Skip to content

Commit c379c14

Browse files
trevinflicktrevinflick
authored
add catcher framing chapter (#7)
1 parent b0cd5bb commit c379c14

File tree

2 files changed

+379
-5
lines changed

2 files changed

+379
-5
lines changed

‎07_catcher-framing.Rmd‎

Lines changed: 379 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -2,9 +2,384 @@
22

33
**Learning objectives:**
44

5-
- THESE ARE NICE TO HAVE BUT NOT ABSOLUTELY NECESSARY
5+
- Plot strike zone
6+
- Learn how to model called strike probability
7+
- Learn how to model catcher framing
68

7-
## SLIDE 1 {-}
9+
## Background
810

9-
- ADD SLIDES AS SECTIONS (`##`).
10-
- TRY TO KEEP THEM RELATIVELY SLIDE-LIKE; THESE ARE NOTES, NOT THE BOOK ITSELF.
11+
- "Historically, scouts and coaches insisted that certain catchers had the ability to “frame” pitches for umpires. The idea was that by holding the glove relatively still, you could trick the umpire into calling a pitch a strike even if it was technically outside of the strike zone"
12+
- "Part of the problem was that until the mid-2000s, pitch-level data was hard to come by. With the advent of PITCHf/x, more sophisticated modeling techniques became viable on these more granular data."
13+
14+
## Framing Examples
15+
16+
![Molina framing a strike](http://ia601701.us.archive.org/31/items/JoseMolinaryanDoumitGifs/MolinaFrame1.gif)
17+
18+
![Doumit framing a ball](http://ia801701.us.archive.org/31/items/JoseMolinaryanDoumitGifs/DoumitFrame1.gif)
19+
20+
![Molina framing a strike](http://ia601701.us.archive.org/31/items/JoseMolinaryanDoumitGifs/MolinaFrame3.gif)
21+
22+
![Doumit framing a ball](http://ia601701.us.archive.org/31/items/JoseMolinaryanDoumitGifs/DoumitFrame3.gif)
23+
24+
25+
## Getting the data
26+
27+
```{r, eval = FALSE}
28+
sc2022 <- here::here("data_large/statcast_rds/statcast_2022.rds") |>
29+
read_rds()
30+
sc2022 <- sc2022 |>
31+
mutate(
32+
Outcome = case_match(
33+
description,
34+
c("ball", "blocked_ball", "pitchout",
35+
"hit_by_pitch") ~ "ball",
36+
c("swinging_strike", "swinging_strike_blocked",
37+
"foul", "foul_bunt", "foul_tip",
38+
"hit_into_play", "missed_bunt" ) ~ "swing",
39+
"called_strike" ~ "called_strike"),
40+
Home = ifelse(inning_topbot == "Bot", 1, 0),
41+
Count = paste(balls, strikes, sep = "-")
42+
)
43+
```
44+
45+
```{r, eval = FALSE}
46+
taken <- sc2022 |>
47+
filter(Outcome != "swing")
48+
taken_select <- select(
49+
taken, pitch_type, release_speed,
50+
description, stand, p_throws, Outcome,
51+
plate_x, plate_z, fielder_2_1,
52+
pitcher, batter, Count, Home, zone
53+
)
54+
write_rds(
55+
taken_select,
56+
here::here("data/sc_taken_2022.rds"),
57+
compress = "xz"
58+
)
59+
```
60+
61+
## Where is the Strike Zone?
62+
63+
- Only part of the ball needs to cross home plate to be a strike
64+
- Home plate is 17 inches wide and ball's circumference is 9 inches
65+
- Outside edges of strike zone vary by plus or minus 0.947 feet
66+
- Top and bottom of strike zone varies by batter height (Midpoint between top of shoulders and top of players pants down to hollow beneath kneecap)
67+
68+
![MLB Strike Zone](https://blogs.fangraphs.com/wp-content/uploads/2024/12/Strike-Zone-Possible-Header-1.png)
69+
70+
```{r, eval = FALSE}
71+
plate_width <- 17 + 2 * (9/pi)
72+
k_zone_plot <- ggplot(
73+
NULL, aes(x = plate_x, y = plate_z)
74+
) +
75+
geom_rect(
76+
xmin = -(plate_width/2)/12,
77+
xmax = (plate_width/2)/12,
78+
ymin = 1.5,
79+
ymax = 3.6, color = crcblue, alpha = 0
80+
) +
81+
coord_equal() +
82+
scale_x_continuous(
83+
"Horizontal location (ft.)",
84+
limits = c(-2, 2)
85+
) +
86+
scale_y_continuous(
87+
"Vertical location (ft.)",
88+
limits = c(0, 5)
89+
)
90+
```
91+
92+
```{r, eval = FALSE}
93+
k_zone_plot %+%
94+
sample_n(taken, size = 2000) +
95+
aes(color = Outcome) +
96+
geom_point(alpha = 0.2) +
97+
scale_color_manual(values = crc_fc)
98+
```
99+
100+
![](https://beanumber.github.io/abdwr3e/07-framing_files/figure-html/fig-what-strikes-1.png)
101+
102+
```{r, eval=FALSE}
103+
zones <- taken |>
104+
group_by(zone) |>
105+
summarize(
106+
N = n(),
107+
right_edge = min(1.5, max(plate_x)),
108+
left_edge = max(-1.5, min(plate_x)),
109+
top_edge = min(5, quantile(plate_z, 0.95, na.rm = TRUE)),
110+
bottom_edge = max(0, quantile(plate_z, 0.05, na.rm = TRUE)),
111+
strike_pct = sum(Outcome == "called_strike") / n(),
112+
plate_x = mean(plate_x),
113+
plate_z = mean(plate_z)
114+
)
115+
```
116+
117+
```{r, eval=FALSE}
118+
library(ggrepel)
119+
k_zone_plot %+% zones +
120+
geom_rect(
121+
aes(
122+
xmax = right_edge, xmin = left_edge,
123+
ymax = top_edge, ymin = bottom_edge,
124+
fill = strike_pct, alpha = strike_pct
125+
),
126+
color = "lightgray"
127+
) +
128+
geom_text_repel(
129+
size = 3,
130+
aes(
131+
label = round(strike_pct, 2),
132+
color = strike_pct < 0.5
133+
)
134+
) +
135+
scale_fill_gradient(low = "gray70", high = crcblue) +
136+
scale_color_manual(values = crc_fc) +
137+
guides(color = FALSE, alpha = FALSE)
138+
```
139+
140+
![](https://beanumber.github.io/abdwr3e/07-framing_files/figure-html/fig-zones-1.png)
141+
142+
## Modeling Called Strike Percentage
143+
144+
- We use a Generalized Additive Model with binomial family
145+
146+
```{r, eval=FALSE}
147+
library(mgcv)
148+
strike_mod <- gam(
149+
Outcome == "called_strike" ~ s(plate_x, plate_z),
150+
family = binomial,
151+
data = taken
152+
)
153+
```
154+
155+
```{r, eval=FALSE}
156+
library(broom)
157+
hats <- strike_mod |>
158+
augment(type.predict = "response")
159+
160+
k_zone_plot %+% sample_n(hats, 10000) +
161+
geom_point(aes(color = .fitted), alpha = 0.1) +
162+
scale_color_gradient(low = "gray70", high = crcblue)
163+
```
164+
165+
![](https://beanumber.github.io/abdwr3e/07-framing_files/figure-html/fig-gam-k-zone-1.png)
166+
167+
- We can build a continuous grid
168+
169+
```{r, eval=FALSE}
170+
library(modelr)
171+
grid <- taken |>
172+
data_grid(
173+
plate_x = seq_range(plate_x, n = 100),
174+
plate_z = seq_range(plate_z, n = 100)
175+
)
176+
177+
grid_hats <- strike_mod |>
178+
augment(type.predict = "response", newdata = grid)
179+
180+
tile_plot <- k_zone_plot %+% grid_hats +
181+
geom_tile(aes(fill = .fitted), alpha = 0.7) +
182+
scale_fill_gradient(low = "gray92", high = crcblue)
183+
tile_plot
184+
```
185+
186+
![](https://beanumber.github.io/abdwr3e/07-framing_files/figure-html/fig-gam-k-zone-tile-1.png)
187+
188+
- Batter and pitcher handedness may have an effect, let's add it to our GAM
189+
190+
```{r, eval=FALSE}
191+
hand_mod <- gam(
192+
Outcome == "called_strike" ~
193+
p_throws + stand + s(plate_x, plate_z),
194+
family = binomial,
195+
data = taken
196+
)
197+
198+
hand_grid <- taken |>
199+
data_grid(
200+
plate_x = seq_range(plate_x, n = 100),
201+
plate_z = seq_range(plate_z, n = 100),
202+
p_throws,
203+
stand
204+
)
205+
hand_grid_hats <- hand_mod |>
206+
augment(type.predict = "response", newdata = hand_grid)
207+
208+
diffs <- hand_grid_hats |>
209+
group_by(plate_x, plate_z) |>
210+
summarize(
211+
N = n(),
212+
.fitted = sd(.fitted),
213+
.groups = "drop"
214+
)
215+
tile_plot %+% diffs
216+
```
217+
218+
![](https://beanumber.github.io/abdwr3e/07-framing_files/figure-html/fig-gam-k-zone-tile-diffs-1.png)
219+
220+
## Modeling Catcher Framing
221+
222+
223+
```{r, eval=FALSE}
224+
taken <- taken |>
225+
filter(
226+
is.na(plate_x) == FALSE,
227+
is.na(plate_z) == FALSE
228+
) |>
229+
mutate(
230+
strike_prob = predict(
231+
strike_mod,
232+
type = "response"
233+
)
234+
)
235+
```
236+
237+
$$\log \frac{p_j}{1 - p_j} = \beta_0 + \beta_1 \cdot strike\_prob_j + \alpha_{c(j)}$$
238+
239+
We fit a generalized linear mixed model using fixed effects from the catcher.
240+
241+
```{r, eval=FALSE}
242+
library(lme4)
243+
mod_a <- glmer(
244+
Outcome == "called_strike" ~
245+
strike_prob + (1|fielder_2_1),
246+
data = taken,
247+
family = binomial
248+
)
249+
```
250+
251+
```{r, eval=FALSE}
252+
fixed.effects(mod_a)
253+
254+
# (Intercept) strike_prob
255+
# -4.00 7.67
256+
257+
VarCorr(mod_a)
258+
259+
# Groups Name Std.Dev.
260+
# fielder_2_1 (Intercept) 0.218
261+
```
262+
263+
```{r, eval=FALSE}
264+
c_effects <- mod_a |>
265+
ranef() |>
266+
as_tibble() |>
267+
transmute(
268+
id = as.numeric(levels(grp)),
269+
effect = condval
270+
)
271+
```
272+
273+
```{r, eval=FALSE}
274+
master_id <- baseballr::chadwick_player_lu() |>
275+
mutate(
276+
mlb_name = paste(name_first, name_last),
277+
mlb_id = key_mlbam
278+
) |>
279+
select(mlb_id, mlb_name) |>
280+
filter(!is.na(mlb_id))
281+
282+
c_effects <- c_effects |>
283+
left_join(
284+
select(master_id, mlb_id, mlb_name),
285+
join_by(id == mlb_id)
286+
) |>
287+
arrange(desc(effect))
288+
289+
c_effects |> slice_head(n = 6)
290+
291+
# A tibble: 6 × 3
292+
# id effect mlb_name
293+
# <dbl> <dbl> <chr>
294+
# 1 664848 0.358 Donny Sands
295+
# 2 669004 0.294 MJ Melendez
296+
# 3 642020 0.287 Chuckie Robinson
297+
# 4 672832 0.275 Israel Pineda
298+
# 5 571912 0.260 Luke Maile
299+
# 6 575929 0.243 Willson Contreras
300+
301+
c_effects |> slice_tail(n = 6)
302+
303+
# A tibble: 6 × 3
304+
# id effect mlb_name
305+
# <dbl> <dbl> <chr>
306+
# 1 664731 -0.293 P. J. Higgins
307+
# 2 455139 -0.304 Robinson Chirinos
308+
# 3 661388 -0.336 William Contreras
309+
# 4 608360 -0.357 Chris Okey
310+
# 5 435559 -0.357 Kurt Suzuki
311+
# 6 595956 -0.390 Cam Gallagher
312+
```
313+
314+
$$\log \frac{p_j}{1 - p_j} = \beta_0 + \beta_1 strike\_prob_j + \alpha_{c(j)} + \gamma_{p(j)} + \delta_{b(j)}$$
315+
316+
We add to the model with pitcher and batter effects.
317+
318+
```{r, eval=FALSE}
319+
mod_b <- glmer(
320+
Outcome == "called_strike" ~ strike_prob +
321+
(1|fielder_2_1) +
322+
(1|batter) + (1|pitcher),
323+
data = taken,
324+
family = binomial
325+
)
326+
327+
VarCorr(mod_b)
328+
329+
# Groups Name Std.Dev.
330+
# pitcher (Intercept) 0.267
331+
# batter (Intercept) 0.251
332+
# fielder_2_1 (Intercept) 0.209
333+
```
334+
335+
336+
```{r, eval=FALSE}
337+
c_effects <- mod_b |>
338+
ranef() |>
339+
as_tibble() |>
340+
filter(grpvar == "fielder_2_1") |>
341+
transmute(
342+
id = as.numeric(as.character(grp)),
343+
effect = condval
344+
)
345+
c_effects <- c_effects |>
346+
left_join(
347+
select(master_id, mlb_id, mlb_name),
348+
join_by(id == mlb_id)
349+
) |>
350+
arrange(desc(effect))
351+
352+
c_effects |> slice_head(n = 6)
353+
354+
# A tibble: 6 × 3
355+
# id effect mlb_name
356+
# <dbl> <dbl> <chr>
357+
# 1 624431 0.313 Jose Trevino
358+
# 2 669221 0.277 Sean Murphy
359+
# 3 425877 0.263 Yadier Molina
360+
# 4 664874 0.253 Seby Zavala
361+
# 5 543309 0.229 Kyle Higashioka
362+
# 6 608700 0.221 Kevin Plawecki
363+
364+
c_effects |> slice_tail(n = 6)
365+
366+
# A tibble: 6 × 3
367+
# id effect mlb_name
368+
# <dbl> <dbl> <chr>
369+
# 1 596117 -0.277 Garrett Stubbs
370+
# 2 435559 -0.281 Kurt Suzuki
371+
# 3 521692 -0.291 Salvador Perez
372+
# 4 553869 -0.327 Elias Díaz
373+
# 5 455139 -0.336 Robinson Chirinos
374+
# 6 669004 -0.347 MJ Melendez
375+
```
376+
377+
## Further Reading
378+
379+
- Turkenkopf [(2008)](https://www.beyondtheboxscore.com/2008/4/5/389840/framing-the-debate)
380+
- Fast [(2011)](https://www.baseballprospectus.com/news/article/15093/spinning-yarn-removing-the-mask-encore-presentation/)
381+
- Lindbergh [(2013)](http://grantland.com/features/studying-art-pitch-framing-catchers-such-francisco-cervelli-chris-stewart-jose-molina-others/)
382+
- Brooks and Pavlidis [(2014)](https://www.baseballprospectus.com/news/article/22934/framing-and-blocking-pitches-a-regressed-probabilistic-model-a-new-method-for-measuring-catcher-defense/)
383+
- Brooks, Pavilidis, and Judge [(2015)](https://www.baseballprospectus.com/news/article/25514/moving-beyond-wowy-a-mixed-approach-to-measuring-catcher-framing/)
384+
- Deshpande and Wyner [(2017)](https://doi.org/10.1515/jqas-2017-0027)
385+
- Judge [(2018)](https://www.baseballprospectus.com/news/article/38289/bayesian-bagging-generate-uncertainty-intervals-catcher-framing-story/)

‎bookclub-r_baseball.Rproj‎

Lines changed: 0 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
Version: 1.0
2-
ProjectId: a276e697-8cc5-47d6-ae2d-de8ce342d9de
32

43
RestoreWorkspace: Default
54
SaveWorkspace: Default

0 commit comments

Comments
 (0)