As a refresher, I set up an experiment using Out of the Park Baseball for opening day 2019. If you did read that post, I updated the results which show several incorrect predictions. Did the baseball simulator fail for opening day 2019?
If you read through my assumptions, you'll see that the simulator did exactly as I suggested it might. The assumptions were that predictions made between two closely matched teams would be similar to tossing a coin for the simulator. These coincide with the strength indicator < 1.15. Some of those closely matched games did get predicted properly, but many didn't.
If you have ever flipped a coin before, this result shouldn't surprise you. Even if the matching isn't an even 50/50 split (most weren't), they were close enough for these predictions to make sense.
On the other hand, most of the predictions where the strength indicator was >= 1.15, the simulator predicted correctly. At the time of this writing the Boston/Mariners was about half way through with the Mariners having a healthy lead. If the game were to end now, this would be the only game where the simulator predicted incorrectly with the strength indicator >= 1.15. In fact, for the game, the strength indicator is 1.46 in favor of Boston. This means Boston should win which makes the simulator less accurate, right?
Not quite. First, it's early in the game, especially for a strong team like Boston. Don't count them out yet. In the morning, I will update this post with the final score and we'll see how the game played out.
Second, another assumption I made was that 5,000 games were played. If it were even possible for the same two teams to play 5,000 games (it's obviously not), then I have every confidence in the law of large numbers would give the results of those 5,000 games to Boston, handily. I mentioned in the previous post that you cannot judge results of one game as every team has an off night or two.
I believe that the experiment was a success and the simulator performed the way I had described it would, and should. It could turn out that the simulator wasn't designed for this type of analysis and as I have stated previously, I am still learning how the simulator works. Therefore, it's possible I didn't run the simulator properly. However, from my perspective, this probably means the simulator would do even better with its predictions after I would learn how to run it properly.
I will post this and future experiments on the forums for OOTP. I will ask if I am setting up the experiments correctly and adjust if I am not. If it turns out that the simulator isn't supposed to work in this manner, I may continue to see how it plays out with my assumptions in place.
Here, again, is the link to the experiment and the results for your review.
