Tidy Tuesday: 2025-01-04, Simpsons Episodes Dataset EDA

TidyTuesday’s 2025 Simpsons Dataset comes in 4 sections:

1. simpsons_characters.csv
2. simpsons_episodes.csv
3. simpsons_locations.csv
4. simpsons_script_lines.csv

I recently constructed a graph on section #2, simpsons_episodes.csv. In this article, I will show you the process I used to explore that set of data.

First, I took a peek at the data dictionary (Data Science Learning Community, 2025):

variable	class	description
id	double	Unique identifier for each episode record.
image_url	character	URL linking to the image associated with the episode record.
imdb_rating	double	IMDb rating for the episode.
imdb_votes	double	Number of votes received on IMDb for the episode.
number_in_season	double	Episode number within the season.
number_in_series	double	Episode number within the series.
original_air_date	date	Date the episode originally aired.
original_air_year	double	Year the episode originally aired.
production_code	character	Code used in production to identify the episode.
season	double	Season number of the episode.
title	character	Title of the episode.
us_viewers_in_millions	double	Number of viewers in the U.S. in millions.
video_url	character	URL linking to the video associated with the record.
views	double	Total number of views recorded for the episode video URL.

Next, I use R-language functions to examine my data a little closer. My favorites are dplyr’s glimpse() and the Skimr R package.

I then came up with a few research questions to help guide my exploration. Here are the ones I got for my previous visualization:

1. How does the distribution of imdb_rating vary with the other variables of the dataset?
2. What are the most relevant groupings for episode data?

I chose to focus on the imdb_rating columns because it seemed to have a well-spread distribution. Columns us_viewers_in_millions, views, and imdb_votes had also piqued my interest for similar reasons, but I decided to reserve those variables for a future visualization. A simple graphic on imdb_rating felt like a great way to begin my analysis.

The season variable felt like a natural grouping choice. In comparing an ordinal variable like season to a quantiative variable like imdb_rating, furthered my analysis by checking out a few styles of graphs.

Here’s a prototype I made for a violin plot. I think its main strength is that it contained a lot of information about the conditional distributions of imdb_rating given each season. The graph, however, felt very noisy to me. This led me to the stacked barchart strategy I posted the other day:

It’s not perfect, but my graph has its strengths. It does a pretty good job of showing the general imdb_rating differences between each season. I’m also proud of how clean it looks.

A lot of information is lost about the conditional distributions of each season, however. I do not know anything about outliers. I do not know anything about the ranges. No modality info either.

And immediately after I posted my previous graph, I remembered the existence of the ridgeline plot:

This graph gives way more information about the distributions. I’m going to post a polished version of this in the upcoming few days.

As another future direction, I’d like to look deeper into the number_in_series and number_in_season columns. They seem like pretty relevant info for a more complex visualization.

References

1. Data Science Learning Community (2025). Tidy Tuesday: A weekly social data project. https://tidytues.day.