Home And Road Exit Velocity

This topic is an iffy area, and I want to say that up front. It is complicated and controversial.  I don’t have all of the facts, and I don’t have access to the best available data.  I’m just going to share, as best I can, the situation as I currently understand it. All of the numbers taken here are from Baseball Savant.  I’ve done nothing to manipulate the numbers in any way.  I say this because, as some of you may know, I generally run an algorithm to clean up and manipulate numbers from Baseball Savant prior to using them for analysis. Today is not one of those days, I’m just going with the raw information.

Trackman Radar, as I understand it, specializes in measuring velocity.  That is its bread and butter, and other metrics measured by Trackman are added on top of, or derived from, this most basic starting point. By all accounts, Trackman is very good radar, and these velocity measurements are very reliable.

Velocity Fluctuations

That said, there are a number of pretty large home/road exit velocity fluctuations worth pointing out.  I’m not exactly sure what may cause these differences. It could be quality of pitching or batting. If your team has great pitchers, you’d expect them to give up fewer hard hit balls, so maybe your home park exit velocity numbers would be suppressed.

Your batters could provide a similar input.  A good offensive team may have overall higher exit velocity contributions, or a bad offense may have below average. I stress *may*. A good offense isn’t necessarily defined by high exit velocity.  Look at guys like Mike Trout and Joey Votto.  Neither of these guys produce particularly high exit velocities, but both are great hitters. Likewise, Manny Machado has had elite exit velocity this season, but with pedestrian numbers.

These things are complicated, there are a lot of moving parts. A great pitching staff and a great offense may cancel each other out, for example. However, we need to talk about these home/away exit velocity numbers.

 

This is a big table, I’m sorry about that.  Especially for mobile users. These are exit velocities generated by the team’s offense in home games versus away games. I’ve highlighted the seasons where road exit velocity is greater than home exit velocity for a given team.

First to note, Arizona. A few weeks ago I wrote an article addressing the addition of the Humidor to Chase Field, and in doing so I addressed a physics based model that claims the low air density and humidity in Arizona should net a roughly 4mph increase in exit velocity in Chase Field when compared to road games. However, in 2015 and 2016 we see a difference of 1.2 and 1.3 respectively. This season, the difference is up to 2.5, much closer to what we would expect.

Arizona is a good standard candle for this sort of analysis, since we expect to see a certain difference between home and away numbers. Unfortunately, that difference is unrealized. Which is unsatisfactory. There are possible explanations.

1. The physics based model is wrong.
2. Unaccounted for variables.
3. Exit velocity numbers are wrong.
4. Small sample size problem, and the differences haven’t had enough time to emerge.

Personally, I feel number 4 is the least likely possibility. I find it hard to believe that an expected difference this large would go unnoticed to this degree after two and a half seasons. Exit velocity seems to sort itself out after a few hundred batted balls, so I find it unlikely that this is due to some statistical anomaly.

The other three options each have their own roadblocks and challenges. The physics model could be wrong, but it has withstood testing over time and it has made accurate predictions in the past. Trackman appears to be very good at measuring velocity numbers, and it has also withstood its own testing. Unaccounted for variables may be the most alluring, if you were to believe both the physics model and velocity numbers, but what could these factors be? The batters eye? Is the pitching somehow superior in Arizona? It is hard to justify.

One thing for sure, though, our standard candle is throwing up a smoke sign. It is warning us about something, but we don’t know exactly what it might be.

Ballpark Bias?

It could be, perhaps, that there are some sort of velocity measuring bias in each stadium. This is very controversial, because there is little evidence to support case. And if it were true, what mechanisms would drive the bias?

In the past, people have pointed towards slight differences in the location of the radar within the stadium. This explanation seems plausible, but there isn’t much evidence to support the conclusion. Think of it this way, if there is a certain location within each field that is ideal for measuring batted balls, don’t you think the engineers installing the radar would do their best to place the radar within that ideal region of the park?

Perhaps your stadium cannot place the radar in the ideal location for whatever reason, and the engineers would have to push towards the extreme edges of the acceptable locations within the ballpark. Maybe it should be in a given location give or take 30 feet, so you place it 25 feet away. That 25 feet is within your 30 foot acceptable radius, but it is towards the extreme end.  In this case, if you knew your radar weren’t ideal, wouldn’t you keep an eye on the numbers over time, and try to correct issues that arise?

This sort of inherent stadium bias seems a bit farfetched, but it may also be the best explanation for the observed differences to date. In my analysis of Statcast data, I have included a stadium based velocity correction for each batted ball, and recently I have gone through and updated these definitions.

 

You may notice that the Diamondbacks, Padres and Astros have the largest adjustments. I’ve already addressed the Diamondbacks as the standard candle here, and after this adjustment their home/away difference increases to 3.87, roughly around where we expected to see it. The Astros and Padres are a little bit more difficult to explain. I don’t know why they require such a large adjustment to equalize their home and road performance.

Again, I want to stress that I am not fully convinced these sorts of adjustments are truly necessary. Or correct. I am sure there are people who can devise far more rigorous examinations of the data and come to better conclusions and probably better adjustments.

My goal here isn’t to provide some be all, end all explanation and solution. Far from it, the more I think about this problem, the more confused I feel.  But, as I have read articles about the increasing home run rates, I haven’t seen very much addressing this particular issue.

Um, Why Is This In Fantasy?

I am sure you may be wondering why I’ve written this in the fantasy section of the website. There are two primary reasons. First, I feel fantasy players stand to disproportionately benefit from an understanding of the Statcast system. Second, I know many of you are using Statcast derived stats on a regular basis.

Here’s an example for how these velocity adjustments could be helpful. I have run the exact same algorithm for predicting home runs two times. First, the it was run on the raw data, the second time it was run on the velocity adjusted data. Check out the results below.

 

The Root Mean Square error for individual batters, as opposed to teams, is 1.34 prior to the adjustment, and 1.09 after. The velocity adjustments help these algorithms create a more descriptive stat. Whether this may be more predictive is another story, and I do not have an answer for you.

Of course, this runs into the issue of overfitting data. That is a problem we’re going to wrestle with for a long time now that Statcast has become a mainstream tool in baseball. There are two camps that I know of regarding this problem.

First, there are those people who firmly believe Statcast should be used to describe what has happened in games to the best of its ability. All efforts should be spent increasing the accuracy of the system, and efforts to normalize data could wash out real changes and differences throughout the course of a game, season, or career.

Second, there are those who wish to normalize the data across the league and focus on the predictive nature of these stats.

None of This Makes Me Comfortable

I feel like I am hopelessly lost between the two extremes, and maybe that is the worst place to be. Maybe it would be better to be more firmly in one camp over the other. The first camp will adamantly poo-poo all of these velocity adjustments, and I can already hear them yelling at me. I want to say to those people:  I know, I agree, this makes me uncomfortable, too.  But, at the same time, our standard candle in Arizona tells me something needs to be adjusted. And when I make the adjustment, my expected stat algorithms are more descriptive.

But, at the same time, these changes make my skin crawl. I know they aren’t firmly rooted in a logical argument. It feels more like a ‘just make it work’ jerry rig. The academic rigor is stripped away and replaced with a hunch.

Of course, this hunch isn’t new.  Baseball Prospectus wrote about this last year, and they released their own park adjustments. Their adjustments are much more rigorous than mine, but they are also out of date and obsolete. They were calculated using only the first few months of the 2016 season, and a lot has changed since then. The summer of 2016, for one. We know the summer heat makes the ball fly.

At the end of the day, this is meant to be a warning. There is a problem here, and I don’t know the solution. I don’t think anyone knows the solution, and I would hesitate to trust any claims otherwise. Just keep it in the back of your mind, every time you quote an exit velocity, or an exit velocity derived stat like xwOBA, xOBA, Barrels, VH%, xBABIP, or whatever else. Ask yourself, what role could these park factors play in the result? As you can see in the home run totals I showed before, very small changes in exit velocity can create rather large changes in team and player totals. These are just home runs, too. Singles, doubles and triples are impacted as well. Batting average, slugging, everything.