22 April 2012

What does April's Pennant Mean?

The Orioles have been atop the AL East for much of April.  The Orioles' success has been largely due to the performances of Matt Wieters, Nolan Reimold, Adam Jones, Jake Arrieta, and Jason Hammel.  This is not exactly a new thing for the team.  Last year, the Orioles stayed above .500 for 11 games with 151 at or below that mark.  This year the team has been able to go 15 strong so far with their heads above water.  These games obviously are not meaningless.  They count in the record book, but with such low expectations coming into the season one has to wonder how meaningful they are.  Six of the eight wins so far have come against two of the least talented teams in the American League (i.e., Minnesota Twins, Chicago White Sox).  The other two have come against a very solid mid-level team (i.e., Toronto Blue Jays).  Games against teams with more elite expectations (i.e., New York Yankees, Los Angeles Angels of Anaheim) have been highly contested.  The end result being that this has been a rather pleasant beginning to the year.

The difficulty is trying to figure out what these 15 games mean to the future.  As an analyst who is somewhat statistically inclined, this is the part of the year where I am a bit uncertain what I can actually write about.  I was discussing baseball with one of my friends at lunch on Friday in front of the Capitol and discussing how everything I have begun writing so far has not inspired me to finish any of the pieces.  She, a peripheral Red Sox fan who enjoys the Fenway experience, replied how it is nice people write critical pieces, but so little has actually happened and so much is left to occur.  In other words, the famed words: Small Sample Size.  SSS has been an often used and often overused term for those of us who engage statistics with more vigor than most.  This means often that in moments like this when our counting stats are minimal in power, we need to lean more on qualitative analysis.

But what does winning April mean?  The Orioles have a shot at it being one game behind the Yankees with a little over a week left to play.

To try to answer this question, I looked at all fourteen American League teams last year.  Specifically, I took their monthly winning percentages and ran some R2s comparing a single month's winning percentage against the average winning percentage of the other months.  By doing this, we can see to how each month's performance was able to relate to the final total.  

Month R2
April 0.05
May 0.11
June 0.05
July 0.07
August 0.59
September 0.05
First off, we see there is an amazing correlation between a team's August record in 2011 and what that team's record was in every month excluding August.  Second, this is just one year with only 14 teams.  I think it is prudent not to automatically assume that August is the defining month.  It is also unfortunate that if it is the defining month then it makes the July trade deadline a bit murky.  Regardless, it appears that maybe April's record is not incredibly useful in predicting the final record for a team.

A secondary question can then emerge: how many games does it take to see how well a team will perform in the future?  To try to answer this, I used the same data and generated R2s comparing cumulative records against the remaining record.

Date R2
May 1st 0.05
June 1st 0.16
July 1st 0.41
August 1st 0.18
September 1st 0.05
The story here is likely that the more games you play, the more certain you are of the true value of a team (measured as wins).  By that I mean, 81 games played gives you a decent idea of the true value of a team and 81 games left to play allows for that talent to represent what they are truly worth.  Having 130 games in the bag will help you know the value of a team even more, but with only 32 games left, you can have some interesting things happen that can skew a record.

Finally, I decided to run a quick regression comparing individuals months (April, May, June, and July) to a club's final record.

Month Coefficient P
April 0.39 0.006
May 0.30 0.02
June 0.25 0.03
July 0.16 0.07
Again, I caution against taking these numbers as anything definitive due to what I assume to be a small sample size.  I find it interesting that the coefficient (weighted value of a month's record related to the final total record) would be the most valuable and that the coefficients trend downward.  I would like to see this with a more robust data set.  Second, I also find it striking that the record for July did not meet the level of significance I set (P = 0.05).  I also wonder how a more robust data set would affect that.  It should also be noted that standard error here is roughly 0.1 for each month's coefficient, so that is a pretty wide range that leaves none of the months statistically significant in difference to each other.

I am not exactly sure what to conclude here other than me want to see what happens with a larger data set.  The first two exercises suggest that winning the April pennant has little bearing on what happens for the rest of the year.  The third exercise provides numbers for a convenient narrative that the beginning of the season appears to establish a perspective for the rest of the year.  That narrative would mean that if a team starts off poorly, then for some reason that performance will have an effect on the rest of the season more so than whatever is accomplished in May, June, or July.  Tempering that perspective is the relatively large standard error.


Patrick said...

As someone that spends the majority of his life buried in SPSS files it is always interesting to see the use of stats on something I love, but at the same time it usually acts as kind of a buzz kill. This article is very interesting and was passed along to me by a baseball friend as he knew I was a stats junky. I usually don't read Camden depot because of its normally, no offense, bleak outlook. I do think it lends a very awesome look at which month explains the greatest amount of variance in terms of outcome but you are right it would be fool hardly to draw conclusions with the dreaded *dun dun dun* SSS. My suggestions to ones that dwell on the high power of the stats from the last 13 losings... We are currently winning, that is better then losing.

Cheers on a solid article.

Jon Shepherd said...

Thanks Patrick.

I hope our site soon is not a bleak place to be. Alas the qualitative and quantitative assessments we do always lead us in a direction that is bleak, but I would like to mention that the past three years we have predicted more wins than the Os wind up getting.

Anyway, here's to better times coming sooner than later.

Patrick said...

Well I never claimed you weren't correct however, sometimes the truth is better to avoid... especially in April ;)

RD said...

You mention after the first set of data that Augusts’ value put the trade deadline as a bit murky. Yet couldn’t it be that the trade deadline is a causative effect on Augusts’ value? After all there is greater discrepancy between top tier and bottom tier teams after August 1st.

Jon Shepherd said...

Wouldn't you expect some carryover to September? Also, exactly how many trades of significance transpire to affect August?

It sounds like a convenient narrative, but it just does not seem to jive completely for me.

RD said...

I guess I was thinking the synergistic effect of more quality on top tier teams and loss of quality players on lower tier teams would push towards August being overwhelmingly higly valued. But you are correct September blows holes through my hypothesis!

Anonymous said...

But you have to admit it has been fun to watch the young players make a diff like Weiters and Reimold