-
Notifications
You must be signed in to change notification settings - Fork 0
/
BSAF23Project.R
1238 lines (739 loc) · 39 KB
/
BSAF23Project.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
## FALL 23 BSA Article
### Evaluating when to go for it on 4th down in the NFL
### STEP 1: create logistic regressions model for all 4th down tries ~2015-2021 (based on yds to go)
### STEP 2: test model on all tries last year, show success rate of model (should prolly split 70/30 instead, can show known/interesting examples)
### - sort by field position for controversial decisions
### - model will suggest ti go for short yardage on any place on the field, independent of score
### - thus we need wpa - i.e. 4th and 1 on own 15 while up 2 points and little time left
### STEP 3: introduce wpa into evaluation of decision:
### - 2 models:
### - wpa on successful conversions based on field position and yds to go (predicts wpa if converted)
### - wpa on unsuccesful conversions based on field position and yds to go (predicts wpa if not converted)
### - Predict wpa_if_converted and wpa_if_failed for all 4th downs (converted/unsuccesful/no try)
### - Create variable should_go_for_it:8
### - if wpa_if_converted > wpa_if_failed ~ TRUE
### - if wpa_if_converted < wpa_if_failed ~ FALSE
### - if wpa_if_failed == wpa_if_converted ~ NA
### - Pick certain instances (include summary of outcome) to describe in which:
### - should_go_for_it == TRUE & went for it
### - should_go_for_it == TRUE & didn't go for it
### - should_go_for_it == FALSE & went for it
### - should_go_for_it == FALSE & didn't go for it
library(dplyr, warn.conflicts = FALSE)
library(ggrepel)
library(nflreadr)
library(nflplotR)
library(stringr)
library(nflfastR)
options(scipen = 9999)
pbp_current <- load_pbp(2012:2022)
pbp_current <- pbp_current %>%
mutate(year = str_sub(game_id, 1, 4))
names(pbp_current)
## subset by: play == 1, down == 4
### Relevant Columns: game_id, ydstogo, year, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed
set.seed(93)
pbp_4th_downs <- pbp_current %>%
select(posteam, wp, field_goal_result, half_seconds_remaining, game_seconds_remaining, epa, wpa, game_id, qtr, score_differential, season_type, penalty, end_clock_time, year, ydstogo, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed, special_teams_play, down) %>%
filter(season_type == "REG", penalty != 1, play_type != "no_play", down == 4, year %in% (2012:2022))
pbp_4th_downs <- pbp_4th_downs %>%
mutate(went_for_it = play_type %in% c("pass", "run"),
fgm = "")
attempts <- pbp_4th_downs %>%
filter(play_type %in% c("pass", "run"), special_teams_play != 1, year %in% (2012:2022))
# subset attempts to meaningful ones (no mail mary's, blowouts, kneel downs, must-convert scenarios inside 5 mins, etc.) - 5+% win prob, 10 or less yards to go, 30+ seconds remaining in the half
meaningful_fourth_downs <- pbp_4th_downs %>%
filter(half_seconds_remaining >= 30,
game_seconds_remaining >= 300,
wp >= .02, ydstogo <= 30,
qtr %in% (1:4),
wp <= 0.99)
meaningful_attempts <- meaningful_fourth_downs %>%
filter(play_type %in% c("pass", "run"),
special_teams_play != 1,
year %in% (2012:2022))
### FG and PUNT
meaningful_fg <- meaningful_fourth_downs %>%
filter(play_type == "field_goal") %>%
mutate(fgm = as.numeric(field_goal_result == "made"))
meaningful_fg$fgm
fg_train_test <- meaningful_fg %>%
mutate(id = row_number())
fg_train <- fg_train_test %>% sample_frac(.70)
# Test set
fg_test <- anti_join(fg_train_test, fg_train, by = 'id')
#############
# FG LOGISTIC REGRESSION GRAPH
plot(conversion_prob ~ ydstogo, data = meaningful_fourth_downs) #
fg_regression <- glm(fgm ~ yardline_100, data = fg_train, family = binomial)
plot(fgm ~ yardline_100, data=fg_train, pch = 16,
col = rgb(red = 0, green = 0, blue = 0, alpha = 0.06),
main = "Probability of Making a Field Goal",
ylab = "Probability of FGM",
xlab = "Yard Line")
newdat <- data.frame(yardline_100 =seq(min(fg_train_test$yardline_100), max(fg_train_test$yardline_100),len=100))
newdat$fgm = predict(fg_regression, newdata=newdat, type="response")
lines(fgm ~ yardline_100, data=newdat, col="dodgerblue", lwd=2)
legend(2, 0.7, legend = c("Predicted Probability", "Field Goal Attempts"),
lty = c(1, NA),
pch = c(NA, 16),
col = c("dodgerblue", "black"))
#############
#############
# FG LOGISTIC REGRESSION SUCCESS %
fg_test_predictor <- predict(fg_regression, newdata=fg_test, type="response")
fg_test_predicted <- cbind(fg_test, fg_test_predictor)
fg_test_predicted <- fg_test_predicted %>%
mutate(pred_outcome = round(fg_test_predictor, 0)) %>%
mutate(correct_pred = (pred_outcome == fgm))
fg_test_results <- with(fg_test_predicted, table(correct_pred))
fg_succcess_rate <- fg_test_results[2]/(fg_test_results[1] + fg_test_results[2])
#############
meaningful_fourth_downs$fgm_prob <- predict(fg_regression, newdata = meaningful_fourth_downs, type = "response")
# need to predict on 2023 here - just for interesting/known scenarios
wpa_fgm <- meaningful_fg %>%
select(fgm, epa, wpa, half_seconds_remaining, game_seconds_remaining, game_id, qtr, score_differential, penalty, end_clock_time, year, ydstogo, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed, special_teams_play, down) %>%
filter(fgm == 1)
head(wpa_fgm)
wpa_fgm_train_test <- wpa_fgm %>%
mutate(id = row_number())
set.seed(93)
wpa_fgm_train <- wpa_fgm_train_test %>% sample_frac(.70)
# Test set
wpa_fgm_test <- anti_join(wpa_fgm_train_test, wpa_fgm_train, by = 'id')
# take into account score differential, results are better
# do individual models so I can graph (maybe just use the better one)
wpa_fgm_model <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining + ydstogo, data = wpa_fgm_train) # removed ydstogo(?)
wpa_fgm_model_noyds <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining, data = wpa_fgm_train) # removed ydstogo(?)
wpa_fgm_model1 <- lm(wpa ~ abs(score_differential), data = wpa_fgm_train)
wpa_fgm_model2 <- lm(wpa ~ ydstogo, data = wpa_fgm_train)
wpa_fgm_model3 <- lm(wpa ~ yardline_100, data = wpa_fgm_train)
wpa_fgm_model4 <- lm(wpa ~ game_seconds_remaining, data = wpa_fgm_train)
summary(wpa_fgm_model_noyds)
summary(wpa_fgm_model1)
summary(wpa_fgm_model2)
summary(wpa_fgm_model3)
summary(wpa_fgm_model4)
#############
# INDIVIDUAL FGM PLOTS
par( mfrow= c(2,2) , mar = c(7,4,4,2))
plot(wpa ~ ydstogo, data = wpa_fgm_train, sub = "R = 0.37",
ylim = c(-0.25, 0.25),
xlim = c(0, 30),
main = "WPA vs. Yards to Go",
ylab = "WPA",
xlab = "Yards to Go")
abline(wpa_fgm_model2, col = "dodgerblue")
plot(wpa ~ abs(score_differential), data = wpa_fgm_train, sub = "R = -0.12",
ylim = c(-0.25, 0.25),
xlim = c(0, 30),
main = "WPA vs. Score Differential",
ylab = "WPA",
xlab = "Score Differential")
abline(wpa_fgm_model1, col = "dodgerblue")
plot(wpa ~ game_seconds_remaining, data = wpa_fgm_train, sub = "R = -0.06",
ylim = c(-0.25, 0.25),
xlim = c(0, 3600),
main = "WPA vs. Seconds Remaining",
ylab = "WPA",
xlab = "Seconds Remaning")
abline(wpa_fgm_model4, col = "dodgerblue")
plot(wpa ~ yardline_100, data = wpa_fgm_train, sub = "R = 0.62",
ylim = c(-0.25, 0.25),
xlim = c(0, 45),
main = "WPA vs. Yard Line",
ylab = "WPA",
xlab = "Yard Line")
abline(wpa_fgm_model3, col = "dodgerblue")
#############
#############
# FGM WPA vs PRED PLOT
par(mfrow = c(1,1), mar = c(8, 8, 5, 5))
wpa_fgm_predict <- predict(wpa_fgm_model_noyds, newdata = wpa_fgm_test)
wpa_fgm_test$wpa_pred <- wpa_fgm_predict
sqrt(.3786)
plot(wpa ~ wpa_pred, data = wpa_fgm_test, ylim = c(-0.3, 0.3), xlim = c(-0.3, 0.3), sub = "R = 0.62",
main = "Predicted WPA vs. WPA\n(Field Goal Makes)",
ylab = "WPA",
xlab = "Predicted WPA")
wpa_fgm_accuracy <- lm(wpa ~ wpa_pred, data = wpa_fgm_test)
summary(wpa_fgm_accuracy)
abline(wpa_fgm_accuracy, col = "dodgerblue")
#############
## FG MISS
wpa_fg_miss <- meaningful_fg %>%
select(fgm, epa, wpa, half_seconds_remaining, game_seconds_remaining, game_id, qtr, score_differential, penalty, end_clock_time, year, ydstogo, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed, special_teams_play, down) %>%
filter(fgm == 0)
head(wpa_fg_miss)
wpa_fg_miss_train_test <- wpa_fg_miss %>%
mutate(id = row_number())
set.seed(93)
wpa_fg_miss_train <- wpa_fg_miss_train_test %>% sample_frac(.70)
# Test set
wpa_fg_miss_test <- anti_join(wpa_fg_miss_train_test, wpa_fg_miss_train, by = 'id')
# take into account score differential, results are better
# do individual models so I can graph (maybe just use the better one)
wpa_fg_miss_model <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining + ydstogo, data = wpa_fg_miss_train) # removed ydstogo(?)
wpa_fg_miss_model_noyds <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining, data = wpa_fg_miss_train) # removed ydstogo(?)
wpa_fg_miss_model1 <- lm(wpa ~ abs(score_differential), data = wpa_fg_miss_train)
wpa_fg_miss_model2 <- lm(wpa ~ ydstogo, data = wpa_fg_miss_train)
wpa_fg_miss_model3 <- lm(wpa ~ yardline_100, data = wpa_fg_miss_train)
wpa_fg_miss_model4 <- lm(wpa ~ game_seconds_remaining, data = wpa_fg_miss_train)
summary(wpa_fg_miss_model_noyds)
summary(wpa_fg_miss_model1)
summary(wpa_fg_miss_model2)
summary(wpa_fg_miss_model3)
summary(wpa_fg_miss_model4)
sqrt(0.4124)
#############
# INDIVIDUAL FG MISS GRAPHS
par( mfrow= c(2,2) , mar = c(7,4,4,2))
plot(wpa ~ ydstogo, data = wpa_fg_miss_train, sub = "R = 0.20",
main = "WPA vs. Yards to Go",
ylab = "WPA",
xlab = "Yards to Go",
ylim = c(-.45, .05),
xlim = c(0, 30))
abline(wpa_fg_miss_model2, col = "dodgerblue")
plot(wpa ~ abs(score_differential), data = wpa_fg_miss_train, sub = "R = 0.60",
main = "WPA vs. Score Differential",
ylab = "WPA",
xlab = "Score Differential",
ylim = c(-.45, .05),
xlim = c(0, 30))
abline(wpa_fg_miss_model1, col = "dodgerblue")
plot(wpa ~ game_seconds_remaining, data = wpa_fg_miss_train, sub = "R = -0.02",
main = "WPA vs. Seconds Remaining",
ylab = "WPA",
xlab = "Seconds Remaining",
ylim = c(-.45, .05),
xlim = c(0, 3600))
abline(wpa_fg_miss_model4, col = "dodgerblue")
plot(wpa ~ yardline_100, data = wpa_fg_miss_train, sub = "R = 0.15",
main = "WPA vs. Yard Line",
ylab = "WPA",
xlab = "Yard Line",
ylim = c(-.45, .05),
xlim = c(0, 45))
abline(wpa_fg_miss_model3, col = "dodgerblue")
#############
#############
# FG MISS WPA vs PRED PLOT
par(mfrow = c(1,1), mar = c(8, 8, 5, 5))
wpa_fg_miss_predict <- predict(wpa_fg_miss_model_noyds, newdata = wpa_fg_miss_test)
wpa_fg_miss_test$wpa_pred <- wpa_fg_miss_predict
plot(wpa ~ wpa_pred, data = wpa_fg_miss_test, ylim = c(-0.3, 0.3), xlim = c(-0.3, 0.3), sub = "R = 0.64",
main = "Predicted WPA vs. WPA\n(Field Goal Misses)",
ylab = "WPA",
xlab = "Predicted WPA")
wpa_fg_miss_accuracy <- lm(wpa ~ wpa_pred, data = wpa_fg_miss_test)
summary(wpa_fg_miss_accuracy)
sqrt(0.4046)
abline(wpa_fg_miss_accuracy, col = "dodgerblue")
#############
pbp_wpa_if_fgm <- predict(wpa_fgm_model_noyds, meaningful_fourth_downs)
meaningful_fourth_downs$wpa_if_fgm <- pbp_wpa_if_fgm
pbp_wpa_if_fg_miss <- predict(wpa_fg_miss_model_noyds, meaningful_fourth_downs)
meaningful_fourth_downs$wpa_if_fg_miss <- pbp_wpa_if_fg_miss
meaningful_fourth_downs <- meaningful_fourth_downs %>%
mutate(fg_expected_wpa = (wpa_if_fgm * fgm_prob) + (wpa_if_fg_miss * (1 - fgm_prob)))
#############
# 4th down attempted rate and success rate 2012-2022
year_succ_table <- with(meaningful_attempts, table(year, fourth_down_converted))
year_succ_table1 <- with(attempts, table(year, fourth_down_converted))
succ_perc_by_year <- data.frame("Year" = 2012:2022,
"Success Rate" = year_succ_table[,2]/(year_succ_table[,1] + year_succ_table[,2]))
succ_perc_by_year1 <- data.frame("Year" = 2012:2022,
"Success Rate" = year_succ_table1[,2]/(year_succ_table1[,1] + year_succ_table1[,2]))
year_freq_table <- with(pbp_4th_downs, table(year, went_for_it))
year_freq_table1 <- with(meaningful_fourth_downs, table(year, went_for_it))
attempt_perc_by_year <- data.frame("Year" = 2012:2022,
"Attempt Rate" = year_freq_table[,2]/(year_freq_table[,1] + year_freq_table[,2]))
attempt_perc_by_year1 <- data.frame("Year" = 2012:2022,
"Attempt Rate" = year_freq_table1[,2]/(year_freq_table1[,1] + year_freq_table1[,2]))
par(mfrow = c(1, 1), mar = c(8,8,8 ,8))
plot(Success.Rate ~ Year, data = succ_perc_by_year, ylim = c(0, 0.7),
pch = 16,
ylab = "4th Down Conversion Rate",
main = "NFL 4th Down Conversion Rate by Year\n(Meaningful Attempts)")
plot(Attempt.Rate ~ Year, data = attempt_perc_by_year, ylim = c(0, 0.25), pch = 16,
ylab = "4th Down Attempt Rate",
main = "NFL 4th Down Attempt Rate by Year\n(Meaningful Attempts)")
#############
# Train and test ydstogo model
train_test <- meaningful_attempts %>%
mutate(id = row_number())
train <- train_test %>% sample_frac(.70)
# Test set
test <- anti_join(train_test, train, by = 'id')
regression <- glm(fourth_down_converted ~ ydstogo, data = train, family = binomial)
#############
# YDSTOGO 4TH DOWN LOGSTIC MODEL GRAPH
plot(fourth_down_converted ~ ydstogo, data=train, pch = 16,
col = rgb(red = 0, green = 0, blue = 0, alpha = 0.09),
main = "Probability of Converting a 4th Down",
ylab = "Probability of Conversion",
xlab = "Yards to Go")
newdat <- data.frame(ydstogo=seq(min(train_test$ydstogo), max(train_test$ydstogo),len=100))
newdat$fourth_down_converted = predict(regression, newdata=newdat, type="response")
lines(fourth_down_converted ~ ydstogo, data=newdat, col="dodgerblue", lwd=2)
legend(15, 0.9, legend = c("Predicted Probability", "4th Down Attempts"),
lty = c(1, NA),
pch = c(NA, 16),
col = c("dodgerblue", "black"))
#############
#############
# YDSTOGO LOGISTIC REGRESSION SUCCESS %
test_predictor <- predict(regression, newdata=test, type="response")
test_predicted <- cbind(test, test_predictor)
test_predicted <- test_predicted %>%
mutate(pred_outcome = round(test_predictor, 0)) %>%
mutate(correct_pred = (pred_outcome == fourth_down_converted))
test_results <- with(test_predicted, table(correct_pred))
succcess_rate <- test_results[2]/(test_results[1] + test_results[2])
#############
meaningful_fourth_downs$conversion_prob <- predict(regression, newdata = meaningful_fourth_downs, type = "response")
### WPA Attempt Models
wpa_conv <- meaningful_attempts %>%
select(epa, wpa, half_seconds_remaining, game_seconds_remaining, game_id, qtr, score_differential, penalty, end_clock_time, year, ydstogo, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed, special_teams_play, down) %>%
filter(fourth_down_converted == 1)
head(wpa_conv)
wpa_conv_train_test <- wpa_conv %>%
mutate(id = row_number())
set.seed(93)
wpa_conv_train <- wpa_conv_train_test %>% sample_frac(.70)
# Test set
wpa_conv_test <- anti_join(wpa_conv_train_test, wpa_conv_train, by = 'id')
# take into account score differential, results are better
# do individual models so I can graph (maybe just use the better one)
wpa_conv_model <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining + ydstogo, data = wpa_conv_train) # removed ydstogo(?)
wpa_conv_model_noyds <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining, data = wpa_conv_train) # removed ydstogo(?)
wpa_conv_model1 <- lm(wpa ~ abs(score_differential), data = wpa_conv_train)
wpa_conv_model2 <- lm(wpa ~ ydstogo, data = wpa_conv_train)
wpa_conv_model3 <- lm(wpa ~ yardline_100, data = wpa_conv_train)
wpa_conv_model4 <- lm(wpa ~ game_seconds_remaining, data = wpa_conv_train)
summary(wpa_conv_model_noyds)
summary(wpa_conv_model1)
summary(wpa_conv_model2)
summary(wpa_conv_model3)
summary(wpa_conv_model4)
sqrt(0.01865)
#############
# INDIVIDUAL CONV GRAPHS
par( mfrow= c(2,2), mar = c(7,4,4,2))
plot(wpa ~ ydstogo, data = wpa_conv_train, sub = "R = 0.09",
main = "WPA vs. Yards to Go",
ylab = "WPA",
xlab = "Yards to Go",
ylim = c(-0.3, 0.3),
xlim = c(0, 20))
abline(wpa_conv_model2, col = "dodgerblue")
plot(wpa ~ abs(score_differential), data = wpa_conv_train, sub = "R = -0.52",
main = "WPA vs. Score Differential",
ylab = "WPA",
xlab = "Score Differential",
ylim = c(-0.3, 0.3),
xlim = c(0, 30))
abline(wpa_conv_model1, col = "dodgerblue")
plot(wpa ~ game_seconds_remaining, data = wpa_conv_train, sub = "R = 0.14",
main = "WPA vs. Seconds Remaining",
ylab = "WPA",
xlab = "Seconds Remaining",
ylim = c(-0.3, 0.3),
xlim = c(0, 3600))
abline(wpa_conv_model4, col = "dodgerblue")
plot(wpa ~ yardline_100, data = wpa_conv_train, sub = "R = -0.12",
main = "WPA vs. Yard Line",
ylab = "WPA",
xlab = "Yard Line",
ylim = c(-0.3, 0.3),
xlim = c(0, 85))
abline(wpa_conv_model3, col = "dodgerblue")
#############
#############
# CONV WPA vs PRED PLOT
par(mfrow = c(1,1), mar = c(8, 8, 5, 5))
wpa_conv_predict <- predict(wpa_conv_model_noyds, newdata = wpa_conv_test)
wpa_conv_test$wpa_pred <- wpa_conv_predict
wpa_conv_accuracy <- lm(wpa ~ wpa_pred, data = wpa_conv_test)
summary(wpa_conv_accuracy)
plot(wpa ~ wpa_pred, data = wpa_conv_test, ylim = c(-0.3, 0.3), xlim = c(-0.3, 0.3), sub = "R = 0.52",
main = "Predicted WPA vs. WPA\n(Fourth Down Conversions)",
ylab = "WPA",
xlab = "Predicted WPA")
abline(wpa_conv_accuracy, col = "dodgerblue")
#############
wpa_fail <- meaningful_attempts %>%
select(game_seconds_remaining, epa, wpa, game_id, qtr, score_differential, penalty, end_clock_time, year, ydstogo, yardline_100, play_type, down, fourth_down_converted, fourth_down_failed, special_teams_play, down) %>%
filter(fourth_down_failed == 1)
head(wpa_fail)
wpa_fail_train_test <- wpa_fail %>%
mutate(id = row_number())
set.seed(90)
wpa_fail_train <- wpa_fail_train_test %>% sample_frac(.70)
# Test set
wpa_fail_test <- anti_join(wpa_fail_train_test, wpa_fail_train, by = 'id')
wpa_fail_model <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining + ydstogo, data = wpa_fail_train)
wpa_fail_model_noyds <- lm(wpa ~ abs(score_differential) + yardline_100 + game_seconds_remaining, data = wpa_fail_train)# removed ydstogo(?)
wpa_fail_model1 <- lm(wpa ~ abs(score_differential), data = wpa_fail_train)
wpa_fail_model2 <- lm(wpa ~ ydstogo, data = wpa_fail_train)
wpa_fail_model3 <- lm(wpa ~ yardline_100, data = wpa_fail_train)
wpa_fail_model4 <- lm(wpa ~ game_seconds_remaining, data = wpa_fail_train)
summary(wpa_fail_model_noyds)
summary(wpa_fail_model1)
summary(wpa_fail_model2)
summary(wpa_fail_model3)
summary(wpa_fail_model4)
#############
# INDIVIDUAL FAIL GRAPHS
sqrt(0.04505)
par( mfrow= c(2,2), mar = c(7,4,4,2))
plot(wpa ~ ydstogo, data = wpa_fail_train, sub = "R = 0.36",
main = "WPA vs. Yards to Go",
ylab = "WPA",
xlab = "Yards to Go",
ylim = c(-0.3, 0.3),
xlim = c(0, 20))
abline(wpa_fail_model2, col = "dodgerblue")
plot(wpa ~ abs(score_differential), data = wpa_fail_train, sub = "R = 0.67",
main = "WPA vs. Score Differential",
ylab = "WPA",
xlab = "Score Differential",
ylim = c(-0.3, 0.3),
xlim = c(0, 30))
abline(wpa_fail_model1, col = "dodgerblue")
plot(wpa ~ game_seconds_remaining, data = wpa_fail_train, sub = "R = -0.21",
main = "WPA vs. Seconds Remaining",
ylab = "WPA",
xlab = "Seconds Remaining",
ylim = c(-0.3, 0.3),
xlim = c(0, 3600))
abline(wpa_fail_model4, col = "dodgerblue")
plot(wpa ~ yardline_100, data = wpa_fail_train, sub = "R = 0.29",
main = "WPA vs. Yard Line",
ylab = "WPA",
xlab = "Yard Line",
ylim = c(-0.3, 0.3),
xlim = c(0, 85))
abline(wpa_fail_model3, col = "dodgerblue")
#############
#############
# FAIL WPA vs PRED PLOT
par(mfrow = c(1,1), mar = c(8, 8, 5, 5))
wpa_fail_predict <- predict(wpa_fail_model_noyds, newdata = wpa_fail_test)
wpa_fail_test$wpa_pred <- wpa_fail_predict
wpa_fail_accuracy <- lm(wpa ~ wpa_pred, data = wpa_fail_test)
summary(wpa_fail_accuracy)
sqrt(0.5017)
plot(wpa ~ wpa_pred, data = wpa_fail_test, ylim = c(-0.3, 0.3), xlim = c(-0.3, 0.3), sub = "R = 0.71",
main = "Predicted WPA vs. WPA\n(Failed Fourth Down Attempts)",
ylab = "WPA",
xlab = "Predicted WPA")
abline(wpa_fail_accuracy, col = "dodgerblue")
#############
# use both models to predict on all fourth downs:
pbp_wpa_if_fail <- predict(wpa_fail_model_noyds, meaningful_fourth_downs)
meaningful_fourth_downs$wpa_if_fail <- pbp_wpa_if_fail
pbp_wpa_if_conv <- predict(wpa_conv_model_noyds, meaningful_fourth_downs)
meaningful_fourth_downs$wpa_if_conv <- pbp_wpa_if_conv
meaningful_fourth_downs <- meaningful_fourth_downs %>%
mutate(expected_wpa = (wpa_if_conv * conversion_prob) + (wpa_if_fail * (1 - conversion_prob)))
meaningful_fourth_downs <- meaningful_fourth_downs %>%
mutate(should_go_for_it1 = (expected_wpa > fg_expected_wpa) & (expected_wpa > 0),
should_kick = (fg_expected_wpa > expected_wpa) & (fg_expected_wpa > 0),
should_punt = !should_go_for_it1 & !should_kick)
meaningful_fourth_downs
# can show % of 4th downs in which they follow the advice (should have and did, shouldn't have and didn't)
# - only shows if this is how how NFL coaches think
meaningful_fourth_downs$followed_model <- rep(NA,30189)
meaningful_fourth_downs[meaningful_fourth_downs$went_for_it,]$followed_model <- meaningful_fourth_downs[meaningful_fourth_downs$went_for_it,]$should_go_for_it1
meaningful_fourth_downs[meaningful_fourth_downs$play_type == "field_goal",]$followed_model <- meaningful_fourth_downs[meaningful_fourth_downs$play_type == "field_goal",]$should_kick
meaningful_fourth_downs[meaningful_fourth_downs$play_type == "punt",]$followed_model <- meaningful_fourth_downs[meaningful_fourth_downs$play_type == "punt",]$should_punt
#############
# % MODEL FOLLOWED
meaningful_fourth_downs %>% pull(followed_model) %>% table
with(meaningful_fourth_downs, barplot(table(followed_model)/nrow(meaningful_fourth_downs),
ylim = c(0, 0.7),
col = "gray",
main = "Does the Model Agree With the Coach's 4th Down Decision?",
xlab = "Model Followed",
ylab = "Proportion"),
)
with(meaningful_fourth_downs, table(followed_model)/nrow(meaningful_fourth_downs))
#############
meaningful_fourth_downs %>% pull(play_type) %>% table
meaningful_fourth_downs %>% filter(!followed_model) %>% with(wpa <= 0) %>% table() # % of plays not following the model that have negative wpa and lower wpa than model expected
meaningful_fourth_downs %>% filter(!followed_model) %>% with(wpa >= 0) %>% table() # % of plays following the model that have positive wpa
(meaningful_fourth_downs %>% with(wpa >= 0) %>% table()) / nrow(meaningful_fourth_downs)
#############
# EFFECTIVENESS OF MODEL
(meaningful_fourth_downs %>% with(wpa >= 0) %>% table()) / nrow(meaningful_fourth_downs) # Coach's decisions are good 54.3% of the time
(10766 + 7427) / nrow(meaningful_fourth_downs) # Coach's decisions are good 62.8% of the time when following the model
(10766 / 17137) # Coach's decisions are good 62.8% of the time when following the model
5625 / 13052 # Coach's decisions are good 43.1% of the time when not following the model
#############
meaningful_fourth_downs %>% filter(play_type != "qb_kneel") %>% pull(followed_model) %>% table
# COACHES FOLLOW NO YDSTOGO MODEL MORE THAN YDSTOGO MODEL - YDSTOGO RECCOMENDS TO GO FOR IT A LOT LESS
meaningful_fourth_downs[meaningful_fourth_downs$went_for_it,]$should_go_for_it1 %>% table()
meaningful_fourth_downs[meaningful_fourth_downs$went_for_it,] %>% identical(meaningful_fourth_downs[meaningful_fourth_downs$play_type %in% c("pass", "run") & meaningful_fourth_downs$special_teams_play != 1,])
meaningful_fourth_downs$should_go_for_it1 %>% table()
#############
# TABLES AND GRAPHS OF PLAY TYPE DISTRIBUTION, REAL AND RECOMMENDED
rec_breakdown <- data.frame("Field Goal" = (meaningful_fourth_downs$should_kick %>% table())[2] / nrow(meaningful_fourth_downs),
"Conv. Attempt" = (meaningful_fourth_downs$should_go_for_it1 %>% table())[2] / nrow(meaningful_fourth_downs),
"Punt" = (meaningful_fourth_downs$should_punt %>% table())[2] / nrow(meaningful_fourth_downs))
real_breakdown <- data.frame("Field Goal" = (meaningful_fourth_downs %>% pull(play_type) %>% table)[1] / nrow(meaningful_fourth_downs),
"Conv. Attempt" = ((meaningful_fourth_downs %>% pull(play_type) %>% table)[2] + (meaningful_fourth_downs %>% pull(play_type) %>% table)[4]) / nrow(meaningful_fourth_downs),
"Punt" = (meaningful_fourth_downs %>% pull(play_type) %>% table)[3] / nrow(meaningful_fourth_downs))
is.matrix(table(meaningful_fourth_downs$qtr, meaningful_fourth_downs$should_go_for_it1))
barplot(matrix(unlist(rbind(real_breakdown[1, , drop = TRUE], rec_breakdown[1, , drop = TRUE])), nrow = 2, ncol = 3),
ylim = c(0, 1),
beside = TRUE,
main = "4th Down Play Type Breakdown \n(2012-2022)",
names = c("Field Goal", "Conv. Attempt", "Punt"),
col = c("gray","dodgerblue"),
ylab = "Proportion",
xlab = "Play Type")
legend("topleft",
legend = c("Real", "Recommended"),
col = c("gray","dodgerblue"),
fill = c("gray","dodgerblue"))
barplot(unlist(rec_breakdown[1, , drop = TRUE]),
ylim = c(0, 0.7),
main = "Model Recommendation \n(2012-2022)",
names = c("Field Goal", "Conv. Attempt", "Punt"),
col = "dodgerblue",
ylab = "Proportion",
xlab = "Play Type")
#############
#############
# eWPA vs. eFGWPA VIZ
plot(expected_wpa ~ abs(score_differential), data = meaningful_fourth_downs,
ylab = "eWPA",
xlab = "Score Differential",
main = "Expected Win Probability Added vs. Score Differential",
col = "gray") # low score idfferential prefers field goals
points(fg_expected_wpa ~ abs(score_differential), data = meaningful_fourth_downs,
col = rgb(blue = 0.7, red = 0.4, green = 0.2, alpha = 0.2))
legend("bottomright",
pch = 16,
col = c(rgb(blue = 0.7, red = 0.4, green = 0.2), "gray"),
legend = c("eFGWPA", "eAWPA"),
cex = 1.5)
# fg wpa expected is more moderate at lower score differentials and both get more moderate until 14, but around 14, fg wpa expected begins to become more extreme than going for it
# the lowest expected wpa for each sharply increases
plot(expected_wpa ~ yardline_100, data = meaningful_fourth_downs %>% filter(abs(score_differential) <= 2)) # low score idfferential prefers field goals
points(fg_expected_wpa ~ yardline_100, data = meaningful_fourth_downs %>% filter(abs(score_differential) <= 16, game_seconds_remaining <= 120),
col = rgb(blue = 1, red = 0, green = 0, alpha = 0.2)) # when it makes it a 3 possession game
plot(fgm_prob ~ yardline_100, data = meaningful_fourth_downs)
plot(expected_wpa ~ game_seconds_remaining, data = meaningful_fourth_downs) # low score differential prefers field goals
points(fg_expected_wpa ~ game_seconds_remaining, data = meaningful_fourth_downs,
col = rgb(blue = 1, red = 0, green = 0, alpha = 0.2))
#############
# can really only evaluate the cases of when they do go for it (evaluates if the difference in wpa predictions are valid)
# - if they should have but didn't get it, is abs(wpa_if_conv) > abs(wpa)? (correct if TRUE)
# - if they should have and got it, is abs(wpa) > abs(wpa_if_fail)? (correct if TRUE)
# - if they should not have but got it, is abs(wpa) < abs(wpa_if_fail)? (correct if TRUE)
# - if they should not have and didn't get it, is abs(wpa) > abs(wpa_if_conv)? (correct if TRUE)
# subset by the above cases and input T or F
# evaluate wpa estimation based on accuracy % of results (whether they actually should have gone for it assuming wpa is motivating factor)
meaningful_attempts1 <- meaningful_fourth_downs %>%
filter(play_type %in% c("pass", "run"), special_teams_play != 1, year %in% (2012:2022))
meaningful_attempts1$wpa_risk_accuracy <- TRUE
meaningful_attempts1$wrong_var <- ""
meaningful_attempts1$wpa_error <- numeric(1)
should_but_fail <- meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1 & meaningful_attempts1$should_go_for_it,]
meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1 & meaningful_attempts1$should_go_for_it,] <- should_but_fail %>%
mutate(wpa_risk_accuracy = abs(wpa_if_conv) >= abs(wpa), wrong_var = "wpa_if_fail too low")
should_n_conv <- meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1 & meaningful_attempts1$should_go_for_it,]
meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1 & meaningful_attempts1$should_go_for_it,] <- should_n_conv %>%
mutate(wpa_risk_accuracy = abs(wpa_if_fail) <= abs(wpa), wrong_var = "wpa_if_conv too high")
shouldnt_but_conv <- meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1 & !meaningful_attempts1$should_go_for_it,]
meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1 & !meaningful_attempts1$should_go_for_it,] <- shouldnt_but_conv %>%
mutate(wpa_risk_accuracy = abs(wpa_if_fail) >= abs(wpa), wrong_var = "wpa_if_conv too low")
shouldnt_n_fail <- meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1 & !meaningful_attempts1$should_go_for_it,]
meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1 & !meaningful_attempts1$should_go_for_it,] <- shouldnt_n_fail %>%
mutate(wpa_risk_accuracy = abs(wpa_if_conv) <= abs(wpa), wrong_var = "wpa_if_fail too high")
failed <- meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1,]
meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1,] <- failed %>%
mutate(wpa_error = abs(wpa_if_fail) - abs(wpa))
conv <- meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1,]
meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1,] <- conv %>%
mutate(wpa_error = abs(wpa_if_conv) - abs(wpa))
with(meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1,], mean(wpa_error))
with(meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1,], mean(wpa_error))
# error for failure is higher than error for conversion for NOYDS model
plot(wpa_error ~ abs(wpa), data = meaningful_attempts1[meaningful_attempts1$fourth_down_converted == 1,])
abline(0, 0)
plot(wpa_error ~ abs(wpa), data = meaningful_attempts1[meaningful_attempts1$fourth_down_failed == 1,])
abline(0, 0)
# more cases where abs(wpa_if_conv) is overestimated, but less off from real wpa than when wpa_if_conv is underestimated
# more cases where abs(wpa_if_fail) is underestimated, and these cases more extremely miscalculate wpa
# overall, the model seems to bias towards going for it on 4th down
should_go_table <- meaningful_fourth_downs$should_go_for_it1 %>% table()
should_go_table[2]/sum(should_go_table)
# Should go for it breakdown:
# NOYDS model: 37.2%
# YDS model: 72.6%
nrow(meaningful_attempts1[meaningful_attempts1$conversion_prob >= 0.5,])/nrow(meaningful_attempts1)
# both models indicate that coaches should start attempting more 4th downs
meaningful_attempts1[meaningful_attempts1$wpa_risk_accuracy,]$wrong_var <- NA
wpa_accuracy <- meaningful_attempts1 %>% pull(wpa_risk_accuracy) %>% table()
meaningful_attempts1 %>% pull(wrong_var) %>% table()
## This table shows that the model's errors in evaluating wpa occurs most often when it says a team should go for it, regardless of result (most often when they shpuld go for it and they get it)
## in other words it overestimates wpa if converted or underestimates wpa if fail
nrow(meaningful_attempts1)
1336/nrow(meaningful_attempts1)
# YDSTOGO model better approximates wpa, but no YDS model more closely follows how NFL coaches currently make decisions and is more conservative
# Show how YDSTOGO model in practical situations may be extreme, can get a coach fired in this era - worth it to trail blaze this much?