Who will die, live in Game of Thrones, Song of Ice and Fire series?

Can’t wait to know the outcome of Game of Thrones? Neither can Dr Richard Vale from Canterbury University who has turned on complex predictive modelling to answer the rather silly problem of who will die or live in the series.

Dr Vale, from the uiversity’s Mathematics and Statistics department, hasn’t been following the addictive series on television, preferring to read the book instead.

Using probability equations, he tries to predict the outcomes in the final two books of the “Song of Ice and Fire” series written by George R R Martin. His findings will be presented at the university's free lecture, on Oct 8 (Wed) as part of its What if series. 

Dr Vake's favourite character is Littlefinger, or Lord Petyr Baelish, a master of coin in the series, a master of court intrigue. “Because I have absolutely no idea what he is up to.”

Think of this as his version of fan art, except for him, his enthusiasm for the fantasy novel is done using his love for equations.

“I was just really just wondering what might happen next, and whether the statistical methods I have learned would help me to find out,” he told Idealog.

He also thought it would be a very good way of getting people interested in statistics.

What went into building his model?

 “The statistics is actually pretty simple as I chose to use a small data set to build the model, which is one reason why the model is probably not very useful.

“But the programming side was actually quite a lot more complicated, and I learned a great deal about how to do Bayesian statistics from trying it out on this rather silly problem. I also learned that I need to learn how to use version control.”

(We don’t know anything about Bayesian statistics but according to what we learnt from the internet, is a predictive model, using mathematics, to infer conclusions based not on a fixed set of data but different variables that are being assigned a probability distribution based on different degrees of what is happening. Confused? We are too! Check out Wiki!)

From a statisticians point of view, the project seeks to answer big questions such as how can we say ‘how uncertain we are about our prediction?’ and ‘how uncertain are we about our uncertainty?’

“One of the most fascinating things about the books is that they are very unpredictable, with long-running and popular characters being unexpectedly killed off and widespread speculation about future plot developments. This makes them an interesting subject for statistical prediction,” Dr Vale says.


He confesses it is impossible to validate the model by applying it to unseen data.

“In this case, that's basically not possible, although I did make an attempt by looking at how the model would have predicted "A Storm of Swords" if it had seen only the first two books.

“But in a sense, this entire project is an attempt to validate the model, because we can see how well it does after "The Winds of Winter" comes out. So it's not right to be impressed yet, because the model might turn out to be completely wrong.”

Wider application?

Is there a wider application of his model to, say predicting stockmarket outcomes or election results?

“This kind of model is not really the way to do predictive modelling, and I certainly wouldn't do it like this if money was on the line.

“But predictive models in general, especially those built with big data, have all sorts of applications in modern life,” he says.

In his lecture, he will shed light on some general aspects of prediction, probability, and forecasting.

He will also describe his model, used in the prediction. “We also discuss the shortcomings of the model and explain why some things probably cannot be predicted in a meaningful way, and why this should not stop us from trying to predict them.”

He applied Bayesian statistics to create a hierarchical model, with data based on the number of point of view chapters assigned to each character in the previous books. The model predicts the number of chapters characters will likely get in the future books.

“From the model you get simulations of the next book so you can see how many point of view chapters each character should have. I did this mostly because, as a fan, I was interested in what characters would have zero predicted chapters, or not, using that as a proxy to discover whether or not they would still be alive.

“I know that Game of Thrones is not really random and it depends entirely on the whim of George R R Martin, but that’s okay. Although this is just a bit of fun, I am using probability to quantify what I don’t know.

“Statistics is meant to be able to answer questions like this. It’s meant to tell us things we don’t know, to put a probability or a price on something.”

Dr Vale adds that people should be aware that it may be hard to avoid major spoilers for the five existing books during the lecture.