    explosive = case_when(
      play_type == "run"  & yards_gained >= 10 ~ 1,
      play_type == "pass" & yards_gained >= 20 ~ 1,
      TRUE ~ 0
    )
  ) %>% 
  filter(explosive == 1) %>% 
  select(
    game_id,
    play_id,
    defteam,          # defensive team
    play_type,        # pass or run
    ydstogo,          # yards to go for first down
    yardline_100,     # field position (how far from opponent's end zone)
    yards_gained,     # actual yards gained on play
    explosive         # your new indicator (1 or 0)
  )

head(pbp_explosive)

## ── nflverse play by play data ──────────────────────────────────────────────────

## ℹ Data updated: 2025-04-30 02:39:06 EDT

## # A tibble: 6 × 8
##   game_id  play_id defteam play_type ydstogo yardline_100 yards_gained explosive
##   <chr>      <dbl> <chr>   <chr>       <dbl>        <dbl>        <dbl>     <dbl>
## 1 2024_01…      83 BUF     pass            7           67           22         1
## 2 2024_01…     622 BUF     pass            4           65           24         1
## 3 2024_01…     707 BUF     run            10           28           11         1
## 4 2024_01…     902 ARI     run            10           64           15         1
## 5 2024_01…     946 ARI     pass            5           44           23         1
## 6 2024_01…    1269 BUF     run             6           56           12         1

Let’s visualize the distribution of explosive plays by defensive team. Use geom_bar() to create a bar plot of the number of explosive plays by defteam. Use fill = defteam to color the bars by team and use labs() to set the axis and plot titles.
Let’s take a closer look at the worst defenses against explosive plays. Group by defensive team and play type, then use reframe to obtain the number of explosive plays in each category, and slice_max() to obtain the top 10 teams. Replot the bar graphs and facet by pass or run plays. Hint: to create a column counting the number of rows, use n() with reframe(). Additionally, use scales = 'free' with facet_wrap() to allow each facet to have its own y-axis scale.
Challenge: Finally, let’s visualize the relationship between explosive plays and yards to go with a heat map. First, run the following code below to obtain the needed columns for this plot.

pbp_explosive_rate <- pbp_2024 %>%
  filter(play_type %in% c("pass", "run"),
         !is.na(yards_gained),
         !is.na(defteam),
         !is.na(down),
         !is.na(yardline_100)) %>%
  mutate(
    explosive = case_when(
      play_type == "run"  & yards_gained >= 10 ~ 1,
      play_type == "pass" & yards_gained >= 20 ~ 1,
      TRUE ~ 0
    )
  ) %>% 
  select(
    game_id,
    play_id,
    defteam,          # defensive team
    play_type,        # pass or run
    ydstogo,          # yards to go for first down
    yardline_100,     # field position (how far from opponent's end zone)
    yards_gained,     # actual yards gained on play
    explosive         # your new indicator (1 or 0)
  )

Then, group by defteam and ydstogo and obtain the explosive play rate, instead of number of explosive plays. Hint: You will need to use n() and sum() for this task. Then, use geom_tile() to create a heat map with defteam on the x-axis, ydstogo on the y-axis, and fill based on the number of explosive plays. Set a color gradient to your liking.

Discuss with fellow students possible conclusions that can be drawn from the visualizations in this section. How are teams’ defenses stacking up against explosive plays? How much should certain bins in this visualization be weighed over others?

Problem Set 5

Data visualization with nflfastR