Baseball strategy

At this point you are probably wondering why I am bothering to run this categorisation. Why don't I just run a regression? Remember, what I am trying to detect here is the presence of an elite group of pitchers, hence why I am segmenting. Technically you could say I should be comparing this group with REST of the population and not just the pitchers in the bottom right corner. If we find a difference between the extremes then lets come back to this.

So my hypothesis is that you may get an elite group of pitchers who don't give up many line drives and they reside in the top left corner of the diagram. Pitchers in this corner include: Mariano Rivera, Johan Santana, Tim Wakefield, Billy Wagner, AJ Burnett and Jose Contreras to name a few. Not a bad list. But in the other corner there were also some A-list names: Jason Isringhausen, Mark Prior and (ouch) Brad Lidge - hmm my hypothesis looks doomed!!.

Anyway, to test this what I did was to run an independent sample t-test of this data using FIP (Field Independent Pitching - developed by Tangotiger) as the test variable. FIP is a good measure of how effective a pitcher is with defence controlled. The two groups were, group 1: where the pitchers had a rank of either 1 or 2 in both 2004 & 2005 and group 2: where pitchers had a rank of ether 5 or 6 in both 2004 and 2005. Everyone else was excluded. Running the analysis it turned out that the test wasn't significant. In other words there was NO difference between the two test groups in their FIP scores therefore disproving my hypothesis.

Not a surprise I suppose given the low year-to-year correlations in line drive percentage and the observations above. But I was still curious whether people like Santana and Rivera who distinguished themselves in have a low line drive % in both 2004 and 2005 did occupy an elite group of pitching. I segment the existing groups into 4:

• Group 1: elite pitchers, ranked 1 in both 2004 and 2005
• Group 2: semi-elite, ranked 1 and 2 or 2 and 2
• Group 3: poor pitchers, ranked 5 and 6 or 5 and 5
• Group 4: worst, ranked 6 in both years

I then ran an ANOVA to compare the different samples. And no surprise the test failed - the overall mean of the data was a better fit to the data than the ANOVA model. Why am I boring you with all this? Well the one interesting thing I found was that there was a significant difference between the worst (group 4) and the rest. FIP for group 4 was almost a whole point higher. Now this is probably because the sample size was small (only 10 in group 3 vs 30 in other groups). But if this is confirmed with a larger dataset it opens the possibility that there are some pitchers who simply shouldn't be in the major leagues becuase they give up too many line drives - which as we know are expensive. I'd also like to look at line drive percentages for a couple of the elite picther like Santana to see if his line drive percentage regressed towards the mean. Given the findings above I would expect this to happen.

# posted by Beamer @ 8:03 PM 1 comments

Thanks for all you feedback about my review of batted ball data from THT. One question that came up a couple of times is do "flyball pitchers" induce more pop-ups than ground-ball pitchers. I want to tackle this question in a couple of ways. Firstly, lets look at a graph of pop-ups vs. flyballs for all pitchers who have more than 40 batted balls in play in 2005:

There is a clear relationship but the Rsq it is only 0.18 (significant at 0.01 level). This means that only 18% of the variance of flyballs is explained by this model (ie, pop-ups). (In case you were wondering including 2004 data gives a similar correlation).

Actually the correlation could be a little stronger than it first appears. Because both variables are a percentage of balls in play, if say, flyballs increase then there is less "room" for pop-ups to increase. This is why groundballs correlate invesely to fly-balls - if you don't have one you have the other (ignoring line drives).

Nothing suprising so far. Another way to look at this is to ask the contrarian question: do groundball pitchers give up fewer pop-ups than flyball pitchers. Given that flyballs and groundballs make up ~70% of batted balls we can simply categorise pitchers according to whether they give up a lot of groundballs or not. Then we can look for a difference in pop-ups in these two populations. Clear? Lets have a look at how it works in practice.

To categorise pitchers into those that give up groundballs and those that don't I'm simply going to cut the sample of pitchers in half. Those above the mean will go in the "groundball" group; those who aren't go in "other". Then we can run an independent t-test on the two populations to see if there is a significant difference in pop-ups. And, again no surprise, there is a difference and calculating an "Rsq" for this gives 0.63, which as expected is much more pronounced. We could further control for line drives but since we know that (for pitchers) they are largely a random event I have ignored them.

So, what does all this mean? We know from linear weights and the run expectancy matrix that a pop-out is worth almost the same as a strikeout. This poses a wider question. We know from my last post that inducing groundballs is very effective for the fielding team because most of them (75%) are turned into outs and those that are not are predominantly singles. But our analysis here tells us that popups, which lets not forget are as valuable as strikeouts, are the domain of flyballers. I haven't run the analysis but it would be interesting to see if groundball or flyball pitchers have a higher propensity to strikeout. Then we could use batted bull run value data to build the profile(s) of what an elite pitcher looks like.

# posted by Beamer @ 8:30 PM 0 comments

No surprise. Pitchers who gave up a lot of groundballs in 2004 did so in 2005. The Rsq is 0.5, which says that 50% of performance in 2005 is explained by performance in 2004 - which is reasonably high. That is why we refer to pitchers as groundball pitchers - Tim Hudson comes to mind. Now, where it gets more interesting is if we look at the same chart for line drives. Here it is:

As you can see the Rsq is very small: 0.01. In other words whether or not a pitcher gives up a line drive is largely luck. I bet you didn't know that (unless year read THT). Doing the same for batters shows a slightly larger Rsq (~0.1) between one year and the next for line drives. Hitters do show a small degree of skill in hitting line drives. THT Annual 2006 does this (and a lot more) for a range of different batter / pitcher events and I encourage you to have a look.

All very interesting but so what, you may ask? To really understand what is going on lets look at another article from, you've guessed it, THT Annual 2006 (no, I am not a contributor). This particular piece works out run value per batted ball above or below an average baseline. Here are some selected examples:

• Line drive: 0.356
• Outfield flyball: 0.035
• Groundball: -0.101
• Strikeout: -0.287

What this is saying is that if a batter slams a line drive then it contributes runs for the offense - the highest value event. This is because line drives only result in an out 25% of the time. And, remember, we said earlier that line drives are largely luck! Pretty amazing. Now take groundballs. Hitting a groundball is bad news. That is because it results in an out 75% of the time, and if it doesn't the chances are that you will only get to first base.

Two things jump out at me that I want to look at further. Firstly, I want to dig into line drives a little deeper. If we can find some pitchers who consistently prevent line drives more than others then they should be more valuable. I guarantee you that if I was on the mound a lot more than 20% of my pitches will go for line drives. Secondly I want to use this to develop a measuring system for pitchers / batters. Now I know this has been done (check out J'S Bradbury PrOPS metric - http://www.sabernomics.com - all his work is excellent if you have time to peruse), but I am curious if we can find a new part of the player population that has been significantly under-valued. Also I'll continue to explore the batted ball data and post anything else that I discover.

# posted by Beamer @ 7:26 PM 0 comments