Cuse Analysis
Max Whalen
2023-01-26
Contested Shots Splits
Not Self-Created Shots
Looking at catch and shoot/1-2 dribble pull-ups shooting stats based on varying degrees of ‘contestion’. Only statistics with 20 attempts from each split were included, few shots were marked as having poor contests so I chose not to include them
Useful supplements
- Clips of different close outs
- Clips demonstrating how overly aggressive closeouts can be negative (i.e. Steph Curry takes it the hoop off a pump fake)
## Vector of Shot Plays that require strong closeouts
shots <- c("1-2D Pull-Up", "Catch & Shoot")
results %>%
filter((off_team == "GSW") & (result_type %in% shots) & self_created == 0) %>%
mutate(made_shot = if_else(points > 1, 1, 0)) %>%
group_by(Player = result_off, result_contest) %>%
summarise(fg_pct = mean(made_shot), shots = n()) %>%
filter(shots >= 20) %>%
select(-shots) %>%
pivot_wider(names_from = result_contest, values_from = fg_pct) %>%
drop_na(Open, Plus) %>%
select(Open, Average, Plus) %>%
arrange(-Open)## # A tibble: 5 × 4
## # Groups: Player [5]
## Player Open Average Plus
## <chr> <dbl> <dbl> <dbl>
## 1 Jordan Poole 0.491 0.357 0.304
## 2 Klay Thompson 0.475 0.446 0.5
## 3 Stephen Curry 0.432 0.384 0.6
## 4 Andrew Wiggins 0.405 0.413 0.333
## 5 Otto Porter Jr. 0.378 0.505 0.326Self-Created shots between midrange and paint
The idea here was to examine how well these players shoot the ball when contested inside the 3-point arc. Which players’ shots should we contest, at the expense of giving up an easy look near the rim.
Useful supplements
- Clips of players chasing the block, at the expense of an easy look near the hoop (Think prime Hassan Whiteside before people realized that was a terrible way of playing defense)
## Shots where contests near the rim may be important, at the expense of creating an opportunity
shots <- c("1-2D Pull-Up", "Floater", "Leaner", "Hook", "PU Fade", "3+D Pull-Up")
## Basically how hard should we contest players in the midrange
results %>%
filter((off_team == "GSW") & (result_type %in% shots) & self_created == 1 &
(grepl("Midrange", result_zone) | grepl("Paint", result_zone) | grepl("Short", result_zone))) %>%
mutate(made_shot = if_else(points > 1, 1, 0)) %>%
group_by(Player = result_off, result_contest) %>%
summarise(fg_pct = mean(made_shot), shots = n()) %>%
filter(shots > 6) %>%
select(-shots) %>%
pivot_wider(names_from = result_contest, values_from = fg_pct) %>%
drop_na(Open) %>%
select(Open, Poor, Average, Plus) %>%
arrange(-Open)## # A tibble: 4 × 5
## # Groups: Player [4]
## Player Open Poor Average Plus
## <chr> <dbl> <dbl> <dbl> <dbl>
## 1 Klay Thompson 0.909 NA 0.581 0.375
## 2 Jordan Poole 0.636 0.4 0.368 0.370
## 3 Stephen Curry 0.5 0.429 0.486 0.375
## 4 Andrew Wiggins 0.412 0.364 0.403 0.338GSW Initiation
General Overview of GSW Initiation Duties
I don’t expect this to be insightful, but this is just a general view of who creates and how well in this offense
## Who's creating the most points when initiating
left_join(pbp, results, by = "result_id") %>%
arrange(poss_id.x) %>%
filter((off_team.x == "GSW") & (play_type == "Initiation") & (player_type == "Ball-Handler")) %>%
group_by(player_name) %>%
mutate(points = if_else(between(points, 0, 4), points, "NA")) %>%
filter(!is.na(points) | points != "NA") %>%
summarise(Avg_Points = mean(as.numeric(points)), Initiations = n()) %>%
filter(Initiations > 30) %>%
arrange(-Initiations)## # A tibble: 4 × 3
## player_name Avg_Points Initiations
## <chr> <dbl> <int>
## 1 Stephen Curry 0.815 168
## 2 Jordan Poole 0.992 130
## 3 Draymond Green 0.409 44
## 4 Andrew Wiggins 0.706 34Where are the primary initiators initiating from? To what effect?
I think here we see a little insight. Steph better from shallow 3 point area, Poole doesn’t matter. Speculation is Poole is more athletic, gets a step and makes you pay. Steph thrives in a “telephone booth”, one misstep and he can pull trigger. Easier to pass and shoot from closer to the basket.
Useful supplements
Any film support of this theory (foremost)
Info on o-reb conversion rate and ability to predict team success against warriors
-> If o-reb is equally big factor in success, then this strategy may be hard to implement
## Highlighting players that initiate
highlighted <- c("Stephen Curry", "Jordan Poole", "Draymond Green")
## Who is most effective creating from where?
left_join(pbp, results, by = "result_id") %>%
arrange(poss_id.x) %>%
filter((player_name %in% highlighted) & (play_type == "Initiation") & (player_type == "Ball-Handler")) %>%
group_by(player_name, play_zone) %>%
mutate(points = if_else(between(points, 0, 4), points, "NA")) %>%
filter(!is.na(points) | points != "NA") %>%
summarise(Avg_Points = mean(as.numeric(points)), Initiations = n()) %>%
filter(Initiations > 10) ## # A tibble: 5 × 4
## # Groups: player_name [3]
## player_name play_zone Avg_Points Initiations
## <chr> <chr> <dbl> <int>
## 1 Draymond Green Deep Backcourt 0.5 30
## 2 Jordan Poole Deep Backcourt 1.03 69
## 3 Jordan Poole Short Backcourt 1.06 47
## 4 Stephen Curry Deep Backcourt 0.660 103
## 5 Stephen Curry Short Backcourt 1.09 56Comparing GSW Pace with NBA Pace
Basically, if GSW show’s a higher propensity for scoring points late in the shot clock, we need to focus on causing havoc early in the posession at risk of exposing ourselves,
{
off_poss_summary <- read_csv("ops.csv")
def_poss_summary <- read_csv("ops.csv")
# Cleaning off_poss_summary
off_poss_summary %>%
filter(elapsed > 0) %>%
mutate(scored = if_else(points > 0, "Points > 0", "Points == 0"),
points = as.numeric(points),
team = "GSW") -> elapsed_df_gsw
# Cleaning off_poss_summary
def_poss_summary %>%
filter(elapsed > 0) %>%
mutate(scored = if_else(points > 0, "Points > 0", "Points == 0"),
points = as.numeric(points),
team = "NBA") -> elapsed_df_nba
}{
library(ggthemes)
library(ggridges)
}
{
gsw_blue <- "#1D428A"
gsw_gold <- "#FFC72C"
nba_blue <- "#1d428a"
nba_red <- "#c8102e"
rbind(elapsed_df_gsw, elapsed_df_nba) %>%
ggplot(aes(x = elapsed, y = team, fill = as.factor(scored))) +
stat_density_ridges(quantile_lines = TRUE, quantiles = 2, alpha = 0.65) +
scale_fill_manual(values = c(gsw_blue, gsw_gold)) +
guides(
fill = guide_legend(
label = TRUE, title = "", label.position = "left",
direction = "vertical",
label.theme = element_text(size = 20)
)
) +
theme_light() +
theme(panel.grid.major.y = element_blank(),
legend.position = c(0.8, 0.9)) +
labs(x = "Time Elapsed in Possesion",
y = "Who's on Offense")
}
We don’t really see a notable difference in density