32 Game of Thrones Data Visualizations
And a bold prediction about who will end up on the Iron Throne
Now that HBO has announced the premiere date for the final season of Game of Thrones (April 14, 2019), I wanted to share some data visualizations I’ve been working on since I published “The Ultimate Game of Thrones Dataset” about a year and a half ago. It’s not imperative to read that article first, but it may help give a a sense for the impetus of this project.
A few notes before getting into things:
- All of these are data visualizations, not (static) infographics. There are loads of wonderfully nerdy Game of Thrones infographics out there, but visualizations will update when new data is added whereas the infographics won’t. As Season 8 proceeds to the series finale, the visualizations will stay current with the show when the datasets are updated.
- For simplicity, I’m posting .png screenshots to medium instead of embedding the visualizations — that’s to expedite load time on this page and make it more mobile-friendly. Feel free to check out any of the live visualizations via the links in the captions.
- Wherever possible, I’ve also included links in the captions to any source visualization or infographic inspiration for a visualization.
- The code for all of these is also shared on github and the visualizations can be viewed on github.io, too. The majority are made using d3.js v4.
- I’ve grouped the visualizations below by similarity: Narrative Chart, Screen Time, Cumulative Time, Character Co-Occurrence, Character Relationships, Actor Co-Occurrence, and Geography.
- The visualizations are built from several data files shared on github, including:
episodesis data about each scene in each episode of the show
characterscaptures data about character relationships, actor names, and character and actor imdb identifiers
locationslists an order of all regions and locations from north-to-south in Westeros then west-to-east in Essos
characters-groupslists the characters who are members of houses and groups
characters-gendercategorizes characters by gender
colorslists style information for various houses and groups (based on London Underground colors)
lands-of-ice-and-fireis GeoJSON data of Westeros, Essos, Sothoryos, Ulthos, and the various islands of the Game of Thrones world
Narrative Chart (through Season 7)
The narrative chart really started this project for Game of Thrones, and it continues to be the most layered of the visualizations below. It was inspired by xkcd and other visualizations (as described here), so just a bit about what it shows:
- Each line represents a character’s journey through the show. Hover over a line to isolate that line, see an image of the character if one is available, and track the regions through which that character travels.
- Every time a character is on screen, their line is wider; the paths between on-screen time infer either stasis in a place or a trajectory from one place to another.
- Each line is color coded by house/group (e.g. Starks are gray, Night’s Watch is black) and title: royalty (e.g. King, Khaleesi) has a solid gold outline and royal assistants (e.g. Hand) have a dotted bronze outline.
- Regions are arranged top to bottom geographically, from north to south in Westeros then west to east in Essos. Seasons alternate colors horizontally, and episode titles show at the top and bottom when hovering over a vertical subdivision of the background.
- Character death is marked by a pulsing red circle.
- The visualization can be explored using the 🔍 tab at the bottom: filter by gender, house, character name, title, and whether or not the character is/has been dead.
In addition to just making the narrative chart, it made sense to explore the datasets and see what other visualizations could be made; some are subsets of what’s shown on the chart above, while others are wholly unique. Some of the easiest things to visualize are representations of when characters are on screen, where those characters are, or when something happens to someone.
When are characters are on screen?
This visualization uses the
episodes dataset to show when each of the 555 characters is on screen. Each bar is color-coded by season, and hovering over a bar lists where the character is located during that scene, e.g.
The Crownlands — King's Landing. The characters are sorted top-to-bottom according to the total amount of time they’ve spent on screen, and the visualization is read left-to-right from the first season through the most recent season.
Where do scenes take place?
This is a similar visualization to the one above, but instead displays the various regions in which the show spends time. Again, colors represent seasons and the regions are sorted by the amount of on-screen time spent in each.
More specifically, where do scenes take place?
Same idea again, but using more-specific locations (e.g. cities, castles, roads) instead of just regions.
When are “special” scenes?
Throughout the show, there are certain events that are captured in the
episodes dataset: death, sex, flashbacks, greensight, and warging. This visualization shows when these happen, and hovering over a bar displays the season/episode and location when it does.
When do people die?
Death. So much death in Game of Thrones. This visualization essentially collects all of the death bubbles from the narrative chart above and shows the pattern of pulses of death throughout the show. Hovering over a bubble shows who died, who killed them, and how they died, e.g.
Oberyn Martell killed by Gregor Clegane via head crush or
Walder Frey killed by Arya Stark via throat slash. As above, the visualization proceeds from left-to-right from the first death (Waymar Royce) to the most recent one (Petyr Baelish).
When do weapons show up?
Weaponry is becoming increasingly important in the show, primarily because of the characters’ discovery that Dragonglass and Valyrian steel can kill White Walkers. This visualization shows when weapons are shown on screen, who has them, and where they’re located, e.g.
(S7E7) Arya Stark has Valyrian Steel Dagger at The North — Winterfell or
(S2E8) Samwell Tarly receives Dragonglass at North of the Wall — Fist of the First Men.
How many characters are on screen?
Periodically throughout the show, weddings and battles collect a higher-than-average number of characters on screen at the same time. This is a simple bar chart showing the number of characters on screen together at a given time. Hovering over a bar shows that character’s name and when else that character is on screen, and again is meant to be read left-to-right.
The same data can be centered to show the pulse of the show more clearly.
What about continuous on screen time?
As a show, there’s frequent bouncing between different characters in various locations, so when characters are on screen together for substantial periods of time, it’s usually quite special (and critical to the story). This visualization shows bubbles sized according to how long a character has been continuously on screen, and is read left-to-right. Circles are color-coded by house using
characters-groups and hovering over a bubble will show all times that character is on screen. Note the gathering of many characters for a substantial amount of time in Season 7, Episode 7 during the meeting at the Dragonpit outside of King’s Landing.
In addition to just representing data directly, it’s also fairly trivial to count time and places: how long are characters on screen (in total or by season)? how long do they spend in certain locations? how many locations are shown in an episode and how long is spent in each?
How long are characters on screen?
Ever wonder how long the supercut would be for a particular character? or who’s been on screen the longest over the course of the show? Here’s your answer (data from
episodes): through the end of Season 7, Tyrion Lannister has been on screen the most (09:09:16) followed closely by Jon Snow (09:07:32).
How long are characters in each season?
Want to know how long each character is on screen in each season? There’s a separate visualization for each season, but the leaders are:
- Season 1: Eddard Stark (02:22:43)
- Season 2: Tyrion Lannister (01:27:37)
- Season 3: Tyrion Lannister (01:09:06)
- Season 4: Tyrion Lannister (01:31:20)
- Season 5: Jon Snow (01:24:26)
- Season 6: Jon Snow (01:32:07)
- Season 7: Jon Snow (02:10:40)
How long are characters in each episode?
Curious whether characters gain or lose screen-time over the course of the show? This graph represents how long each character is in each episode, and episodes are color-coded by season. Each graph also includes a calculated trendline referenced below in the “Prediction” section.
How long is spent in each location?
Curious how long is spent in each region over the course of the show? This visualization accumulates on-screen time in various regions throughout all seasons.
It’s worth noting how the amounts of time spent in a location varies from season to season: relatively little time was spent at the Wall in Seasons 2 and 7, but quite a bit happened there in Seasons 1, 4, and 5.
How long is spent in each more-specific location?
Same idea as above, but with more specific places: cities, castles, roads, etc.
What percentage of a character’s time is in each season?
Some characters are important early on in the show and fade, while others have maintained importance throughout. This visualization shows how characters’ time on screen is dispersed throughout the show. Note how Tyrion has been pretty consistently on screen throughout the show, whereas someone like Davos Seaworth gets to be more important in later seasons.
How long are characters on screen relative to all time?
This visualization shows how much time a particular character is on screen relative to all of the time spent by all of the characters on screen. It’s not particularly useful, but it helps to distinguish between supporting characters (e.g. Lord Varys or Beric Dondarrion) and peripheral ones. Characters are grouped according to
(Really I just wanted to make a treemap.)
How many locations are in each episode?
This visualization counts how many different regions are shown per episode and how much screen time is spent in each.
Season 6, Episode 3 “Oathbreaker” shows the most different regions (9) followed by Season 6, Episode 10 “The Winds of Winter” (8), while Season 2, Episode 9 “Blackwater” takes place completely in The Crownlands, while Season 4, Episode 9 “The Watchers on the Wall” and Season 6, Episode 9 “Battle of the Bastards” each take place across only 2 regions.
What percentage of time do characters spend in various locations?
We can also see who the various traveling characters are by counting how many different regions various characters appear in and the relative amount of time they spend in each.
Daenerys Targaryen and Jorah Mormont each travel through 14 regions, while Tyrion Lannister and Daenerys’s dragons each appear in 12.
Character co-occurrence can be used to group characters according to the amount of time they spend together on-screen, but it can also be used to tease out interactions between groups that may be otherwise unexpected.
Which characters are on screen together?
One of the most fun things about Game of Thrones is when characters from very different narrative threads intersect and interact. The matrix visualization below shows which of the main characters have appeared on screen together, and hovering over a square shows how long those two characters have spent on screen together. Characters of the same house/group are also color-coded and the characters are sorted by those groups, demonstrating that not everyone within the group necessarily spends time together in the show (e.g. Yoren and the Night’s Watch).
… or as a force-directed network?
The same co-occurrence data can also be used to make a force-directed network that clusters characters based on how much time they spend together on screen. It’s quite cluttered given how many characters there are in Game of Thrones, but is still useful to see how certain characters group together (e.g. the younger versions of Ned Stark, Nan, Rodrik Cassel, Lyanna Stark, Wyllis, etc. in flashbacks to Winterfell).
… or as a chord diagram?
That same data can also be used to build a chord diagram where lines are drawn representing characters who spend time together on screen. This one is a bit of overload, but is fun to explore nevertheless: hover over a chord connecting two characters to see how long they’ve spent together on screen, or hover over a character along the circumference to see all of the other characters they’ve been on screen with.
Representing the ways in which characters are related — whether genealogically, socially, or otherwise — demonstrates the complexity of the Game of Thrones world.
Which characters are connected?
This is a relatively simple force-directed network visualization that links characters based on a few relationships captured in the
characters dataset: parent-child relationships are solid gray, spouse relationships (married or engaged) are dashed blue, and someone who was killed by someone else is shown by a solid black arrow pointing toward the person who was killed. You can quickly see who the real killers are because they have quite a few black arrows pointing away from them (Daenerys Targaryen, Sandor Clegane, Arya Stark, Jon Snow, etc.)
Which characters are connected?
This visualization asks the same question, but draws on a different source for inspiration: the infographic HBO put out to explain the R+L=J revelation of Jon Snow’s true parentage. Taking that data and extending it using the
characters dataset, the visualization below shows all of the parent-child, married/engaged, and killer/killed relationships in the show.
Which characters have sex?
Also drawing inspiration from elsewhere, this visualization links characters who’ve had sex with each other.
As an ensemble, the cast of Game of Thrones includes a mix of both newcomers and seasoned actors with quite lengthy careers (e.g. Charles Dance and Jim Broadbent). In the same way that we can look at the characters who have been on screen together, we can also look at which actors have worked with other cast members in different projects.
This data was aggregated using the
characters dataset along with theimdbapi.org (now defunct), but there are some other open imdb-like APIs out there now, too, which could be used to replicate this.
Which actors have been in movies with other Game of Thrones actors?
Not everything needs to be a visualization (I guess?), so I made a list of films with multiples Game of Thrones actors and posted it to medium here: “List of movies with Game of Thrones actors”.
… or as a matrix?
But it could also be visualized as a co-occurrence matrix. Hover over any square below to see how many films two actors have been in together.
The geography of the Lands of Ice and Fire is one of the unsung characters in the Game of Thrones world. The
episodes dataset includes location information for each scene and each episode’s (award-winning) opening sequence, so I was curious if I could visualize patterns in the places of Game of Thrones.
Is there a pattern in the opening locations?
I had long wondered whether it would be possible to use network analysis algorithms to find a single path through all of the opening sequence locations in order to construct something of a supercut of all of the places shown throughout the Game of Thrones intro.
To start with, the force-directed visualization below shows why that’s just not possible. Notice how there are too many “loose ends”: Eastwatch, Dorne, Astapor, and Yunkai are all dangling, so a single path can’t be drawn without backtracking at some point. It is also worth noting, however, that all episodes link Winterfell to The Wall; that is the only consistent transition throughout the show (even though the representation of Winterfell changes based on who is occupying it, etc.).
The opening sequence for Season 6, Episode 9 “Battle of the Bastards” is shown below; explore all episodes in the live visualization.
Could I show them in geographic space?
This visualization draws upon the same idea, but instead represents all of the places based on the geography of this map. As above, this is the opening sequence for Season 6, Episode 9 “Battle of the Bastards”, and you can explore all episodes in the live visualization.
What about on a globe?
Showing geography in 2D space was good, but 3D would be even better, right? Well, turns out that was non-trivial since the geographic data for the Lands of Ice and Fire didn’t really exist (in a format that worked). It’s probably a topic for another post, but the short version is that I built upon what others had already done (this book and this map), converted that map (.png) to a vector image (.svg) in Illustrator and cleaned it up quite a bit, imported those outlines through a few online converters (mygeodata.cloud and mapshaper.org), then hand rolled a bit of the resulting data to make the GeoJSON which I’ve shared on github as
lands-of-ice-and-fire. There was a fair bit of trial and error in making that file, and there’s a lot to say about how to do that smoothly (and freely), but I’ll leave that for another time (if you’re interested , please leave a comment and I’ll write that up).
One caveat about this geographic data: there’s no canonical globe version of these continents, so I made some assumptions when placing them on a sphere. Most importantly, I assumed that it’s cold at the poles and hot at the equator (like on Earth), so I’ve placed the Lands of Always Winter up around where the Arctic Circle would be, and I’ve placed Dorne and the Summer Isles in the Tropics. Beyond that, I didn’t create any new geographies in the world that may/may not be known, so there are some wonky cutoffs at the edges of the ‘known’ world that aren’t meant to represent the actual borders of the land masses.
This is what the resulting globe of Westeros, Essos, Sothyros, Ulthos, and loads of islands looks like with markers for the cities, castles, and ruins shown on the map. In the live visualization you can also zoom and rotate the world:
The opening sequences could look like this?
With a working geography, it then became possible to show the opening sequence locations and the path between them on the globe itself. As above, this is the opening sequence for Season 6, Episode 9 “Battle of the Bastards”, and you can explore all episodes in the live visualization.
… or this?
Taking the globe one step further, this visualization shows the path of each opening sequence while the shadows below represent all possible paths taken across all of the opening sequences. It’s a mix between the globe and the force-directed visualizations above.
Where have various characters travelled?
Lastly, what paths have characters taken throughout Game of Thrones? This visualization maps data from the
episodes dataset onto the globe and traces where different characters have gone. It isn’t perfect since it’s missing some locations (so far), and assumes travel ‘as the crow flies’, but it still shows some epic travel. Below is the path Jorah Mormont has taken, and you can explore all characters who’ve been to four or more locations in the live visualization.
… And a Bold Prediction
Wouldn’t it be nice to take a data-driven approach to predicting who will end up on the Iron Throne? A major caveat: there may not really be enough data (given the few people who’ve sat on the Iron Throne), so I don’t know how accurate a machine learning approach to answering this question could be. That said, it is worth asking who Game of Thrones is really about. And if the amount of time spent on screen is any indicator, Game of Thrones is (so far) the story of Tyrion Lannister and Jon Snow.
We can take that idea one step further (because, why not?) by plotting the slope and y-intercept of each trendline in the graph above (☝️☝️) that shows how long are characters appear in each episode. That gives this scatter plot where markers represent whether a character is alive (circle) or dead (x) and colors represent the last season a character appears in:
Hover over various points in the live visualization to see where different characters fall. And you may also start to notice some patterns; in a very unscientific way, I drew a few boundaries based on my impressions of the characters. All of the dead characters cluster together to the left; that’s expected since they’d have a negative slope by not showing up on screen often post-mortem. The minor characters are bundled in the center; they may have positive or negative slopes, but generally will have quite small y-intercepts. The contenders have presumably all been on screen quite a bit since the beginning (high y-intercept), and that won’t change drastically over time (slope between -2 and 2). And the dutiful servants have tended to become more important over the course of the show (positive slope) and have varying levels of initial importance (a range of y-intercept values).
Now, say what you will, but the two points that (continue to) stand out for me are Tyrion Lannister (top center just to the right of the y-axis) and Jon Snow (far right, halfway up the quadrant). I don’t know what it means that Jon Snow is in the Dutiful Servants cluster (even though he’s an outlier there), but my hunch is that he’ll once-again die (not to be brought back) during the final battles. So that leaves Tyrion Lannister who I think will either be the King/Queen-maker, or will be the King himself. Maybe we’ll learn more in Season 8 about why the dragons took so kindly to Tyrion?
So: King Tyrion of House Lannister, first of his name? A bold prediction, but one as good as any other. We’ll see how the data changes over the course of Season 8. Stay tuned!
Oh, and a Few More Things
The big reveal of Jon Snow’s true identity (birth rite and relations with Daenerys) piqued my interest in visualizing the genealogical trees of the various families of Westeros. However, the majority of the genealogical information available comes from the books since the show doesn’t go too many generations back in time. This data has already been visualized quite exhaustively, so I’m not going to recreate that. I’m not sure whether the underlying data is shared anywhere, but that’d be nice...
Scripts from Game of Thrones (and other shows) are available on some online forums. I may one day add dialogue to the
episodes dataset, but that’ll be a pretty massive undertaking...
There’s no reason the visualizations above wouldn’t also work for other shows. As a proof of concept, I’ve applied the same code to another custom dataset I’ve made for Netflix’s Stranger Things. It’s worth noting, however, that not all shows will have the same range of data available or suitable geography to do every visualization.
I was also made aware of Amazon’s X-ray data, and have used that data (without modification) to build several similar visualizations for The Man in the High Castle, but again, that dataset doesn’t include things like geography, so not every visualization is possible. Here, it’s worth noting that Amazon X-ray data isn’t perfectly clean/consistent, so it requires cleanup of things like character names and intro sequences. My hope for these visualizations is that they’ll work on any show where Amazon X-ray data available.
I hope you’ve enjoyed exploring these various Game of Thrones data visualizations. If you have ideas for others or improvements, please leave a comment. Thanks for reading!
Update (Apr. 2019): Now that the final season has begun, I’m posting a weekly data-driven recap of each episode:
- “Winterfell” (Season 8, Episode 1) Data Visualization Recap
- “A Knight of the Seven Kingdoms” (Season 8, Episode 2) Data Visualization Recap
- “The Long Night” (Season 8, Episode 3) Data Visualization Recap
- “The Last of the Starks” (Season 8, Episode 4) Data Visualization Recap
- “The Bells” (Season 8, Episode 5) Data Visualization Recap
- “The Iron Throne” (Season 8, Episode 6) Data Visualization Recap