You Are Not Your April Record...or Run Differential

Does this matter, O's & Yankees both 15-8, but....run differential
NY plus-43 (1st in AL)
O's plus-1 (8th in AL)

— Steve Melewski (@masnSteve) May 1, 2017

This tweet was brought to my attention.  Let us not presume what Melewski's intent is here in writing and simply answer the question: what is more important in determining what is likely to occur for the rest of the season?  April Record or April Run Differential.

I took team performance from 2014-2016.  I ran linear modeling for April winning percentage against the rest of the season winning percentage.  Similarly, I ran run differential (as a rate for 20 games) against the rest of season winning percentage.

Utility as a predictor:
April Wins and Losses - r2 = 0.07
April Run Differential - r2 = 0.21

Both are rather poor predictors of future performance, but in practice a difference would be made.  If you were running a population study and saw a correlation of 0.21, then you would likely highlight that as a potential variable that plays a role in whatever you are looking at.  A variable with a 0.07 coefficient is one that will almost always be tossed out as having little to no predictive value.

OK, but what about very successful April teams?  How do the metrics work for them (n=18)?
April Wins and Losses - r2 = 0.03
April Run Differential - r2 = 0.30

This also seems to affirm that April Run Differential is more meaningful, but it may be over or under stating things due to a much smaller sample size.

But what about a team like that Yankees with a plus 40 run differential and a team like the Orioles with a neutral run differential?  Do teams like that act differently?

The Orioles are performing at a .147 winning percentage above their Pythagorean win expectation. 

For teams performing .070 and above (n=11):
April Wins and Losses - r2 = 0.15
April Run Differential - r2 = 0.24

For teams underperforming between -0.048 and -0.058 (n=11):
April Wins and Losses - r2 = 0.45
April Run Differential - r2 = 0.45

Amazingly, it looks like these metrics work much better for teams in the Yankees' position.  For the Orioles, Run Differential still looks like the preferred method, but Wins and Losses became more interesting.  That all said, I would prefer to use the correlation values from the whole 90 team data set than these sub-divisions because there may be an issue with small sample size.

Regardless, for the Orioles, there does not appear to be a great reason to ignore run differential or prefer team record over run differential for expecting how the team will perform in the future.  Instead, one should probably look to the impact of Chris Tillman and Zach Britton returning, but, of course, that means to ignore the probability of other injuries to key players occurring.