For this week’s data dive, I want to investigate what contributes to a team’s ability to win games. In order to do this, I will focus on the “wl_home” variable, which is a binary “w” or “l” depending on whether the home team won or lost the game in question.
Potential explanatory variables include a the home teams’ field goal percentage (how efficient the home team is shooting), their rebounding totals, their turnover totals, and assist totals. I selected these because each are important statistics that contribute to a team’s success on the court, but also are isolated enough where we should be able to limit collinearity.
log_model <- glm(
home_win ~ fg_pct_home + reb_home + tov_home + ast_home,
data = NBA_Data,
family = binomial
)
summary(log_model)##
## Call:
## glm(formula = home_win ~ fg_pct_home + reb_home + tov_home +
## ast_home, family = binomial, data = NBA_Data)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -17.644267 0.202500 -87.132 < 2e-16 ***
## fg_pct_home 28.168233 0.359937 78.259 < 2e-16 ***
## reb_home 0.166227 0.002373 70.058 < 2e-16 ***
## tov_home -0.122756 0.003321 -36.961 < 2e-16 ***
## ast_home -0.010968 0.002994 -3.663 0.000249 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 53314 on 39801 degrees of freedom
## Residual deviance: 37681 on 39797 degrees of freedom
## AIC: 37691
##
## Number of Fisher Scoring iterations: 5The model estimates the log‑odds of a home win using four performance variables. Each coefficient represents the change in log‑odds of winning for a one‑unit increase in the predictor, holding the others constant.
This is the dominant predictor in the model. A coefficient of 28.17 means that even small changes in shooting percentage have a large effect on win probability. Because FG% is on a 0–1 scale, a one‑percentage‑point increase (0.01) changes the log‑odds by about 0.28, which corresponds to roughly a 32% increase in the odds of winning. This aligns with basketball intuition: shooting efficiency is one of the strongest determinants of game outcomes.
The coefficient of 0.166 indicates that each additional rebound increases the log‑odds of winning. Converting to an odds ratio, a single rebound raises the odds of winning by about 18%. Rebounding extends possessions and limits opponent opportunities, so this effect is consistent with game dynamics.
The negative coefficient (–0.123) shows that turnovers reduce win probability. Each turnover decreases the odds of winning by about 11–12%, reflecting the cost of lost possessions and transition opportunities for opponents.
The coefficient is small (–0.011) and negative, which may seem counterintuitive. This often happens when assists are highly correlated with shooting percentage and field goals made. Once FG% is in the model, assists add little unique information, and the sign can flip due to multicollinearity. The effect is statistically significant but practically small.
Using the standard error for fg_pct_home, the formula for a confidence interval is:
(28.168233 - (z * 0.359937), 28.168233 + (z * 0.359937))
For a confidence interval of 95% (wanting to minimize standard error, but stay within reason), this yields approximately:
Lower bound: 27.46
Upper bound: 28.88
Because the entire interval is positive and far from zero, the data strongly support the conclusion that higher shooting percentage increases the probability of winning. In practical terms, even the lower end of the interval implies that small improvements in shooting efficiency have a meaningful impact on game outcomes.
The model captures the essential logic of NBA games:
Teams win when they shoot well.
Rebounding helps secure extra possessions.
Turnovers hurt.
Assists matter, but their effect is overshadowed by shooting efficiency.