Final Hugo Predictions 2023

Edit 7/7: 2024 list posted! https://mrphilipslibrary.wordpress.com/2023/07/06/hugo-predictions-2024/

I thought it would be a good idea to start a brand new post with my final predictions to make it all more legit and streamlined than just editing the post I’ve been updating throughout the year. I won’t be updating these further. I included a seventh slot below each category only to be considered in the event that it’s later revealed one of the six official predictions declined a nomination, as has been the case in the past. Keep reading below for an explanation of the prediction model as well as some general discussion.

Novel:

Novella:

Notes:

• Just to make it clear, these lists are not a representation of the works I think should be finalists, or even the works I think will be finalists, they’re the result of predictive modeling based on data. My gut sometimes disagrees with the predictions.
• Having said that, I think both of these prediction lists look pretty good. I wouldn’t be surprised if Sea of Tranquility does not make the cut because the author’s work is generally considered to be in the realm of literary fiction versus SFF. Three additional novels I could see show up are The Spare Man by Mary Robinette Kowal, Seasonal Fears by Seanan McGuire, and Legends and Lattes by Travis Baldree. Mind you, these are gut feelings – TSM and SF are both actually way down on the prediction list. Other possible novellas are The Sins of Our Fathers by James S.A. Corey and Ogres by Adrian Tchaikovsky (although again, those are just personal thoughts), or High Times in the Low Parliament by Kelly Robson (which is more likely according to the model).
• Spear by Nicola Griffith is technically eligible as a novel and a novella under official Hugo rules. As such, I’m including it in the novella list even though it has been nominated for other awards as a novel. I also included it in the novel prediction model, but it didn’t crack the final list.

Methodology:

The model I use to make the predictions is continually a work in progress and I regularly train it to make it as accurate as possible (although at this point it’s all just very minor tweaks) which is the reason it sometimes changes seemingly without any new information. I’ll explain a specific example of this as well as my general methodology for further context.

As a disclaimer, I’m not a coder and do not use any sophisticated programming. I’m a pseudo-statistician who has researched predictive modeling to design a formula for something that interests me. I first noticed certain patterns among Hugo finalists that made me think it would be cool to try and compile those patterns into an actual working formula. I use a discriminant function analysis (DFA) which uses predictors (independent variables) to predict membership in a group (dependent variable). In this case the group is whether a book will be a Hugo finalist.

I’ve compiled a database of past Hugo finalists that currently goes back to 2008. Each year I use a dataset that includes information from the previous 5 years to reflect current trends that are more indicative of the final outcome than many years of past data (Pre-Puppy era data is vastly different than the current Post-Puppy era despite not being that long ago.) I also compile a database of books that have been or are being published during the current eligibility year. Analyzing those databases generates a structure matrix that provides function values for different variables/predictors. For these 2023 predictions, 28 total predictors were used. Each predictor is assigned value based on how it presented in previous finalists, and how it presents in the database of current books. My rankings are simply sums of the values each book receives based on which predictors are present.

Predictors cover four general areas: “Specs” such as genre, publisher, and standalone/sequel; “Awards” meaning performance in other awards leading up to the Hugos; “History” meaning an author’s past Hugo history; and ”Buzz” such as inclusion on various reader lists, bestseller performance, and whether a book receives a starred review from a prominent publication. 

Although the prediction model has historically been pretty accurate (>80% correct on average), it can also be pretty off. For example, last year it predicted 5/6 of the novel finalists, but the other one, She Who Became the Sun by Shelley Parker-Chan, wasn’t even in the top 10 for various reasons. For the novellas, the work that was considered most likely to become a finalist by the model, Remote Control by Nnedi Okorafor, didn’t make the cut.

Sometimes I’ll consider a new variable and evaluate whether I have enough previous data to use it, and whether its predictive power makes it worth including. Sometimes I’ll re-evaluate the data I’m already using and determine if I’m utilizing it effectively. Here’s an example regarding Hugo History as a predictor. Previously I assigned separate predictor values to books written by authors who are previous winners, previous finalists, and have been previously longlisted, and included each separate variable in the model. I since determined this made the variables redundant leading the model to overfit the data, and transitioned to assigning works value based on whether they were written by previous Hugo winners/finalists or previously longlisted authors.

Feel free to let me know if you have any questions about how it all works!

Discussion:

Chengdu Worldcon posted a Twitter update today saying they’re looking to release the Hugo finalist list by the end of June and are contacting individuals in preparation. A previous tweet said “due to the amount and complexity of this year’s nominations we have received, the committee still needs some time to work out the official nomination list.” I wonder what that means. It’s wild that it’s taken two whole months. Maybe I’m naive, but I don’t think it should be that difficult to just run the E Pluribus Hugo nomination process and figure it out. My overactive imagination can’t help but speculate that there’s some juicy behind-the-scenes drama between Chengdu and the Hugo administrators, and something untoward is going on. It has been mentioned in various places that because Worldcon will take place in China this year, after a big effort by Chinese fans, it’s very possible that native authors will take up a lot of the nomination slots. This would essentially make my predictions useless because all of my data comes from English-language websites, lists, awards, reviews, etc. No way to account for that!

I’d love feedback on these posts in the future! I have always just updated my previous prediction posts by editing them throughout the year, but I’m not sure if that is the most effective way to do it? Should I archive my various edits as reply comments as I go? Feel free, readers, to comment and discuss!