Statistician Unpacks Forthcoming Game of Thrones Novel Using Bayesian Models

Game of Thrones has already been the subject of several scientific papers, including the one that yielded the production of this awesome geologic map. Now mathematicians are vying to get a slice of the fame and glory that comes from studying the epic fantasy series.

While many have predicted the fate of our favorite characters in the final books of George R.R. Martin's A Song of Ice and Fire series, few sources have the credibility of Dr. Richard Vale. Vale, Professor of Mathematics at the University of Canterbury, has just released his predictions for the number of chapters told from the point of view of each character in last two highly-anticipated novels.


Prologues and epilogues aside, so far, 24 characters have had chapters told from their point of view. Based on data from a French fansite, Vale has built a statistical model that predicts the number of point-of-view chapters for each character in Martin's forthcoming novel, The Winds of Winter.

"In general, the best predictions are obtained by a combination of modeling and common sense," Vale writes. "Here we focus entirely on the modeling side and leave common sense behind. The question to be answered could be expressed as: What could be predicted about future books if we knew nothing about the existing books except for Table 1?"

Table1: The data. Novel titles are abbreviated using their initials, so for example AGOT= A Game of Thrones etc.


I'll spare readers the nuances of the model, but point any with an interest in Bayesian statistics to the original paper. Figure 1 sums up the results in a series of histograms (probability distributions that give the likelihood of a given number of POV chapters for a given character):


Perhaps unsurprisingly, characters that have already been killed off have a high probability of zero POV chapters in The Winds of Winter. The distribution for Tyrion has the highest variance, followed by that of Jon Snow. Interestingly, Figure 1 also tells us that the probability of Jon Snow not being dead is at least 60%. While discerning readers of book 5 may place the probability of Jon Snow being dead at much higher than 50%, Vale cautions that the model is unaware of events in the books and relies only on the data presented in Table 1.


Other limitations of the model include the assumption of an underlying Poisson distribution, the lack of independence among predictor variables (the model ignores the fact that if one character has more POV chapters, other characters will necessarily have fewer), and the potential introduction of new characters, as has happened in every book so far.

But for those willing to accept a few caveats, it's comforting to know that some aspects of the fantasy saga's conclusion are destined by science. That is, as long as George R.R. Martin himself doesn't pull a Robert Jordan on us.


Madeleine Stone is the author of this blog and The Lonely Spore. Follow her on Twitter.

Share This Story

Get our newsletter