Are the Red Sox Going to Win at least 116 Games this Year?

Sweet Caroline has been played at every home game at Fenway Park since at least 1997. Bap Bap Bahhh

​No. No, they are not.

Wait, don’t go! Hear me out. And I still have some good news for any Red Sox fans out there at the end of this post.

In case you haven’t seen, the Red Sox are in the process of breaking baseball. Over the weekend, they swept the Yankees at Fenway in what could probably be a preview of the ALDS this postseason. As of this writing, on August 7, the Red Sox have a record of 79-34, for a winning percentage of 0.699. They are in first place in the AL East, with a 9 game lead over the Yankees, who just happen to have the third best record in all of baseball. ​

With this fiery winning percentage, an article came out on MLB this morning discussing the potential of this team to tie, or even exceed, the current modern single season win total of 116 games, set by the 2001 Seattle Mariners. The article does a great job discussing the history of teams chasing the record and trying to exceed the 0.700 winning percentage barrier, but does not delve too much into the projections and mathematics.

Since MLB entered the Expansion Era in 1961, this year's Red Sox are the fifth club to win 79 of their first 113 games, joining the '69 Orioles (79 wins), '98 Yankees (84), ‘01 Mariners (82) and '17 Dodgers (80). Of those four teams, only the ‘98 Yankees and ‘01 Mariners finished with a wp% over 0.700.

I have been writing extensively in the last few weeks on team winning percentages, expected wins, and the Bayesian approach to modeling a team’s season. As such, I couldn’t help but dig into the numbers on the 2018 Red Sox.

So why won’t they win 116 games? Well, the Red Sox have a 0.699 wp% with 49 games left to play. If they win those games at the same clip, then they would expect to go 34-15, for a final record of 113-49. They would have to exceed their current insane wp% over their last month and half to reach 116 wins. Well, if they have been deemed ‘unlucky’ so far this season, then this could be possible.

Unfortunately, their Pythagorean Record, as of August 7, was 76-37, due to their 600 Runs and 409 Runs Allowed. This works out to a wp% of 0.6725. If we expect the Red Sox to win at that rate over the last 49 games, then we would expect them to go 33-16 for a final record of 112-50. So not only do they need to exceed their current winning rate, but also be luckier than they already have been up to this point in the season.

Quick intermission. This is insane though. Being “on-pace” or having a win expectation this high is ridiculous, and would obviously be the highest win total in the MLB since the ‘01 Mariners. The only other teams that have come close since 2001 are the 2004 Cardinals (105) and the 2017 Dodgers (104). The Red Sox should be the story of baseball right now. They have an unmatched dominance which all of the numbers support.

But onto my actual opinion and analysis. I recommend reading my last post on creating a Bayesian prior for MLB wp% in 2017, as I did the exact same thing for the 2018 season. I used three over/under predictions from bookies for win totals for each team, as well as preseason win total forecasts from Fangraphs and Baseball Prospectus to come up with my average expected win total, and the standard deviation from 2017 winning percentages, to establish my regression-to-the-mean constant for each team and create my prior.

The prior for the 2018 Red Sox is Beta(40.81, 31.68) which looks like this:

​As you know, if you read my other posts, this is the hardest part. Now I just added the win total to the alpha, first, parameter, and the number of losses to the beta, or second parameter. This means the posterior distribution is modeled by Beta(119.81, 65.68), which looks like this:

This model represents the underlying “true” winning percentage of the Red Sox, based off of our prior expectations of the team and the evidence we have gathered so far. The mean, average, of this distribution is 0.646. This makes sense when compared to the Pythagorean Record. While slightly lower, our prior expectation for their season was only in the upper 0.500s. The model hints that the Red Sox have been a little lucky, and if they were to replay the season up to this point, they would have a slightly lower wp%.

I assume that Red Sox are more likely to regress towards this average wp% over the last 49 games of the season. If so, they would still go 32-17, for an expected win total of 111, which I can’t emphasize enough, is still absolutely insane.

However, since this is a distribution, there is a chance that their true winning percentage is actually much higher, and therefore if they regress to their true wp%, they might actually win 116 games. By sampling from the posterior distribution and multiplying by 49, I came up with a probability that the Red Sox would actually go 37-12 (0.755 wp%). It’s essentially zero. This is confirmed by the 95% confidence interval I created around the expected wins. For those curious, the probability is 95% that the Red Sox finish with between 107-114 total wins. The 99% interval is only slightly wider, at 106-115. Since 116 lies beyond the 99% CI, the probability of it occurring is essentially 0.

If we wanted to get all frequentist up in here, gross, we could say that the Red Sox not winning 116 games is statistically significant at the alpha = .01 level as of the data gathered through August 6th.  

I ended up modeling the probability that the Red Sox would win at least 116 games from every day so far of the regular season. This was done by creating a posterior distribution for every day, then sampling for a wp% which was then multiplied by the number of games remaining in the season. Add that to their current wins and you got your estimate. This was basically what we just did in the preceding couple of paragraphs.

The plot is interactive, so try out hovering over it, zooming in and out, or even download as a PNG. Probably works best in browser.

But all is not lost! The Red Sox have a franchise record of “only” 105 wins, set in 1912. I have already mentioned that I expect them to end the season with 111 wins, so it seems likely that they will break that record. I modeled the probability that the Red Sox would win at least 106 wins, guaranteeing a new franchise record, every day of the regular season as well. That plot is below.

While it is unlikely that the Red Sox will set a new MLB single season win total, they are more than likely to set a new franchise record. If you liked this research, why not check out my new Playoff Odds? I use the same methodology to calculate expected wins for each team on every day in the regular season. Since these are taken from a probability distribution, I could also turn those into Total Playoff, Division, and WildCard playoff estimates. Let me know what you think!

Currently I have every chart up for the 2017 MLB season, as well as probabilities up through the end of day on August 6th for the 2018 MLB season. I am currently in the process of calculating the same for the NBA, since all ESPN can talk about is how the Lakers will do with LeBron.

If you’re curious, I recommend comparing these predictions to Fangraphs and the BaseballGauge. My concern with Fangraphs’ is the high inaccuracy of modelling every season game based on runs scored and allowed, which I only use for predicting postseason games when I have a large enough sample. The BaseballGauge, on the other hand, begins each season by giving every team the same odds of winning their division or making the playoffs. This makes small samples biased, as I discussed in my last post on choosing prior expectations for a team for a season.

While no forecasting model is perfectly accurate, I believe my results from 2017 show that my forecasts converge faster and identify quicker the teams that make the postseason, as well as more accurately predict final win totals when confronted with small samples.

Anyways, thanks for stopping by, I appreciate you! Feel free to let me know your thoughts in the comments below, or on Twitter. As always though, our code, data, and beta parameters can be found on Github. The coding language used was Python, and the API I used to get daily win/loss records was XMLStats.

The SaberSmart Team
​
P.S. If you enjoyed this article, and need something off Amazon anyway, why not support this site by clicking through the banner at the bottom of the page? As a member of the Amazon Affiliates program, we may receive a commission on any purchases. All revenue goes towards the continued hosting of this site.