Can Ryan Howard Hit Lefties?

When it comes to righty/lefty matchups, my nerdy senses start tingling. I think it is one of the most fascinating subjects in baseball, mostly because it’s a cool combination of game theory, statistics, and psychology. Although Ryan Howard‘s name is in the title, this isn’t really about him (I promise my time at Fangraphs won’t be spent solely on Ryan Howard and philosophy).

As a baseball fan, I watch a lot of games. As a follower of advanced analysis, I know this can be both helpful and hurtful. When it comes to righties and lefties, I sometimes let my gut take over, and to be honest, I feel a little guilty. From watching games, my gut tells me that Ryan Howard stinks against lefties. The numbers back it up. For his career, Howard is hitting .226/.304/.441 against southpaws in 1144 plate appearnces. But a few months ago, the incredibly wise Mitchel Lichtman (MGL) wrote the following:

IOW, how a batter does against RH pitchers informs us on how he will likely do against LH pitchers and vice versa. Why? Because there is not much of a spread in true platoon splits among ML baseball players yet there is a large spread in overall true hitting talent among ML baseball players. So if we see a large platoon split, like for a player like Howard, it is likely a fluke. If a player does really well versus RH pitchers but terrible against LH pitchers, both the “really well” and the “terrible” numbers are likely fluky and the “truth” is somewhere in between…

…First we’ll estimate his overall true OPS. In his career, it is .966. We’ll regress that and call it .930, which is a typical projection for him. Now we have to take his observed platoon ratio (I like to use ratio – some people use a differential) and regress that. His observed platoon ratio for those 4 years is 1.052/.719, or 1.46. For the average lefty, it is 1.20 and we just don’t see that much variation among players in their true platoon splits. IOW, that 1.46 is likely very (but not completely) flukey. We might regress that 1.46 80% toward the league average of 1.20, to get 1.25. That is Howard’s “true” estimated platoon split.

Now we simply apply that to his overall estimated true OPS of .930 and the fact that he faced 62% RHP. That gives us an OPS of .805 versus LHP and 1.006 versus RHP.

Matt Swartz of Baseball Prospectus also looked at the numbers before the start of the 2009 season and concluded, “…a mixture of an inability to avoid comparing Howard’s skills versus lefties from his skills versus righties, and unwillingness to actually look up the numbers has led the sabermetric community to be just as inaccurate and groupthinking as the mainstream media.”

Maybe I should just stop right there when two really smart guys like Litchman and Swartz are agreeing on an issue. However, my main area of contention is the argument that because Howard is good at hitting righties, it shows he is a good hitter and that his performance versus lefties is a “fluke.” I understand the argument, but what about the possibility that Howard is just bad against lefties and very, very good against righties? From watching games, this seems to be the case. This is what my “baseball mind” is telling me. The issues with resting on that gut feeling are obvious and many, but that doesn’t mean the conclusion is necessarily wrong.

I guess I should also clarify what I mean by “bad.” He’s not 2010 Aramis Ramirez bad against lefties, just more like 2007 Jason Bay bad. His immense success against righties also makes things relative.

Below are Howard’s wRC+ from 2005-2010, going from overall to versus lefties and then versus righties:

2005: 135, 4, 169
2006: 166, 133, 182
2007: 140, 110, 159
2008: 123, 91, 143
2009: 141, 71, 178
2010: 109, 75, 129

Using the splits section, we can also see that the quality of balls hit by Howard against lefties is much worse than against righties (fewer line drives, more infield flies, etc). Am I saying that Howard should be sat versus lefties? No. I also think that MGL is right, that Howard is more likely to be better than his career OPS versus lefties than worse. The question is how much. Howard’s career walk rate against lefties is 9.4%. This year it’s 3.6%. As we discussed the other day, Howard is having a rough year. The question I want to pose to everyone is, how long do we need to wait until we can “tell” that someone is just not very good against lefties? Is this an instance where we need our gut to take over a bit because the sample size needed for this platoon is big enough to take many years to get to? You tell me, because I still haven’t figured this one out.




Print This Post



Pat Andriola is an Analyst at Bloomberg Sports who formerly worked in Major League Baseball's Labor Relations Department. You can contact him at Patrick.Andriola@tufts.edu or follow him on Twitter @tuftspat

16 Responses to “Can Ryan Howard Hit Lefties?”

You can follow any responses to this entry through the RSS 2.0 feed.
  1. Tommy Bennett says:

    “…but what about the possibility that Howard is just bad against lefties and very, very good against righties? From watching games, this seems to be the case.”

    It is certainly a possibility. But what MGL was trying to do is create an estimate–and remember estimates always carry a certain degree of error–of what Howard’s true split is. That is to say, we have relatively little data about how Howard hits lefties but we have a TON of data about how Howard hits all pitchers. If we are going to try to figure out how good he is against lefties, then, we ought to use all available information, including what we know about other baseball players. And one of the things we know about other baseball players is that there really aren’t very many who have a career platoon split as big as Howard’s. That leaves us with two possibilities (which may be combined to various degrees): 1) Howard’s observed split is a bit of a fluke, or 2) he’s a unique player with a unique failing.

    That doesn’t mean #2 is impossible, I just think it is less likely than #1.

    Vote -1 Vote +1

    • Pat Andriola says:

      Tommy, I completely agree, and I don’t mean to “question” MGL’s findings as much as to discuss whether people who watch the game find this to be accurate, or maybe it leans a little more to #2 than we think.

      I guess an example would be if I were in the major leagues. I’d embarass myself, but even if I went 0-5 with 5 strikeouts, it really wouldn’t be enough of a sample size to definitively declare that I was terrible and would get out almost every time. But from watching you’d know that.

      At the least I find it an interesting thought.

      Vote -1 Vote +1

      • deadpool says:

        I wonder that maybe the problem with guys like Howard is that platoon splits are really an artifact of an actual skill gap that simply coinsides with the platoon splits.

        Since I worded that horribly, what I meant to say was that over his career Ryan Howard has been pretty weak against sliders. Only an extreme spike in 2009 keeps him above average in weighted runs against sliders, and that came at the expense of effectiveness against every other pitch. What if Howard’s weakness isn’t really that he can’t hit lefties, but that he can’t hit sliders?

        So far this season he’s hitting .188 against sliders, and only fractionally below that against lefty sliders. He’s probably seeing more sliders from lefties than righties, because he’s so good with inside pitches they’d be afraid to lay one in to him. So its possible that his weakness against lefties is really just an artifact of his weakness against sliders in general.

        Vote -1 Vote +1

  2. Adam M says:

    Just because two very smart people agree doesn’t make it true. And I think 1144 plate appearances against lefties is a sufficient sample size to draw some conclusions.

    Nevertheless, I understand where they’re coming from, and Tommy Bennett put it better than I could that when trying to project the future, it is wiser to try and use more information than less; we should use information from other hitters, especially other hitters who statistically resemble Ryan Howard.

    But once you get down to the individual level, there are flaws to that. As Tommy points out, he could be a true outlier, which means calling it a ‘fluke’ is statistically ‘accurate’ by potentially practically fallible. I think, in cases of analysis like this, judgement and reason can help us see past the statistics and step out of paradigms (both traditional and new) and not be stuck in them because, well, really smart people are wrong too (though just less often than some of the rest of us). Stick to your guns Pat; I think you’re right.

    Vote -1 Vote +1

  3. don says:

    Jim Thome has hit .294/.429/.613 in almost 7000 PA vs. RHP and .238/.340/.421 in almost 3000 PA vs. LHP. It’s not quite as extreme as the Howard platoon split, but it’s close. If the ‘true’ platoon split were 1.20, what would be the odds of an outlier like that?

    Vote -1 Vote +1

  4. bballer319 says:

    I think Deadpool is onto something. From watching dozens of games, Howard struggles with the slider (breaking balls) from lefties.

    Is there a way to see how he handles each pitch, separated by lefty/righty? I think that will paint the clearest picture.

    Vote -1 Vote +1

    • deadpool says:

      Yeah, ESPN inside edge. They only have batting average against each pitch, but he’s .188 overall against sliders, .153 against lefties and .231 against righties. The other problem is their some sample size issues because only current season data is available.

      Just skimming some numbers and it seems like there’s a trend that players that hit over .230 or so against sliders actually don’t hit sliders worse (and in some cases better) against oppossite handed pitching, which while counterintuitive makes an interesting point that being able to hit certain pitches translates relatively well between the two, but a general weakness to a pitch may be made worse by the handedness of the pitcher.

      Vote -1 Vote +1

      • deadpool says:

        Found some better data. He’s seen 122 sliders from lefties this season 50% have been for strikes and he swung at 38% of them, he’s put 12% in play and whiffed on 16%. so he’s had 61 sliders he should have swung at, he’s swung and missed at 20, and of the 122 he’s only put 15 in play. So he’s only putting 26% of the sliders on the plate in play (give or take for swings outside the zone).

        From righties he’s only seen 119. 70% were for strikes, but he’s put 19% in play and the whiff rate drops to 12%. So 83 he should have swung at, he did swing at 64. and he put 22 in play. That’s 27% folks. He swings and misses more often, but he puts the ball in play more or less the same against sliders from both. He just sees more from lefties. (he’s seen 3 more in dramatically fewer at bats).

        Vote -1 Vote +1

  5. intricatenick says:

    So regress 80% to get the “true” split? Any reason for that? ;)

    It would be best to regress the “causative variable”, but all statisticians can prove are correlative variables. The con is getting people to agree that the correlative variables you find are causative in nature.

    Also, the word fluke is not really defined in a vacuum. I could call all the stats from the 1980s and 1990s a fluke when measured to the whole of baseball history. But it wouldn’t make sense to look at say, Will Clark’s career and call every single one of his seasons a fluke relative to every other one.

    Imagine an alien race playing baseball for 100,000 years with the same rules. The entire history of baseball as played on Earth would be “flukey” and the sample sizes would be way too small here on Earth to compare the two types of baseball being played to one another.

    One man’s “flukey” stats are anothers “predictive indicators”. Prediction is a rough business, but I think at some point everyone needs to realize that the best is measured by what predicts the most things right (on some sort of agreed temporal weighted basis – more points for things further off) the most number of times.

    If your model has a lot of “flukey” things that happen that you didn’t predict, then you might need to go back to the drawing board. Every time anyone chalks something up to luck they need to look at themselves in the mirror and realize that this is a situation where you are way less smart than you thought you were. Luck is nature’s way of telling us that we are stupid.

    Vote -1 Vote +1

  6. circlechange says:

    Howard has seen 1/3 of his PAs against lefties. That’s not an insignificant amount.

    Regarding deadpool’s comment, breaking pitches from same-handed pitching are one of the reasons platoon splits exist. Batters don’t hit balls that break away as well, and because of this pitchers throw more breaking pitches to same-handed batters than to opposite handed batters.

    Vote -1 Vote +1

    • deadpool says:

      And I would agree with that, that’s why MGL came up with his 1.2 split. My argument is that for a guy with a weakness to breaking balls in general his splits will be much more severe than the splits of a guy who’s big weakness is to the changeup, because same side hitters tend to “take away” the changeup. Those people shouldn’t be termed flukes, they’re true outlier’s and the numbers are representative of an actual a skill level. If Howard were hitting above .200 against sliders for righties I might sing a different tune, but he just can’t seem to hit them in general, the fact that he sees more from lefties isn’t the sole cause of the split (because its always going to be there) but it gives a more legitimate explination than saying that Howard’s entire career splits so far have been a “fluke”.

      Vote -1 Vote +1

      • circlechange says:

        Okay, then I just misunderstood what you were saying. I completely agree with you that I can see no logical reason to call Howard’s performance against lefties a fluke.

        Vote -1 Vote +1

  7. MattSullivan says:

    I don’t think the two view points are as contradictory as they seem. Howard is much better against righties but that doesn’t mean that the splits aren’t flukey. Trying to discover “true talent levels” is important for projection, where the goal is to represent future results from past ones. Doing so is full of regression techniques that provide (hopefully) the closest matches to the newly created actual data. That means distrusting wide swings in the expected numbers, but that doesn’t mean that the numbers aren’t showing the truth to some degree. It is in fact, just a question of degree.
    To make good projections, you have to regress this type of gap as much as is sensible, that doesn’t mean that the gap doesn’t exist, just that you feel low confidence in it based on the total data. Should it expand even more, you would likely regress less, but you would still not abandon all skepticism. Howard can be “bad” against lefties in comparison to his hitting against righties, but his overall skills suggest he is a good enough hitter to manage better than he has. 80% regression seems to large to me, but I’ll trust MGL to know those things better than I.
    The issue of the slider is more compelling, because it is an issue consistent with his overall skills.

    Vote -1 Vote +1

    • Rich says:

      “To make good projections, you have to regress this type of gap as much as is sensible, that doesn’t mean that the gap doesn’t exist, just that you feel low confidence in it based on the total data.”

      This should say :

      “To make good projections FOR THE LEAGUE AS A WHOLE”

      A model that matches the league closely isn’t necessarily a model that matches a specific player well, or is even logical for a specific player.

      “but his overall skills suggest he is a good enough hitter to manage better than he has.”
      Why? Neither one of the people quoted above provide any evidence that this is true, or that hitting RHP affects hitting LHP. It is an assumption, nothing more.

      Vote -1 Vote +1

  8. MGL says:

    You can certainly use other information other than these statistics – the player’s overall production, his production versus RHP and LHP, all LHB platoon ratios, and the spread of platoon talent among all LHB – to inform yourself of the answer to the question:

    “What is Howard’s (or any other player’s) true platoon ratio?”

    We do that all the time with other statistical questions. “What is Hitter A’s most likely overall projection?” We can use his height, weight, age, original draft status, etc.

    “What is Pitcher B’s most likely projection?” How fast is his fastball? What was his draft status? What are his other pitches and are they considered bad, good, average, etc.

    If you are trying to determine the true platoon ratio of a pitcher, other than the numbers, what kinds of pitches does he throw (fastballs and sliders have the most platoon differential, change-ups the least, and curves somewhere in between)? What is his arm angle?

    If what you mean by “gut” for Howard are some of those things that we can use to inform us, then by all means use them. For example, does he visibly seem to “bail out” against lefties or does he hang in there? Does he seem to not be able to lay off curves and sliders outside, more so than the average lefty? Is there other useful information you can glean from the pitch f/x data or from batted ball data in general?

    But, be careful about using the general, “Well he looks awful against lefties and great against righties” as your “data” to inform the answer to the question. If a player happens to have a flukey (or somewhat flukey) performance that we are correct to regress, of course he is going to look good or terrible when you watch him. That is what naturally tends to happen when a player has a flukey performance! He looks great or terrible. For example, we watch a hitter go 1 for 13 and rather than regress the appropriate amount (99% probably) toward league average, we say to ourselves, “Well, I watched him for all 13 AB, and he looked terrible so I am not going to regress him 99%.” Well of course most players are going to look terrible when they go 1 for 13 (not all players or all the time of course).

    That being said, if you watch Player A go 1 for 13 you may be able to see things that enable you to only regress that 1 for 13 97% (or even 90% or if it were you or I in the batter’s box, probably 5%), and in other cases, maybe 99.9%. You just have to be careful about making inferences that are a reflection of the anomalous performance itself, if that makes any sense.

    Vote -1 Vote +1

  9. Rich says:

    Lichman’s argument is terrible.

    It is “Because your average hitter doesn’t have as large a platoon split as howard, howard can’t have that large of a platoon split.”

    It is utter garbage.

    We’re talking 1200 PAs now(against lefties, righties 2200) . He is as much a .750 OPS against lefties as he is a 1.050 OPS against righties.

    Vote -1 Vote +1

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Current day month ye@r *