The Great Momentum Debate Round 4
Research: Here is an interesting article that proves there might be momentum from [[link|url=http://www.beyondtheboxscore.com/2005/10/24/10107/273]]Beyond the Boxscore[[end-link]] , a blog dedicated to the study of statistics and baseball.
Article opposing momentum from the Wall Street Journal: http://online.wsj.com/article/SB119093290865641816.html?mod=todays_us_weekend_journal.
And of course, the debates of WVBR sportscasters [[link|url=http://wvbr.com/sportsblog/393]] Jay Sage [[end-link]], [[link|url=http://wvbr.com/sportsblog/394]] Eugene Karlik [[end-link]], and [[link|url=http://wvbr.com/sportsblog/395]] Rahul Desai [[end-link]].
Method: I decided to try and find some way to obtain data that would somehow help determine if momentum existed. I decided to look to see if there was some sort of correlation of finishing the season strongly and performing well in the post-season. I gathered all of the overall winning percentages of every team (N = 112) that has made the post-season during the wild-card era. I also obtained the splits of winning percentage before September and the regular season winning percentage in September and October.
After obtaining the data, I performed numerous tests to determine independence between several factors of my data, and I determined based on a 95% confidence interval that the winning percentage of a team before September and the difference in winning percentage after September 1 would be the factors I would use. I called them PRE and DIF.
I performed several ordinal regression tests to see if the factors PRE, DIF, and their INTERACTION had any affect on team’s PROGRESS in the playoffs.
Results: I first performed an ordinal logistic regression test involving the PRE, DIF, and INTERACTION. The Person Goodness-of-Fit test (p = .318) seemed to suggest that the full model did in fact fit the data. Upon closer review, none of the effects seemed to be significant based on a 95% confidence interval: DIF (p=.110), PRE (p=.078), and INTERACTION (.103).
I decided to drop INTERACTION and perform another ordinal regression using only PRE and DIF. Also INTERACTION had a lower p-value than DIF, I felt that for the purpose of this study, DIF would be easier to interpret. Again, the Pearson Goodness-of-Fit test (p = .490) suggested that the model did fit the data. Also, the Log-Likelihood test (2.80 < 3.84) allowed me to use the reduced model. Unfortunately, the reduced model only using DIF (p = .696) and PRE (p = .141) both indicated that neither indicator was significant in determining post-season success.
I decided to drop the DIF variable because of its very large p-value and see if there was any significance just using just PRE. Again, this reduced model passed the Pearson Goodness-of-Fit test (p=.119) and the Log-Likelihood test (2.95 < 5.99). This time, the main effect PRE was significant (p = .041) at with 95% confidence. It had an odds-ration of 0.00028 and a coefficient of -8.169.
After determining the relationship between pre-September winning percentage and post-season success, the next logical step would be to look at overall winning percentage and post-season success. I performed an ordinal logistic regression model using just OVERALL, which is the winning percentage of overall winning percentage of all the playoff teams and post-season success. I found out that overall winning percentage (p = 0.098) was actually insignificant using a 95% confidence interval.
Discussion: Despite pre-September winning percentage having significance with a 95% confidence interval, the overall winning percentage suggested otherwise. This is only a 95% confidence interval, which means there is some error. Both the p-values for PRE and OVERALL were extremely close to 5%, so it is a matter of what one feels comfortable at being confident. Both are not overwhelmingly confident, but both are not so large that they should be ignored. If one really wants to be correct, ignore these factors when the playoffs come around. If one wants to see a very loose general trend, using the coefficients for OVERALL (-7.975) or PRE would not be a bad determinate.
The biggest limitation in this study was independence. Unfortunately when you look at team winning percentages, each team is related to another because the winning percentage is based on how one team performs against another team. Also, with playoffs, each team has to face each other, which was a problem I knew going into the study. The large N size of 112 teams did a good job reducing the dependence for this factor, but it is impossible for complete independence. If I were to perform this study again, I would look at American League and National League performance separately and see if there was some sort of relationship between each league’s results.
Conclusion: Based on my results, it would appear that a hot September does not have an effect on playoff success. Based on my some research and my own empirical observations, once a team makes it to the playoffs, the results are random. All teams that make it to the playoffs have to be talented, and winning a short series involves a lot of luck that would otherwise be minimalized if each team played a whole season of 162 games against each other.
Personal Opinion: Although there are a lot of flaws to my little study, it did exemplify some of my own personal beliefs on the subject of momentum in baseball. Baseball is a game of 25 players each doing one’s own part. Players do go on hot streaks and others go on cold streaks. I would agree with Jay that there is individual momentum. One play could get an individual into a groove and one play could ruin one’s game for awhile.
As for team momentum though, I am a non-believer. In my opinion, everyone’s individual momentum balances out to the basics of whether a team is talented or not. Although there might be some positive and negative momentum based on individuals, the sum is so minimal that is has very little impact on a team.
So the debate has raged to an intense level among WVBR sportscasters debating the question of whether momentum exists in baseball. I decided to take a statistical approach to find a solution to the answer. Based on tools and time, my study is not perfect (I will point that out right away), but it is very thorough and reproducible (if someone wants the data, let me know).