I recently shared my possession data and wrote a brief post on how to use it to dig deeper into pace. Today I’m going to do some exploratory analysis on that data to breakdown the efficiency on possessions following a missed field goal based on the position of the rebounder. Intuition would suggest that when a primary ball handler gets a defensive rebound it should create more fast break opportunities and lead to a more efficient offense. Let’s see if the data backs that up.
import pandas as pd
import json
possessions = pd.read_csv('possession_details_00217.csv')
Possessions off a live ball rebound will have a player id in the ‘PreviousPossessionEndReboundPlayerId’ column. I’m also going to exclude possessions following missed free throws since possessions after missed free throws are less free flowing than after missed field goals.
off_fg_rebound = possessions[(possessions['StartType'] != 'OffFTMiss') | (possessions['PreviousPossessionEndReboundPlayerId'] != 0)]
off_fg_rebound.is_copy = False
The ‘PlayerStats’ column contains all stats in a nested dictionary. Here is a helper function that can be used to sum up a list of keys.
def sum_player_stats_keys_for_possession(row, keys):
"""
keys - list if keys to be summed up
"""
team_id = str(row['OffenseTeamId'])
player_stats_dict = json.loads(row['PlayerStats'])
value = 0
for lineup_id in player_stats_dict[team_id].keys():
for opponent_lineup_id in player_stats_dict[team_id][lineup_id].keys():
for player_id in player_stats_dict[team_id][lineup_id][opponent_lineup_id].keys():
player_stats = player_stats_dict[team_id][lineup_id][opponent_lineup_id][player_id]
for key in keys:
value += player_stats.get(key, 0)
return value
Here are all the possible stat keys to sum up to get total points.
points_keys = ['FtsMade', 'AssistedAtRim', 'AssistedAtRim', 'AssistedShortMidRange', 'AssistedShortMidRange', 'AssistedLongMidRange', 'AssistedLongMidRange', 'AssistedCorner3', 'AssistedCorner3', 'AssistedCorner3', 'AssistedArc3', 'AssistedArc3', 'AssistedArc3','UnassistedAtRim', 'UnassistedAtRim', 'UnassistedShortMidRange', 'UnassistedShortMidRange', 'UnassistedLongMidRange', 'UnassistedLongMidRange', 'UnassistedCorner3', 'UnassistedCorner3', 'UnassistedCorner3', 'UnassistedArc3', 'UnassistedArc3', 'UnassistedArc3']
We can then use the apply function to add a new column to the table with the sum of the keys. Note for the possession counts we need to divide the total by 5 since there is a value for each player on the floor for each possession.
off_fg_rebound['points'] = off_fg_rebound.apply(sum_player_stats_keys_for_possession, axis=1, args=(points_keys,))
off_fg_rebound['possessions'] = off_fg_rebound.apply(sum_player_stats_keys_for_possession, axis=1, args=(['OffPoss'],)) / 5
I’m going to use the positions on basketball-reference . These aren’t perfect but are fine for some exploratory analysis.
positions = pd.read_csv('positions.csv')
We need to merge the two tables by the player id.
off_fg_rebound = off_fg_rebound.merge(positions, left_on='PreviousPossessionEndReboundPlayerId', right_on='player_id')
We can now group by position and sum up the total points and possessions for each position
by_position = off_fg_rebound[['position', 'points', 'possessions']].groupby('position').sum()
by_position['efficiency'] = by_position['points'] / by_position['possessions']
by_position
So there is a small boost to efficiency following defensive rebounds by point guards. What about fast break opportunities? Let’s look at how often teams score within 5 seconds of securing a rebound.
score_within_5_seconds = off_fg_rebound[(off_fg_rebound['points'] > 0) & (off_fg_rebound['StartTime'] - off_fg_rebound['EndTime'] < 5)]
score_within_5_seconds[['position', 'possessions']].groupby('position').sum()
Teams score within 5 seconds 7.3% (1744/23888) of the time after a center gets a defensive rebound and 13.2% (1628/12355) of the time after a point guard gets a defensive rebound. So when point guards get defensive rebounds there are more transition baskets scored.
This is nothing ground breaking. It doesn’t mean teams should go out of their way to have their point guard gather all defensive rebounds, it could just be that point guards get longer rebounds which lead to more fast breaks. Deeper analysis should be done before we draw a real conclusion, but it’s nice to see the data match the intuition on the surface.