How Dr. TrueLine's NBA Model Works

The NBA is one of the most heavily bet sports in the world. Lines move fast, markets are efficient, and sportsbooks have decades of data refining their pricing. Finding an edge requires more than checking who's hot and who's not.

Our NBA model takes a different approach than our MLB engine. Where MLB is moneyline-based and pitcher-driven, the NBA model is spread-based and efficiency-driven. The core question is the same — is the book's line accurate? — but the variables and the filtering strategy are built for basketball's unique dynamics.

Spreads, not moneylines

Our MLB model predicts moneyline outcomes — who wins. Our NBA model predicts spreads — by how much.

This is a deliberate choice. In the NBA, Vegas is extremely good at picking winners. The favorite wins roughly 67% of games. But Vegas is less precise on the margin — how many points the favorite wins by. That's where the edge lives.

The model calculates what we think the true spread should be, compares it to the book's posted spread, and identifies games where the sportsbook's number is off. When the disagreement exceeds our proprietary threshold, the pick is flagged as Confirmed — our highest conviction tier.

The key variables

Our NBA engine evaluates several efficiency and situational factors for every game:

Net rating is the foundation — offensive rating minus defensive rating, measuring points scored vs allowed per 100 possessions. This is the single most predictive team-level stat in basketball, correlating with win percentage at r > 0.95.

Four factors composite breaks down why a team is winning or losing using Dean Oliver's framework: effective field goal percentage, turnover rate, offensive rebounding rate, and free throw rate. We weight these with proprietary ratios informed by academic research on what drives NBA wins.

Pythagorean regression compares a team's actual record to their expected record based on points scored and allowed. Teams significantly overperforming their Pythagorean expectation are candidates to regress. This is one of the most validated concepts in basketball analytics.

Rest and schedule fatigue is critical in the NBA in a way it isn't in baseball. Teams on the second night of a back-to-back win only about 43.6% of the time. The model adjusts for rest days, back-to-backs, and travel.

Home court advantage is a significant factor in the post-COVID era. Certain venues — like Denver's altitude — are consistently underpriced by books.

Turnover margin regression catches teams running hot or cold on turnovers. High turnover variance is unstable — teams forcing or committing unusual numbers of turnovers will regress.

Three-point shooting regression identifies teams shooting well above or below their expected rate from deep. Teams running hot from three will cool off. Teams in a cold streak will heat up. The model adjusts accordingly.

Direction disagree — our secret weapon

One of the most interesting signals in our NBA data is what we call a direction disagree. This happens when our model's calculated spread puts a different team as the favorite than Vegas does.

When our model and Vegas agree on who's favored (just disagreeing on the margin), our picks perform at a baseline level. But when our model outright disagrees with Vegas on which team should be favored — those picks have been strongly profitable with a significantly higher win rate.

These are rare — only about 10% of games — but they represent the model's strongest signal. A direction disagree means the model sees something fundamentally different about the matchup than the market does. When that happens, the model has been right far more often than not.

Direction disagree is tracked as one of our tier qualifiers. Second Opinion picks specifically capture games where the edge is moderate but the model disagrees with Vegas on direction.

The away underdog exclusion

Our analysis across our tracked NBA games revealed a pattern identical to what we found in MLB: away underdogs are a consistent money pit.

The numbers are stark. Away underdogs had a deeply negative ROI regardless of edge size. Every single edge bucket for away underdogs was negative. It didn't matter whether the model saw a 3-point edge or an 8-point edge — away underdogs lost money regardless.

Meanwhile, home picks were consistently profitable. Home underdogs specifically were the star performer. The model captures home court advantage well, but something about away underdog situations — whether it's travel, hostile crowd, or just how books price road dogs — creates a consistent leak.

So we filter them out. All away underdogs are now classified as No Finding regardless of edge size. This single change dramatically improved our included ROI.

The tier system

After filtering, picks fall into three tiers:

Confirmed picks exceed our highest-conviction edge threshold between our spread and the book's spread. These are our strongest plays — the model sees a significant disagreement with the market. Consistently our strongest performing tier.

Second Opinion picks have a moderate edge AND a direction disagree with Vegas. The edge is smaller, but the model fundamentally disagrees with which team should be favored. A small sample but a powerful signal historically.

No Finding covers everything else — games below our edge thresholds, games where the model and Vegas roughly agree, and all away underdogs. These are tracked but not recommended.

What the model doesn't do

We're transparent about limitations. The model does not account for player injuries beyond what's reflected in the team's recent performance. A star player ruled out an hour before tip fundamentally changes the game — our model won't catch that in real time.

The model uses full-season statistics without EWMA (exponentially weighted moving average) — meaning November blowouts count the same as last night's game. Adding recency weighting is on our development roadmap.

The NBA market is more efficient than MLB, which means edges are smaller and harder to find. Our model works — the data shows it — but it requires patience and volume. Individual game outcomes are noisy. The edge shows up over dozens of games, not one night.

See it in action

Every NBA pick is published on our Today's Plays page alongside MLB. Enable push alerts in your Account to get notified when official picks lock during the final hour before tip-off. The scoreboard tracks NBA performance separately — Confirmed, Second Opinion, and No Finding results are all visible. You can verify every claim made in this article by checking the data yourself.