Morbidity, Mortality & Murder in Westeros

Morbidity, Mortality & Murder in WesterosSurvival analysis and a data-based look at death in A Game of ThronesChris BrownlieBlockedUnblockFollowFollowingMay 24So the long-running hit TV show A Game of Thrones has finally come to an end after 8 years, 73 episodes and over 200 deaths.

To ensure this is a spoiler free article I’ll end the discussion of Season 8 there…This is the third article in an analytical series about the Game of Thrones TV show.

The first two posts examined transcripts from the first 7 seasons, looking at how characters’ words represent their storylines and how the emotion behind their words can give insight to their motives.

This article will differ in that it will look at a different topic all together, but one that is central the success and storylines of Game of Thrones – death.

To ensure this article is a continuation of my first two and to avoid spoilers so soon after the end of season 8, I will continue to only look at seasons 1–7.

This article mostly uses data from Jeffrey Lancaster’s excellent Game of Thrones dataset.

Fire and BloodTo begin with we’ll have a look at who has the most kills throughout the series and how they like to do their killing.

Note that ‘kills’ here refers to kills of named characters.

This usually means characters who speak at least one line.

List of the top 14 biggest killers in Season 1–7 and their most common method of killing.

Unsurprisingly Dany tops the list and unsurprisingly she favours fire.

Her second most common method of killing was death by dragon.

Dany and Cersei are the biggest mass murderers with 8 and 6 of their kills respectively coming in a single scene each.

Dany burning the Khals in Vaes Dothrak and Cersei attacking the Sept of Baelor with wildfire.

Impactful deathsNext we can consider how the distribution of deaths is spread across the first seven seasons.

Originally I planned to simply show the number of deaths in each episode but that didn’t feel particularly meaningful.

Two different character deaths can have wildly different effects on the viewer, particularly if one is a cameo appearance and the other is a long-running fan favourite.

So to analyse instead how the distribution of ‘death impact’ is spread across the seasons, I looked at the word count of killed characters in each episode.

This shows us not only the number of deaths but also a proxy of the impact those deaths would have had on the viewer.

The results can be seen below, the blue lines help differentiate between the different seasons:The word count of characters killed in each episode.

Note: this does not include Jon’s death.

We can see straight away that two of the most talked about episodes, “The Winds of Winter” (Cersei destroying the Sept of Baelor) and “The Rains of Castamere” (The Red Wedding), are also the two where the most words were killed off.

Unsurprisingly, deaths of main characters are more impactful and therefore generate more discussion.

We can also see how each season sees a spike in the penultimate or final episode.

This is a well known characteristic of A Game of Thrones — that they save the most shocking and impactful episodes for one of these two slots.

For information:Season 4 Episode 10 saw the deaths of Jojen Reed, Shae & Tywin Lannister, all of whom had a relatively high word count at the time of their death.

Season 5 Episode 10 sees 8 deaths, most notably of Stannis Baratheon but also the death of Jaqen H’ghar at the hands of Jaqen H’ghar (a potentially contentious inclusion but at the time of the episode it appeared he had killed himself).

The spike at the end of Season 7 is solely down to the demise of Littlefinger.

How to survive in A Game of ThronesThe rest of the article will use survival analysis to look at which characteristics are correlated with a longer survival time in the show.

I will be using Kaplan-Meier estimators – which involves splitting characters into groups and measuring the difference in survival times between them – and a Cox proportional hazard model – which helps to determine which attributes of a character are associated with longer survival in the show.

If you are a non-technical person or don’t quite understand these descriptions, don’t worry – the graphs and explanations below may help to explain the concept.

The survival outcome to be measured is the number of minutes after first appearance, until either death or the end of the show (which would be an example of Type I right-censoring).

This can be interpreted as follows: If all episodes from the first through to the end of Season 7 were played back to back (excluding credits), how many minutes are there between a character’s first appearance and their death (or the end of the series).

This gives us a measure of survival that we can use in our analysis.

I will be looking at data for 187 characters, of which 148 died before the end of the series.

Analysis was performed in R using the survival and survminer packages.

Below shows the Kaplan-Meier plots for 3 character attributes that I found to have a significant association with survival.

Westerosi is Best(erosi)The plot below shows the survival probability of Westerosi characters (those who have a culture that originates in Westeros) vs non-Westerosi characters (e.

g.

Dothraki, Meerenese, Braavosi etc.

).

Kaplan-Meier plot for GoT characters split by Westerosi culture.

The number in brackets indicates the number of characters falling into each group.

The p-value of the log rank test for these two groups is p=0.

0012, indicating a high level of confidence that the two groups have different survival rates.

Here we can see that Non-Westerosi characters have a significantly worse survival rate than Westerosi characters.

This could be due to a number of reasons: one being that Westeros tends to be more developed than Essos.

Dothraki like to kill each other regularly and the residents of Slavers Bay also struggle with death constantly which could lead to Non-Westerosi living more dangerous lives.

Another possibility is simply that Westeros is home to more prominent characters, who survive longer because they are more central to the show’s storylines.

Hazardous HousesNext up we’ll compare those who owe their allegiance to a noble House compared to those who don’t.

The former group includes anyone who is born into a noble House or who works directly for a noble House.

Kaplan-Meier plot for GoT characters split by allegiance to a noble house.

The number in brackets indicates the number of characters falling into each group.

The p-value of the log rank test for these two groups is p=0.

0092, indicating a high level of confidence that the two groups have different survival rates.

Here we see a similiar result to the one above, with the possible reasons being the same: nobles have less dangerous lives or nobles are more likely to be main characters.

Keep it movingBelow the characters are split by whether or not they are “Static”.

A character is classed as static if they spent at least 75% of their screen time in a single location.

Kaplan-Meier plot for GoT characters split by whether or not they are static.

The number in brackets indicates the number of characters falling into each group.

The p-value of the log rank test for these two groups is p<0.

001, indicating a high level of confidence that the two groups have different survival rates.

Here it appears that characters who move around tend to do better in terms of survival.

Characters who spend all or most of their time in a single location are more likely to die sooner.

ComparisonsNext up I’m going to take two of the variables above – allegiance to a house and level of movement – and apply a Cox proportional hazard model.

Below is a forest plot showing the results.

The groupings used are slightly different but derive from the same base variables as the plots above.

HouseAllegiance compares how allegiance to House Stark, Targaryen or Lannister affects chances of survival compared to all other characters.

Allegiance is defined again as either belonging to that house or being a direct servant/advisor of someone in that house.

LevelOfMovement characters who are only ever present in a single location (Single), appear in two locations (Static), appear in three locations (Dynamic) or appear in 4 or more different locations (Nomadic).

The x-axis here is the hazard ratio.

A hazard ratio greater than 1 (to the right of the dotted line) indicates that having that particular attribute increases your chances of death.

Likewise, a hazard ratio less than 1 (to the left of the dotted line) indicates that having that attribute decreases your chance of death.

Forest plot for the results of the Cox Proportional Hazards model.

The numbers on the far right indicate the p-value associated with that group, showing that most groups are significantly different to their baseline group.

Here we can see that owing your allegiance to House Stark or House Lannister significantly improves your chances of survival (lowers your hazard ratio).

Allegiance to House Targaryen increases the hazard to your life (it is to the right of the dotted line) but not significantly (there is some overlap).

This suggests that Daenerys has always been a risky person to put your faith in.

We can also see that there is a strong relationship between the level of movement a character has and their chances of survival.

Characters who move around and are ‘active’ have a much better chance of survival than those who stay in the same place.

ConclusionsSo it appears that in order to survive in A Game of Thrones you should make sure you: are born in Westeros, don’t stay in the same place for too long and ally yourself with either House Stark or House Lannister!There are of course limitations to this analysis as well.

Each minute in an episode of Game of Thrones is not necessarily equivalent, some span an actual minute whereas others can skip several days.

This is a problem that is incredibly hard to overcome in TV shows and analysis of this nature but it could affect the results.

It is also important to note this analysis does not consider causal links.

Although there is a link between being Westerosi and surviving that does not mean one causes the other or vice versa, simply that there is an observed correlation between being Westerosi and not having died in the TV show.

I hope you enjoyed reading this and if you haven’t read my first two posts then please check them out here and here.

This is the first time I’ve delved into survival analysis so I’m happy to take any constructive feedback.

Like and share and follow me for more interesting Data Science posts!.. More details

Leave a Reply