Chapter 16 Quiz: NBA Modeling
Test your understanding of NBA-specific modeling concepts, data, and market structure.
Question 1. Name the Four Factors of basketball identified by Dean Oliver and state the approximate weight each carries in explaining team success.
Answer
The Four Factors are: (1) Effective Field Goal Percentage (eFG%) -- approximately 40% importance; (2) Turnover Rate (TOV%) -- approximately 25% importance; (3) Offensive Rebound Rate (ORB%) -- approximately 20% importance; (4) Free Throw Rate (FTR, free throws per field goal attempt) -- approximately 15% importance. These weights apply to both offense and defense, with the defensive equivalents being opponent eFG%, forced turnover rate, defensive rebound rate, and opponent free throw rate.Question 2. What is the formula for Offensive Rating and why is it preferred over points per game?
Answer
Offensive Rating = (Points Scored / Possessions) * 100. It is preferred over points per game because it controls for pace. A team that scores 115 points per game at a pace of 105 possessions is less efficient than a team that scores 110 points at a pace of 95 possessions. Offensive Rating isolates efficiency from opportunity, making it a fairer comparison across teams with different playing styles.Question 3. How do you estimate the number of possessions in an NBA game from box score data?
Answer
Possessions = FGA - OREB + TOV + 0.44 * FTA. This formula accounts for field goal attempts (each ending a possession unless there is an offensive rebound), turnovers (which end possessions), and free throws (where 0.44 approximates the fraction of free throw trips that end possessions, accounting for and-ones, technical free throws, and three-shot fouls). Both teams will have approximately the same number of possessions in a game.Question 4. What is the typical home-court advantage in the NBA expressed in points, and how has it changed in recent years?
Answer
The typical NBA home-court advantage has historically been approximately 3-3.5 points. In recent seasons it has declined to approximately 2-2.5 points. The decline has been attributed to improved travel conditions, better scouting and preparation technology, and league-wide changes in playing style that reduce the impact of home crowd noise on free throw shooting and officiating.Question 5. Why does the "back-to-back" (B2B) schedule situation matter for NBA modeling?
Answer
Teams playing the second game of a back-to-back typically show reduced performance due to fatigue, especially when combined with travel. Historical data shows B2B teams perform approximately 1-2 points worse than expected. The effect is amplified when the team is on the road, has traveled a long distance, or is playing against a rested opponent. Modeling B2B effects is essential because they occur frequently in the 82-game season and the market does not always fully price them in.Question 6. Explain True Shooting Percentage (TS%) and why it is preferred over standard field goal percentage.
Answer
TS% = Points / (2 * (FGA + 0.44 * FTA)). It is preferred over standard FG% because it accounts for three-point shooting (which produces more points per made shot) and free throws (which produce points without a field goal attempt). A player who shoots 45% from three is more efficient than one who shoots 50% from two, but standard FG% would suggest the opposite. TS% captures the true scoring efficiency by considering all scoring methods.Question 7. What is pace and how should it be incorporated into a totals prediction model?
Answer
Pace is the estimated number of possessions per 48 minutes for a team. To predict the total for a game, you must estimate the pace of the game (typically the average of the two teams' pace, adjusted for home/away and other factors), then multiply by the combined points per possession. Total = Game Pace * (Home ORtg/100 + Away ORtg/100). Without pace adjustment, a model cannot distinguish between efficient low-pace teams and inefficient high-pace teams.Question 8. What is the approximate standard deviation of NBA game point differentials, and how does it compare to the NFL?
Answer
The standard deviation of NBA game point differentials is approximately 12-13 points. This is similar in absolute terms to the NFL's approximately 13.5 points, but because NBA scores are much higher (200+ total points versus roughly 45), the coefficient of variation is much lower. This means NBA games are more predictable in relative terms, and the point spread market tends to be more efficient.Question 9. Why is net rating a better predictor of future performance than win-loss record in the NBA?
Answer
Net rating (offensive rating minus defensive rating) strips out the influence of close-game variance. A team that wins many close games will have a better win-loss record than their net rating suggests, and research consistently shows that close-game performance (games decided by 5 or fewer points) is not very predictive -- it is largely driven by luck in the short run. Net rating is the NBA equivalent of pythagorean wins, and teams that deviate significantly from their net-rating-implied record tend to regress toward it.Question 10. Describe how you would model the impact of a star player's absence on the point spread.
Answer
Use the player's on/off court net rating differential to estimate their point impact per 48 minutes, then scale by their typical minutes played. For example, if a team's net rating is +8.0 with the player on court and +1.0 with the player off court, the player's on/off differential is +7.0. If the player plays 34 minutes per game (70.8% of 48 minutes), their absence would cost approximately 7.0 * 0.708 = 5.0 points. This should be regressed toward a prior based on the player's box score statistics and role, as raw on/off data has significant noise due to lineup interactions.Question 11. What role does three-point shooting variance play in NBA totals betting?
Answer
Three-point shooting has high game-to-game variance. A team that averages 36% from three might shoot 28% in one game and 44% in the next. Since teams attempt 30-40 threes per game, each percentage point of three-point variance translates to roughly 0.3-0.4 points of scoring variance. This makes three-point shooting a significant contributor to game total uncertainty and explains why NBA totals are difficult to predict precisely. Models should use regressed three-point percentages rather than recent hot or cold streaks.Question 12. How should a modeler handle the first two weeks of the NBA season when current-season data is very limited?
Answer
Use Bayesian priors derived from: (1) previous season's performance regressed toward the mean (approximately 25-30% regression); (2) offseason roster changes, weighted by the impact of players gained and lost; (3) preseason market information such as win total lines; (4) coaching changes and scheme shifts if applicable. The prior should dominate predictions in weeks 1-2 and gradually give way to current-season data. By approximately 20-25 games, current-season data should carry most of the weight.Question 13. What is "closing line value" (CLV) and why is it considered the gold standard for evaluating a bettor's skill?
Answer
CLV measures whether a bettor consistently gets better numbers than the closing line. If you bet Team A at -3 and the line closes at -4, you captured 1 point of CLV. Positive CLV indicates that, over time, the bettor is beating the market's best estimate of the true probability. CLV is preferred over raw profit because it is less subject to short-term variance. A bettor with consistent positive CLV will be profitable in the long run even if short-term results are negative due to variance.Question 14. Explain the concept of "garbage time" in NBA games and its impact on ATS results.
Answer
Garbage time in the NBA occurs when the outcome is effectively decided, typically a lead of 20+ points in the fourth quarter. During garbage time, the losing team often scores at a higher rate (against bench players and relaxed defense) while the winning team scores at a lower rate. This causes large favorites to cover less often than expected -- the losing team narrows the gap in meaningless minutes. This systematic effect on ATS results is sometimes called the "garbage time backdoor cover" and should be considered when modeling large-favorite outcomes.Question 15. How does the NBA's conference-based schedule affect strength of schedule adjustments?
Answer
NBA teams play more games against teams in their own conference (and especially their own division) than against the opposite conference. This means raw win-loss records are biased by conference strength. A team in a weak conference will have an inflated record compared to an equally talented team in a strong conference. Modelers must apply opponent adjustments to account for this scheduling imbalance. Net rating with opponent adjustment is the best way to compare teams across conferences fairly.Question 16. What is the importance of free throw rate (FTR) in the Four Factors framework?
Answer
Free throw rate (FTA/FGA) captures a team's ability to get to the free throw line, which is valuable because free throws are the most efficient shot in basketball (approximately 75% hit rate, yielding 1.5 points per two-shot trip). While FTR has the smallest weight among the Four Factors (~15%), it still contributes meaningfully to offensive efficiency. On the defensive side, preventing opponents from getting to the line is also important. FTR is more stable than eFG% game-to-game, making it a reliable feature for modeling.Question 17. Describe how you would build an NBA player prop model for points scored.
Answer
Start with the player's season average points per game. Adjust for: (1) opponent defensive rating and their ability to defend the player's position; (2) game pace -- faster games create more scoring opportunities; (3) home/away split; (4) rest status (back-to-back, etc.); (5) teammate availability -- if a high-usage teammate is out, the player's usage and points may increase; (6) recent form (last 5-10 games), regressed toward the season mean. Use a regression model trained on historical player-game data with these features. Evaluate against posted prop lines using log-loss or calibration analysis.Question 18. Why do NBA playoff games tend to have lower totals than regular-season games between the same teams?
Answer
Playoff games typically feature slower pace, better defensive effort and preparation, and tighter officiating. Teams have time to game-plan specifically for their opponent, and the heightened stakes lead to more half-court basketball. Historical data shows playoff games average 3-5 fewer total points than regular-season games between the same teams. Totals models should include a playoff adjustment factor, and bettors should be aware that the market sometimes underadjusts for this effect early in the playoffs.Question 19. What is a "look-ahead" spot in NBA scheduling and how might it affect betting?
Answer
A look-ahead spot occurs when a team has a high-profile game coming up (e.g., a rivalry game or a nationally televised matchup) and may not give full effort in the preceding game against a lesser opponent. This is a form of motivational bias that is difficult to model quantitatively but has some empirical support. Bettors who track scheduling context can sometimes identify games where a favorite may underperform due to a look-ahead spot. The market often partially accounts for this, but not always fully.Question 20. How would you model the impact of a mid-season coaching change on team performance?
Answer
Mid-season coaching changes typically produce a short-term "bump" of 2-3 points in performance, likely due to renewed effort and attention. However, this effect tends to fade over 10-15 games. A modeler should: (1) apply a temporary positive adjustment of approximately 2 points for the first 5-10 games after a coaching change; (2) gradually reduce the adjustment over the next 5-10 games; (3) update the team's efficiency metrics as new data under the new coach accumulates. The market typically adjusts quickly, so the edge from coaching changes is short-lived.Question 21. Explain the relationship between three-point attempt rate and offensive rating variance.
Answer
Teams that rely heavily on three-point shooting have higher game-to-game variance in offensive performance because three-point shooting is inherently more variable than two-point shooting. A team that takes 45 three-point attempts per game will have wider swings in scoring efficiency than a team that takes 30. This means high-three-point-volume teams are harder to predict game-to-game and may present more totals betting opportunities. The variance also means their ATS record may be more variable -- they will exceed expectations more often but also fall short more often.Question 22. What data sources are most valuable for NBA modeling?
Answer
The most valuable data sources are: (1) NBA.com/stats API -- official box scores, play-by-play data, tracking data, and advanced metrics; (2) Basketball Reference -- historical data, player stats, team ratings; (3) Cleaning the Glass -- filtered statistics removing garbage time and end-of-quarter heaves; (4) PBPStats -- detailed play-by-play analysis; (5) ESPN and CBS Sports APIs for schedule and lineup data; (6) injury report aggregators for real-time availability updates. For tracking data (player movement, shot location), the NBA's official stats portal provides the richest public data.Question 23. Describe the "revenge game" narrative in NBA betting. Is there empirical evidence for it?
Answer
The "revenge game" narrative suggests that a player or team performs better when facing their former team. Empirical evidence is mixed to negative. While individual players may occasionally have notable performances against former teams, systematic studies show no statistically significant improvement in team ATS performance in revenge game spots. The narrative is a cognitive bias (the availability heuristic makes memorable revenge performances salient) rather than a reliable predictive factor. Models should not include a revenge game feature.Question 24. How does the NBA All-Star break affect team ratings and predictions for the second half of the season?
Answer
The All-Star break provides a natural midpoint to reassess team ratings. First-half performance (approximately 50-55 games) provides a substantial sample for estimating team quality. However, several factors change at the break: (1) the trade deadline occurs nearby, reshuffling rosters; (2) some teams begin to tank or rest players; (3) playoff-contending teams may adjust rotations. A good model should down-weight pre-All-Star data slightly in post-break predictions and be alert to roster changes. The market adjusts gradually, creating potential edges in the first 1-2 weeks after the break.Question 25. What is the typical edge a sharp NBA bettor needs to overcome the vig and be profitable?