As with last year, I made a spreadsheet with predicted IFComp scores and rankings, based on a linear regression of the IFComp scores from 2018-2021, using IFDB scores, game characteristics, and the number of public reviews as predictor variables. The sheet contains rankings based on 2018-2021 results, rankings based only on 2021 results, and the current IFDB ranking. Right now, it’s sorted using the 2021 model.
My personal prediction is that According to Cain will rank 1st place, because first place is always a parser game or rarely a parser-like choice-based game, and it seems to have the characteristics that high-placing games tend to have (I have not played According to Cain). There aren’t any games that are obviously dominating the way And Then You Came To a Place… did last year. The top choice-based games will probably be Grown-up Detective Agency and Absence of Miriam Lane.
That seems basically right to me. Beyond those three, I’d guess the top five would be rounded out by A Walk Around the Neighborhood and A Long Way to the Nearest Star, with slots 6-10 (not in order) being Prism, Trouble in Sector 471, Esther’s, No One Else is Doing This, and the Archivist and the Revolution. I confess I’m not really sure where Prom Dress will fit – I could see it doing quite well, but it’s such a long game and folks sometimes rank those lower. I’m also curious where Admiration Point will land.
My personal highest-rated games were January, Chinese Room, and Archivist; I tested According to Cain but I’d put it in the same bucket. It’s notable how downbeat these all are! Usually I think comedies will do slightly better in the rankings, but they’re thin on the ground this year - Grown-Up Detective Agency is funny but also has a lot of sad pieces; Esther’s and Walk Around the Neighborhood are probably the most polished, best-written straight comedies but they’re both a bit too slight for me to expect they’ll take the top prizes.
I feel like a low-key trend this year was many of the best games featured middle-aged moms: Miriam Lane, Chinese Room, Admiration Point, Prom Dress, even I think your game though I wasn’t always clear on how old folks were. That’s a way cooler commonality than any claw-foot tub!
The main characters of A&R are in their late 30s-early 40s, so I suppose they’re middle-aged? Also multiple of these games have the middle-aged mom committing an affair, which is another interesting commonality.
I hope in a few year’s time “sad divorced middle-aged mom” becomes a trend in video game protagonists, like the “sad murderdad”.
I’ve just looked at the coefficients. Am I right that the prediction is influenced to a very large extent by the IFDB rating? And then, to a much lesser extent, by the number of Public reviews and whether the game is a parser game or not?
And then “time” and the number of IFDB reviews don’t really seem to have that much influence?
I unreservedly love the amount of math and spreadsheets applied to this Comp. It feels like you guys really see me. Or maybe more accurately, I see you.
It’s not really cute or funny, though, which is also a major predictor. It’s not my personal favorite, but if I had to bet money on it, I’d go with Grown-up Detective Agency for first place.
Yes, that’s right. Interestingly, the model inferred from the 2018-2021 data has rather different parameters, but still gives a similar ranking. That model gives greater weight to the interaction term (rating count * rating).
Also, I’ve just updated the IFDB rating data, since some people posted reviews after the deadline but presumably voted. I also have a “vibes”-based prediction sheet, which has changes to the predicted ratings based on my impressions of the discussions around the games.
It’s also a Biblical / religious game that opens with four quotes from religious texts. When I first started working on Cain, I knew the subject matter and presentation would throw people off if they assumed it was preachy or Sunday school IF.
Sometimes you don’t choose the story; the story chooses you.
I’m curious about the correlation between public reviews and final score. In theory, causation could go either way (or both): people post about games they like, or people play and rate highly games they’ve heard about.
I worry a bit if the latter is true: it’s the reason that there used to be a moratorium on public reviews in the past. I don’t suppose it’s possible to separate the reviews into ‘reviews posted by choice’ vs. ‘reviews posted based on the random order from ifcomp’? I suspect there’s already little enough data that the answer is ‘no’, but I do wonder. I would feel weird about bumping the scores of games handed me by the RNG just because I didn’t get through everything.
The results are out now, and I’ve updated the spreadsheet:
For the most part, the model was rather accurate - the biggest difference between a predicted and actual score was 1.2 points, and the average difference between the predicted and actual rankings was 6.6 (similar to last year). Of the games predicted to be in the top 10, all were at least in the top 20.
I might do some more analysis on the results later…