Adding a 4-point line to the NBA using the NBA API
Like millions of other people, I am a big fan of JxmyHighroller’s youtube channel. He makes thought provoking videos on various NBA topics, with data analysis being at the core of his content. Recently, he added a video in which he considers the potential addition of a 4-point line to the NBA.
There are many things to consider. The first important question is where should the line be? Jimmy argues that it should be around the 30ft distance, so that the efficiency of the 4-point shot matches the 2 and 3-point shots.
Once the line is set at 30ft, he manually (painstakingly he says) goes through every game of the 2022–23 season (including playoffs), and concludes that only 2 games would have had their result flipped by the addition of a 4-point line at 30ft from the basket. To get this result he uses Basketball-Reference.com to check for any made shots from 30ft or further, and if there are enough to flip the final score, he concludes that the game would have had a different result.
In doing it by hand, he unfortunately makes a number of mistakes, due to not considering the impact of regulation vs overtime. This article aims to rectify his result, while extending it to the last 23 seasons. This is achieved through the use of the NBA API package.
We begin by importing the relevant packages and endpoints from the NBA API.
import requests
import json
import pandas as pd
from nba_api.stats.endpoints import shotchartdetail
from nba_api.stats.endpoints import leaguegamefinderWe now build a function that for a given season, returns a data frame containing info of all the games played, including playoffs.
def findAllGames(season):
# We query leaguegamefinder to get a record of
# all regular season games in the season
result_regular = leaguegamefinder.LeagueGameFinder(
season_nullable = season,
league_id_nullable = '00',
season_type_nullable = 'Regular Season')
# We store this in a data frame
all_games_regular = pd.DataFrame(result_regular.get_data_frames()[0])
# We query leaguegamefinder to get a record
# of all playoff games in the season
result_playoffs = leaguegamefinder.LeagueGameFinder(
season_nullable = '2022-23',
league_id_nullable = '00',
season_type_nullable = 'Playoffs')
# We store this in a data frame
all_games_playoffs = pd.DataFrame(result_playoffs.get_data_frames()[0])
#Finally we combine the two data frames
all_games = pd.concat([all_games_regular,all_games_playoffs])
return all_gamesWe can use the shotchartdetail endpoint to get a record of all made shots in a specific game.
def allShots(season):
#The following record contains all made shots of the regular season,
#note the use of '0' to allow any value
shot_details_regular = shotchartdetail.ShotChartDetail(
team_id = '0',
player_id = '0',
context_measure_simple = 'PTS',
season_nullable = season,
season_type_all_star = 'Regular Season')
#The following record contains all made shots of the playoffs
shot_details_playoffs = shotchartdetail.ShotChartDetail(
team_id = '0',
player_id = '0',
context_measure_simple = 'PTS',
season_nullable = season,
season_type_all_star = 'Playoffs')
#We combine the information in a single data frame containing
#all made shots of the season
shot_df_regular = pd.DataFrame(shot_details_regular.get_data_frames()[0])
shot_df_playoffs = pd.DataFrame(shot_details_playoffs.get_data_frames()[0])
shot_df_all = pd.concat([shot_df_regular,shot_df_playoffs])
return shot_df_allGiven a specific game id, we can now filter through the database of shots, find all the shots made in a single game, and check if the final result would have changed with a 4-point line. We output two arrays, the first one representing the game ids of the games that would have changed result from a win to a loss, and a second one if the game would have gone to an extra overtime.
def findAllGamesChanged(season,line_distance):
#this is used to store the ids of the games were the
#result would have been different
games_changed = []
games_to_overtime = []
#get all the games played
all_games = findAllGames(season)
#get all the shots
shot_df_all = allShots(season)
#get all ids making sure that all games are in both databases,
#to avoid errors
allIds = set(shot_df_all['GAME_ID'].unique()).intersection(
set(all_games['GAME_ID'].unique()))
#iterate through all the unique game ids of the season
for g_id in allIds:
#get the teams names and the points scored for the specific game
full_game = all_games[all_games.GAME_ID == g_id]
two_teams = full_game['TEAM_NAME'].unique()
points_team_zero = full_game[
full_game.TEAM_NAME == two_teams[0]]['PTS'].unique()[0]
points_team_one = full_game[
full_game.TEAM_NAME == two_teams[1]]['PTS'].unique()[0]
#the value pm denoted the difference in score,
#in particular team_zero won the game iff pm is positive
pm = points_team_zero-points_team_one
#filter the shots dataframe to this specific game
shot_df = shot_df_all[shot_df_all.GAME_ID== g_id]
#we need to pay attention to possible overtimes,
#the array periods contains the number of quarters
#if there was no overtime, periods = [1,2,3,4]
periods = shot_df['PERIOD'].unique()
#filter the shots to the 30+ footers
four_pointers = shot_df[shot_df.SHOT_DISTANCE >= line_distance]
#in order to take care of possible overtimes,
#we split the 4 pointers according to whether they were made
#in regular time or in a specific overtime
team_zero_gains = [len(four_pointers[(
four_pointers.TEAM_NAME == two_teams[0]) &
(four_pointers.PERIOD <= x)]) for x in periods]
team_one_gains = [len(four_pointers[(
four_pointers.TEAM_NAME == two_teams[1]) &
(four_pointers.PERIOD <= x)]) for x in periods]
#here team_zero_gains has length 1+ number of overtimes,
#and contains the appropriate number of 4 pointers made by team_zero
team_zero_gains = ([team_zero_gains[3]] +
[team_zero_gains[i]-team_zero_gains[i-1] for i in range(
4,len(team_zero_gains))])
team_one_gains = ([team_one_gains[3]] +
[team_one_gains[i]-team_one_gains[i-1] for i in range(
4,len(team_one_gains))])
#the array pms contains the various swings in points
#given by the 4-point line, with the inclusion of the final difference
#at the end
pms = [0 for x in range(len(team_zero_gains)-1)] + [pm]
#we are now ready to check if the result would have changed
for i in range(len(team_zero_gains)):
#in the following case, the team that won the game,
#scored more 4-pointers before going into another overtime,
#so we stop, knowing that the result would have stayed the same
if pms[i] == 0 and (pm)*(pms[i]+team_zero_gains[i]-
team_one_gains[i]) > 0:
break
#if there is a 4-point differential of opposite sign to the final
#score difference, it now means that the result
#would have changed. Note that changing from a strict inequality
#to <=, includes all games that would have been forced into overtime
if (pm)*(pms[i]+team_zero_gains[i]-team_one_gains[i]) <= 0:
if (pm)*(pms[i]+team_zero_gains[i]-team_one_gains[i]) == 0:
games_to_overtime.append(g_id)
else:
games_changed.append(g_id)
break
return([games_changed,games_to_overtime])Now that we have built the function, we can use it to automatically get Jimmy’s result, and see where he went wrong.
games_changed = findAllGamesChanged('2022-23',30)
print(len(games_changed[0]))The answer for the 2022–23 season is 9! There are 9 games that would have had the result flipped. In his video, JxmyHighroller claimed that there were only 2 games where this happened. The following prints all the relevant matchups for us to see.
#we can print all the relevant matchups,
#where the first team is the one that originally lost
all_games = findAllGames('2022-23')
full_game = all_games[(all_games.GAME_ID.isin(games_changed[0])) &
(all_games.WL == 'L')]
print(full_game[['MATCHUP','GAME_DATE']].to_string(index=False)) MATCHUP GAME_DATE
LAC vs. SAC 2023-02-24
LAL @ BOS 2023-01-28
GSW @ BOS 2023-01-19
ATL @ GSW 2023-01-02
CHA vs. DET 2022-12-14
CHI vs. NYK 2022-12-14
CHI @ ATL 2022-12-11
BOS vs. MIA 2022-12-02
GSW @ CHA 2022-10-29Consider for example the last game, which ended 120–113 in favour of CHA after one overtime. In this game Jordan Poole was the only one making a 30+ footer, and since this happened in regular time, it would have won GSW the game! However if one counts this as only one extra point for GSW, since the score was +7 for CHA, one gets the wrong conclusion!
We can now easily extend the result to the last 23 seasons!
all_seasons = (["200"+str(x)+"-0"+str(x+1) for x in range(9)]+
["2009-10"]+["20"+str(x)+"-"+str(x+1) for x in range(10,23)])
all_changes = []
for season in all_seasons:
all_changes.append([len(x) for x in findAllGamesChanged(season,30)])
df2 = pd.DataFrame(all_changes, columns=["Games Flipped","Games to Overtime"],
index=all_seasons)
We plot the number of games flipped (from win to loss or viceversa).
plot1 = df2[['Games Flipped']].plot.bar(stacked=True).set(xlabel="Season",
ylabel="Number of games")
We conclude by looking at the impact of different distances for the 4-point line on last season.
games_flipped_by_distance = []
for d in range(23,50):
games_flipped_by_distance.append(findAllGamesChanged('2022-23',d))
number_flipped_games = [len(x[0]) for x in games_flipped_by_distance]
df3 = pd.DataFrame(number_flipped_games, columns=["Games Flipped"],
index=range(23,50))
plot2 = df3.plot.bar(stacked=True).set(xlabel="Distance in feet",
ylabel="Number of games")
We conclude our journey by checking what game is surviving all distances, giving the tail of the graph.
all_games = findAllGames('2022-23')
full_game = all_games[(all_games.GAME_ID==
games_flipped_by_distance[-1][0][0]) & (all_games.WL == 'L')]
print(full_game[['MATCHUP','GAME_DATE']].to_string(index=False)) MATCHUP GAME_DATE
GSW @ BOS 2023-01-19Indeed Steph Curry scored from beyond half court, and that would have won the game in regulation, instead of losing in overtime.
