My preference is for people who finish the year strong, rather than those who start off with a bang and end with a whimper. To the best of my knowledge, there’s no reason to believe this. There’s no credible study that shows if you hoard guys who finished strong in the previous year that you’ll find success in the following season. But wouldn’t it be nice if this was the case?
Part of the issue is defining what a strong finish is. Is September enough? The final six weeks? The entire second half? How you frame the question will impact the results. In 2013, Justin Turner had a .764 OPS after the All-Star break. That’s a solid total but where he really did his damage was in September. That month, Turner had a .929 OPS.
In the first half of the year, Turner had a .637 OPS. In July and August, his OPS was .658 – or pretty much what we would expect from his previous numbers. September is where it all came together for him. But his entire second half consisted of 112 PA and only 42 of those came in September. If we’re going to look at this, we have to accept that we’re dealing with small samples.
So, we’re going to use second-half results to be our strong finish indicator, since that will give us the biggest sample. We’re going to require 100 PA in both the first and second half and we’re going to require an OPS difference of 50 points in either direction. Here are the results from the 2014-2018 seasons. We’ll start with the guys who improved 50 or more points in the second half:
|Year||Name||1st Half||2nd Half||Diff|
That’s a fair number of players who improved in the second half. But with the second-half kicks of 2015, 2016 and 2018, it’s not all that surprising. It’s nice to see that 11 of the 18 people in this grouping improved by over 100 points. Now let’s take a look at the players from this same time period who saw their OPS drop by at least 50 points:
|Year||Name||1st Half||2nd Half||Diff|
There aren’t nearly as many people in this grouping as there were in the first one. But 2014 was well represented here. A tiny bit surprising, as the Mets were five games below .500 in the first half and a game over in the second half. Of course, Jacob deGrom going 6-1 with a 2.18 ERA in the second half influenced things some.
Now the question is: Did the second half performance indicate how they would do the following season? Let’s take a look, starting with the players who improved in the second half. We’ll compare their season-long OPS from one year to the next:
2018 Conforto .797 OPS, 2019 .856 OPS
2018 Nimmo .886 OPS, 2019 .783
2018 Bruce .680 OPS, 2019 .784
2017 Reyes .728, 2018 .580
2017 Cabrera .785 OPS,,2018 .774
2017 Flores .795 OPS, 2018 .736
2017 Cespedes .892 OPS, 2018 .821
2016 Cabrera .810 OPS, 2017 .785
2016 Walker .823 OPS, 2017 .801
2016 Flores .788 OPS, 2017 .795
2015 Granderson .821 OPS, 2016 .799
2015 Duda .838 OPS, 2016 .714
2015 Lagares .647 OPS, 2016 .682
2015 Flores .703 OPS, 2016 .788
2015 Cuddyer .699 OPS, 2016 Did Not Play
2015 Murphy .770 OPS, 2016 .985
2015 Tejada .688 OPS, 2016 .489
2014 d’Arnaud .718 OPS, 2015 .825
At first glance this is an overwhelming refutation of the hypothesis that a big second-half performance is a good indicator of a better follow-up season. But things are never what they seem. We have guys like Nimmo, Cespedes and Duda who saw their playing time the following year drop off significantly due to injuries. Then we have guys like Walker and Cabrera who were traded the next season in mid-year. Plus we have guys like Flores and Lagares who were never able to make the jump into solid starter, regardless if they had a strong finish the previous year or a weak one.
Now let’s look at the decliners:
2018 Flores .736 OPS, 2019 .848
2018 Bautista .727 OPS, 2019 Did Not Play
2018 Plawecki .685 OPS, .2019 .629
2017 Lagares .661 OPS, 2018 .765
2016 Cespedes .884 OPS, 2017 .892
2016 Loney .703 OPS, 2017 Did Not Play
2014 Murphy .734 OPS, 2015 .770
2014 Wright .698 OPS, 2015 .814 OPS
2014 Granderson .714 OPS, 2015 .821
2014 Lagares .703 OPS, 2015 .647
Again, we don’t see much information to support the initial hypothesis.
A normal person would give up on the hypothesis. But no one ever accused me of being normal. My next piece will look at the question in a different way, with an attempt to control for quality a little bit better. The default assumption should be that we don’t find much different than what we found here. But you never know.