Mortality and Nobility in the Wars of the Roses and Game of Thrones

The Game of Thrones universe is famous for its high mortality rates, but for persons belonging to the nobility the death tolls are quite similar to those of nobles living in England during the Wars of the Roses.

"Siege of London" by Unknown, possibly Jean Spifame, image from French Ministry of Culture. Licensed under Public Domain via Commons -

Game of Thrones1 (GoT) is a widely popular television show, famous (or rather infamous, some would say) for the high mortality suffered among its main characters and its general violence. In addition to its popularity the show has become a subject to some serious analyses concerning its morally ambiguous characters, political intrigues, gender issues and historical background. It is generally recognized that the civil war raging in the first seasons of the show (the War of Five Kings) bear some striking resemblances (or obvious references) to the historical civil war raging in England between 1455 and 1487, the Wars of the Roses. The civil war in GoT starts for example as a conflict between the fictional noble houses of Lannister and Stark, while the Wars of the Roses concern the historical noble houses of Lancaster and York.

One aspect worth investigating is whether the civil war in GoT is bloodier than its (assumed) historical inspiration. The data and some of the inspiration for this post come from this article, which among other things compares the mortality among characters in GoT to the general mortality in modern countries (like Afghanistan). This comparison is obviously problematic, because the characters in Game of Thrones can hardly be assumed to be a random sample of the imaginary population living in Westeros (≈ England in GoT) at the time of the events in the series. A more natural analysis would be to compare the mortality among characters in GoT with the mortality suffered by the historical persons involved (directly or very indirectly) in the Wars of the Roses. In order to obtain such a dataset I used Wikipedia, and acquired the age at death of as many persons as possible living in Great Britain between 1455 and 1487. Neither of these datasets are random samples of the populations living in Westeros or Great Britain, but before looking at the data I expected both datasets to be biased in a similar fashion, towards notable persons, and more specifically towards adult, noble, males. Children of peasants dying at a young age have a low chance to be portrayed in GoT and also a low chance of having a Wikipedia page. The analysis in this post should not be taken as a comparison between the general mortality in GoT and 15th century Great Britain, but as an analysis concerning a specific, yet somewhat loosely defined, subset of the population: important/notable persons.

Table 1: A non-random sample of characters from the GoT dataset. Note that not all characters in this dataset are dead.

As mentioned I obtained the GoT data from this post, and the interested reader can consult it for information on how the dataset was assembled. The original data contained age, gender and affiliation (which noble house/group the character belongs to) for 369 (mostly) named characters in the TV series and books. The dataset also had a variable ("dead") displaying whether the character is still alive in the show/books or not. I removed 41 characters in the dataset whom I knew to be historical in the GoT world (long dead kings/nobles, whose age at death would not be relevant for the civil war period I am considering). I also replaced the affiliation variable with "nobility" displaying whether a given character is a nobleman/women or not. Parts of the data are given in Table 1 (no major spoilers if you have seen season 1!).

Table 2: A non-random sample of persons from the Wars of the Roses dataset. Note that all persons are dead. Also note that Thomas Wolsey is listed as a noble, although he actually was the son of a butcher. He rose however to become a cardinal, and I have defined all higher church officials as nobles. Old Tom Parr was removed from the dataset as his age was deemed unlikely.

For the Wars of the Roses dataset, I first acquired the names of all persons from the four Wikipedia categories "15th-century English people", "People of the Wars of the Roses", "15th-century Scottish people" and "15th-century Welsh people", by using the API webservice (API:Categorymembers). I used the R package jsonlite and obtained a list of 1172 names (after removing duplicates), then I wrote an R-script accessing the Wikipedia-articles of all these persons and obtaining their date of birth, date of death, gender and profession/occupation. Many persons had incomplete birth or death dates and were removed from the dataset. After restricting myself to persons alive between 1455 and 1487 and removing one unlikely observation (see Table 2), I ended up with 407 persons. As in the GoT data I manually entered a "nobility" variable.

Figure 1: Kaplan-Meier curves for GoT characters (red) and historical Wars of the Roses persons (blue).

As expected both datasets consist of more nobles than commoners and of many more men than women. In the GoT dataset there are 55% nobles and 76% men, while for the historical dataset the numbers are 68% nobles and 88% men. The attentive reader might have noticed a very important difference between the two datasets: in the Wars of the Roses data all persons are dead (obviously), but in the GoT data some persons are still alive. In fact 53% of the 328 characters are alive. We are thus faced with the statistical concept of censoring, we do not know the age at death for the living characters, we only know that their age at death must be at least their current age (and they might become much older, or not). One way to present this kind of data is looking at so-called Kaplan-Meier plots. The red survival curve in Figure 1 is the GoT curve, the crosses representing the ages of characters still alive. At age 0 everyone is still alive (the survival curve start at 1) and at some age past 100 every person is dead (the survival curves reach 0). From the curves, it seems clear that the mortality among the GoT characters is higher than among the Wars of the Roses persons (although there are more very old characters in GoT than in the historical data). We can test whether the curves are significantly different with a nonparametric log-rank test, and they are (p-value=0.0055).

In order to do the rest of the analysis, I merged the two datasets and added a variable "origin" with value 1 if the character belongs to GoT and 0 otherwise. Then I investigated the effects of the others covariates in the data, gender and nobility (and their possible interactions with each other and with origin). There were no significant gender differences, but nobility proved to be more interesting. Looking at only nobility across both origins, there seems to be no difference between commoners and nobles (the black curve, in Figure 2 A) (note that now there are GoT characters and Wars of the Roses characters in both groups), but when I look at nobility and origin together a surprising pattern emerges: the differences in mortality between nobles and commoners have opposite patterns. In the Wars of the Roses, commoners seem to have lower mortality than nobles, but in the GoT-world the opposite is true; it is more dangerous to be a commoner than a noble. In fact, when looking only at nobles, the survival curves from GoT and Wars of the Roses are not significantly different (p-value=0.728; note that the dark blue and dark red curves overlap in Figure 2 B).

Figure 2: A: Kaplan-Meier curves for nobles (black) and commoners (grey). B: Kaplan-Meier curves for commoners from GoT (light blue), nobles form GoT (dark blue), nobles from Wars of the Roses (dark red) and commoners from Wars of the Roses (light red).

One could thus argue that George R. R. Martin (the writer of the book series) correctly captures the mortality suffered by the nobility in his fantasy version of the Wars of the Roses, but that the fictional war in GoT seems much bloodier for the commoners. This aligns well with my prior knowledge: the Wars of the Roses were bloody for the competing noble families (see this nice diagram), but the fighting generally did not involve commoners (unless you were a soldier; see Encyclopedia of the Wars of the Roses, by John A. Wagner, p 59). In the civil war in GoT however, it is clear that commoners suffer extensively, there are multiple reports of massacres, refugees, hunger and a general breakdown of society. However this remains mostly speculation; as I wrote in the beginning I cannot say anything about the general mortality from this analysis. Another explanation is that there are other mechanisms driving the likelihood of a commoner being mentioned in GoT, than the likelihood of an historical person having a Wikipedia page. In GoT a person (and possibly especially a commoner) might be mentioned specifically because he/she dies, while a historical person dying at a young age might have a smaller chance of landing in Wikipedia, because such a person has had less time to do something memorable, like for example being elected to parliament. This brings me to my last attempt at an explanation: the society in GoT and the historical 15th century England have (at least) one fundamental difference: Parliament. The society in GoT is a classic feudal society, while 15th century England was not, amongst other things England had a parliament, which was not always very powerful, nor very democratic by today's standards, but nonetheless allowed commoners to excercise some influence. In fact, a large portion of the commoners in the dataset are actually members of parliament. These peaceful (?) and rich commoners were probably not involved in the fighting, and their long and prosperous lives might explain the lower mortality for commoners in the Wars of the Roses dataset.

I might come back to this dataset in later posts. Both datasets can probably be extended, and I also want to try some parametric and semi-parametric models on the data.


1 The show is based on the book series A Song of Ice and Fire by George R. R. Martin. The analysis in this post is actually mostly based on data from the book series, but I will refer to it as GoT anyway (because it is shorter, and possibly more well-known)

Tags: History, Game of Thrones, Survival Analysis By Céline Cunen
Published Oct. 2, 2015 4:18 PM - Last modified Aug. 25, 2016 10:58 AM

Cool stuff indeed. I've used your data to fit my gamma process level crossing models, and get sensible answers. Here each person in WoR has a gamma process

Z_0(t) \sim Gamma( a_0 M(t), 1)

governing his or her fate; when Z_0(t) crosses the level

c_0 = \exp(\beta_0 + \beta_1 x_1 + \beta_2 x_2),

he or she dies, and with the threshold determined by (x_1, x_2) = (nobility, gender). Correspondingly, each person in GoT has a gamma process

Z_1(t) \sim Gamma( a_1 M(t), 1)

in his or her rucksack, and death ensues when the process crosses the level

c_1 = \exp(\gamma_0 + \gamma_1 x_1 + \gamma_2 x_2).

This model has two a parameters + three betas (for WoR) + three gammas (for GoT) + one more parameter for the M(t) home function. The point is then to both estimate all parameters and check for significances and differences (I've done so, and will write it up, after checking details with you, Céline). Apparently, belonging to the nobility crowd give you worse chances in WoR, but better survival chances in GoT.

We would also need to invent decent goodness-of-fit tests for these models, incidentally.

Nils Lid Hjort - Oct. 3, 2015 12:29 AM

Céline Cunens foredrag "Er Game Of Thrones blodigere enn middelalderkriger?", holdt 9. mars 2017 på UiOs Åpen dag, ble filmet, og ligger i NRKs arkiv "Skole", viser det seg:

Nils Lid Hjort - July 13, 2017 9:24 PM
Add comment

Log in to comment

Not UiO or Feide account?
Create a WebID account to comment