Don’t blame Pedro

Lost in the gnashing of teeth and rending of garments that followed
the Mets’ distastrous 4-13 collapse to finish the 2007 season was the
return of Pedro Martinez. As you know, Pedro missed most of the year
after undergoing shoulder surgery last off-season, and in fact did not
pitch until Sept. 3, when he faced the Reds in Cincinnati. Pedro
was decent that day, but not as good as the next day’s newspaper
accounts would have you believe. He pitched five innings, giving up
two earned runs and getting the win in the Mets’ 10-4 victory.
But he also allowed eight baserunners (including three walks).
He did
record his 3,000th career strikeout in that game, so the mood afterwards was definitely upbeat. Pedro was
back.

But, how was he throwing? Was the shoulder back to full
strength? How hard was he throwing? I had heard reports from
his minor league rehab starts that Pedro was throwing in the high
eighties, 87-88 mph. My Pedro-fan friends (not that I’m not a Pedro fan
myself, because I am) thought that was fine for rehab starts, that
Pedro would certainly build up arm strength and throw harder once he
was back. So, I was very curious about how much speed Pedro could
muster on his return.

I managed to catch some of that first start on the tube and the radar
gun generally had Pedro’s fastball right around 87 or 88 mph. Oh boy, I
thought, this is not good. Not that you can’t succeed with an 87-mph
fastball, you can. But it’s hard to be Pedro Martinez when you
throw as hard as, oh, Josh Fogg, maybe.

But you know what? When you look at the bottom line, Pedro did a
reasonable imitation of himself in his five starts this season. No, we can’t blame
Pedro for the Mets’ woeful finish. His 2007 numbers:

                 GS W  L  IP  H   R   ER  HR  BB  SO  ERA  WHIP
Martinez, Pedro  5  3  1  28  33  11  8   0   7   32  2.57 1.43

This is a small sample size, of course, but that does not mean there
aren’t some nuggets of truth to be mined from it. The first thing that
jumped out at me was the 32 strikeouts in 28 innings
pitched. That comes out to 10.3 strikeouts per nine innings. Now I know he
only did it for a few games, but that strikeout rate was better than any
other ERA qualifier in the NL this year. And he’s doing it with an
87-mph heater. Curious.

His control looks fine: the seven walks work out to 2.1 per nine innings,
which is typical Pedro command. What sticks out as un-Pedro-like in the above
line are the hits and home runs allowed. If we translate those
into rates and compare to Pedro’s career numbers, here’s what we get:

Martinez, Pedro 
                    H/9IP    HR/9IP
2007                10.6     0.00
Career               6.9     0.72

In other words, in 2007 his hit rate was way up and his home run rate
was way down. Both of these measures, hit and home run rates, are
particularly susceptible to large fluctuations when the sample size is
small, so it would probably be best to take the 2007 numbers with a grain
of salt. Pedro’s FIP, which estimates what his ERA would have been given an average batting average
on balls in play (BABIP), was only 1.84.

However, FIP doesn’t account
for the (likely) downward fluctuation in home run rate, so we can turn
to xFIP, which does. This latter stat takes FIP a step further by assuming that a pitcher’s home
run rate depends only on his fly ball rate, an assumption that
that is controversial. Nevertheless, xFIP may give us the best
estimate for Pedro’s “true talent” level in 2007, due to the small
sample size. Pedro’s xFIP in 2007 was 3.76, roughly a run worse than
his actual ERA.

Ok, that’s what he did, how did he do it? We have lots of detailed information
on Pedro’s pitches, thanks to the pitch-f/x system. I already
mentioned that Pedro was throwing his fastball around 87 mph, but
let’s have a deeper look.

On the right you see what I call the speed/movement plot—each point
represents a pitch thrown by Pedro Martinez in 2007. The vertical and
horizontal positions of each point, show, from the catcher’s
viewpoint, the vertical and horizontal movement on the pitch. The
color shows you the speed. Have a look at my Pitch
Identification Tutorial for lots more info on these kinds of
plots.

The first thing to note about the plot is the absence of red points,
i.e., pitches thrown above 90 mph. Well, perhaps you can’t tell from
the plot, since the points tend to pile up on top of each other, but
of the 414 pitches recorded by the pitch-f/x system, only one broke the 90-mph
barrier.

The curveballs are easily spotted—they are the cluster of blue
points in the lower right part of the plot. At first glance, the rest
of the points appear to be a chaotic jumble, but if you look closely
you can see a cluster of light blue points toward the left side of the
plot: those are change-ups, thrown around 75 mph.

It’s a bit hard to tell what’s going on with the rest of the pitches,
so let’s separate out the different speeds.

As you can see on the graphic on the right, I’m now plotting the movements in a
separate plot (let’s call them panels) for each speed group.

Now you can clearly see the
fastballs in the first two panels, although in the second panel
(orange points), you see a bigger blob: some pitches are drifting over
to the right (cutters?) and others downward (2-seamers?). In the third
panel, the 80-84 mph pitches, we see mostly pitches with moderate
vertical movement and horizontal movement ranging between minus five
and five. These are some sort of cutters/sliders, although it’s hard to give
these pitches a definitive label, since they really aren’t very distinct.

In the 76-80 mph group the change-up starts to show up (clump on the
left), plus there are still some sliders at this speed. The last two
panels show the rest of the change-ups, plus the curveballs.

A Hardball Times Update

by Rachael McDaniel

Goodbye for now.

We can use something called a clustering algorithm to classify each of
Pedro’s pitches. A clustering algorithm is just a method for taking a
bunch of data and grouping all the data points into distinct groups or
clusters. In our case, each resulting cluster is a different type of
pitch. The particular algorithm that I have used (kmeans, for you
statistical types) requires that you tell it how many clusters to
form.

So how many different pitches does Pedro throw? Well, after various
trials, I believe the answer is six. The plot on the right shows the
results when asking for six pitches. Of course, the clustering puts the
pitches into groups, but it doesn’t tell you which group corresponds
to which pitch. The legend on the plot shows my attempts to classify
the pitches—here’s a handy table that you can refer back to,
if need be.

FB - fastball
sFB - sinking fastball (2-seamer)
cFB - cut fastball
SL - slider
CU - change-up
CB - curveball

Here are the average movement and speed measures for the six
different pitches, along with the number thrown (and percentage of total thrown) of each type.

 Pitch Horiz  Vert  Speed  NP  Pct
    FB  -8.3   9.6   86.8 102  .25
   sFB -10.7   5.8   85.8 106  .26
   cFB  -3.5   7.1   84.4  59  .14
    SL   1.9   5.2   79.9  61  .15
    CU -10.7   3.1   76.7  56  .14
    CB   7.6  -5.4   70.9  30  .07

So, while you will hear that Pedro can “reach 90 mph” with the
fastball (technically true&mdash he did it once), the average speed
of his fastest pitch is actually a shade below 87 mph. In any case, he
used his fastball (including the sinking variety) about half the time, threw the
cutter, slider and change-up around 15% of the time each, and mixed in a curveball
only rarely.

It would be interesting to break down Pedro’s pitch selection
according to count, opposing batter and a number of other
factors. Another thing to look at would be how batters fared against
the different pitch types. But with only around 400 total pitches, we
can’t really slice and dice the data as much as we’d like, not yet anyway. One split
we can look at, though, is batter-handedness. The following table
shows how Pedro varied his pitch selection according to
batter-handedness.

 Pitch   RHB  LHB
    FB   .28  .20
   sFB   .30  .19
   cFB   .11  .18
    SL   .16  .13
    CU   .05  .26
    CB   .09  .04

When facing righties, Pedro throws predominantly fastballs (2- and
4-seamers), with few curveballs and even fewer change-ups. When lefty
swingers are at the plate, though, Pedro relies heavily on the
change-up, in fact, he throws it more than any other pitch. He also goes to
the cutter more often against lefties.

But, what I really want to know about is the strikeouts. How did Pedro
manage to strike out more than a batter per inning, with a very
average, at least in terms of speed, fastball? Well, let’s have a look.

The following table shows each of the 24 strikeout pitches (that were
captured by pitch-f/x) thrown by Pedro in 2007.

Pedro Martinez Strikeout Pitches
+-----------------+------+------+-------+--------+-------+
| Batter          | Bats | Type | Speed | Strike | Count |
+-----------------+------+------+-------+--------+-------+
| Cairo_Miguel    | R    | FB   |    89 | S      | 2-2   |
| Ross_Cody       | R    | FB   |    88 | C      | 3-2   |
| Pineiro_Joel    | R    | FB   |    88 | C      | 0-2   |
| Werth_Jayson    | R    | FB   |    88 | S      | 3-2   |
| Werth_Jayson    | R    | sFB  |    88 | C      | 0-2   |
| Harang_Aaron    | R    | cFB  |    88 | S      | 0-2   |
| Uggla_Dan       | R    | FB   |    87 | S      | 3-2   |
| Pineiro_Joel    | R    | FB   |    87 | S      | 0-2   |
| Pineiro_Joel    | R    | FB   |    87 | C      | 0-2   |
| Burrell_Pat     | R    | FB   |    87 | C      | 1-2   |
| Harang_Aaron    | R    | cFB  |    87 | S      | 1-2   |
| Dobbs_Greg      | L    | SL   |    80 | S      | 1-2   |
| Lee_Carlos      | R    | CU   |    79 | S      | 0-2   |
| Dobbs_Greg      | L    | CU   |    79 | S      | 3-2   |
| Berkman_Lance   | L    | CU   |    78 | S      | 2-2   |
| Biggio_Craig    | R    | CU   |    78 | S      | 1-2   |
| Schumaker_Skip  | L    | CU   |    78 | S      | 1-2   |
| Rowand_Aaron    | R    | CU   |    78 | S      | 2-2   |
| Rollins_Jimmy   | L    | CU   |    77 | S      | 2-2   |
| Schumaker_Skip  | L    | CU   |    76 | S      | 2-2   |
| Olivo_Miguel    | R    | CB   |    73 | C      | 0-2   |
| Ludwick_Ryan    | R    | CB   |    72 | S      | 2-2   |
| Amezaga_Alfredo | L    | CB   |    71 | S      | 2-2   |
| Ludwick_Ryan    | R    | CB   |    69 | S      | 0-2   |
+-----------------+------+------+-------+--------+-------+
Strike - S means swinging, C means called

I could probably write a thousand words just on this table (don’t worry, I won’t!),
but let me just point out a
couple of things. Pedro struck out 17 of 63 right-handers (27%) in
this sample and seven of 43 left-handers (16%), so he’s punching out righties much
more often. Given that, you might think he’s getting the righties with
breaking balls—sliders and curves—which might be supposed to work
better against same-handed batters.

You’d be wrong, though, as you can see in the table above. Pedro is
getting most of his strikeouts against righties with the fastball. The slider,
for many pitchers a strikeout pitch, is hardly to be seen—Pedro dispatched only one batter with a slider, the lefty-swinging
Greg Dobbs.

Summary of Strikeout Pitches+------+------+----+-------+| Bats | Type | NP | Speed |+------+------+----+-------+| R    | FB   |  8 |    88 || R    | CU   |  3 |    78 || R    | CB   |  3 |    71 || R    | cFB  |  2 |    87 || R    | sFB  |  1 |    88 || L    | CU   |  5 |    78 || L    | CB   |  1 |    71 || L    | SL   |  1 |    80 |+------+------+----+-------+

Another thing to note is the relatively large number of punch outs on
change-ups. Martinez is justly known for his superior slow-ball, but I wondered if
the change-up would lose effectiveness as the fastball dropped in velocity. Apparently not, based
on this admittedly small sample. Note the the speed difference between fastball and change-up is around 10 mph,
which is pretty typical of a good major-league change-up.

Note that every single change-up strikeout was of the swinging
variety. In fact, all of the off-speed pitches that notched a strikeout were
swung at, except the curveball to Miguel Olivo.

This information is summarized in the table on the left, where I’ve
totaled the number of strikeouts achieved on the various pitch types
to left- and right-handed batters. Now you can easily see the
dominance of the fastball in striking out right handers and the
change-up against lefties.

Before wrapping things up, I wanted to take a look at the location of
Pedro’s strikeout pitches. I’ve been focusing on movement, speed and
pitch type, but location may be more important than anything in
determining how effective a pitch is.

The graphic to the right shows the location of the strikeout pitches
as they crossed the front of home plate. This is from the catcher’s
viewpoint (as always), so a right-handed batter would be standing
toward the left side of the plot. The different pitch types are
color-coded, as above. Furthermore, the shape of the plotted symbol
indicates the handedness of the batter, circles for righties,
triangles for lefties.

The main things to observe here is that Pedro kept his change-up down
and away to lefties and his fastball on the outside part of the plate
to righties. Pedro was helped out a bit by Miguel Olivo, who looked at a
hanging curveball on the inner half that went for strike three and Craig Biggio, who
swung and missed, not surprisingly, at an ankle-high change-up.

You know what’d be really interesting? Having pitch-f/x data on Pedro from his great late-nineties seasons and then
more from the years 2002-2005, a period that followed a 2001 shoulder injury (treated with rest and rehab, rather than surgery). We could see how Pedro
adjusted to the 2001 injury (he lost a few mph from his fastball then, too) and then examine his subsequent transition to the Pedro of 2007. Alas, pitch-f/x was born this season,
there is no such data from previous years. In any case, given Pedro’s post-injury performance in 2002-2005—65-25, 2.83 ERA, 820 innings—I would not be at
all surprised to see him make the necessary adjustments to stay one of the top pitchers in the National League.