After seeing folks comment on the in-game predictions, I got to wondering…just how good are these estimates the game comes up with? Looking at the forum, I saw Recte posted this year’s predictions in the Sim #1 thread, and added a couple numbers together to have it dawn on me that maybe the predictions for last year were in the 2036 threads.
That’s right. I mean. You gotta get up pretty early to put one past me.
So, using last year’s data, I grabbed the most valuable and Excel-lent tool ever devised, and did some number crunching. How good are the predictions? How many are right or wrong? When they’re off, how far are they off? How many games do they miss? How do they look across the divisions? In other words, can they tell us anything about that infamous “which division is the best in baseball argument?”
As always, here’s the data…I’ll get to analyzing in a bit.
The Big Misses
Bottom line, the 2036 Opening Day predictions had some big misses both ways. The game under-estimated Atlantic City by 22 games to the worse, and Jacksonville by 16. In the Frick, it judged Seattle to be 21 worse than they performed. Going the opposite direction, the software scouts gave Phoenix the benefit of the doubt by 22 games, Calgary by 23 games, and Edmonton by 20. It expected Louisville to win 14 more than it did.
In other words, it missed seven of our 30 teams by 14-23 games.
Close Enough For Horse Shoes and Hand Grenades
On the other hand, the game did get 14 teams “right” within 4 games, and even called Mexico City at 95 right on the button. So roughly half the league it gets "right." Not horrible when you look at it that way, eh?
Of course, this and $5 will get you a decent book on Amazon if you look around close enough. No one knows which teams are "right" and which are "wrong" until it's tool late. I suppose this would be fun to do with our Media Guide predictions and team previews, too. How close are we vs. the program. Maybe I don't want to know.
League-wide Look:
Here’s the league/division breakdown. Again, data first.
- 2036-projections-vs-actuals-divs.PNG (10.39 KiB) Viewed 918 times
Look at the divisions, though.
How to read the data: Reading across the JLA, for example, the game projected the average JLA team would win 80.4 games, with a standard deviation of 17.1 games. That STDEV suggests that the game thought the JLA would be the least competitive division top-to-bottom (the spread between best and worst would be wider than the other divisions). The 2036 Actual results showed that the division was a little bit better than expected (83 wins), and, yes, at 16.2 STDEV, was not great, but it was better than the FLF’s 18.9.
Now that we’ve been through that, we can see that the game projected the division’s power rankings to be:
2036:
FLP – 84 wins (13.8 STDEV))
JLSB – 83.6 (13.0)
FLH – 83 (13.0)
JLA – 80.4 (17.1)
JLSE – 79.4 (7.7)
FLF – 77 (10.4)
In other words, for example, the JLSE was supposed to be the most competitive league top-to-bottom (it turned out to be #3). And it over-estimated the Frontier and Southeast by an average of 4.6 and 4 game each. The game under-estimated the Heartland the most (3.4 games), the JLA by a bit less (2.6), and the Southeast by only 2 games apiece. Its average error for the Pacific was a pretty reasonable .6 games per team.
Bottom line: the Heartland was the best division in baseball during the regular season, but the JLA and Pacific will be obviously right in lording their Landis appearances over the rest.
