The CFB Zero-Score Bias: Why Underdog Team Totals Are Mispriced Every Season Opener

College football Week 1 kicks off September 6. The major books have posted lines for LSU vs. Clemson, Notre Dame vs. Wisconsin, and several other openers. Sharp bettors are already looking at one specific market most recreational players ignore: team totals.

A peer-reviewed paper published in the Journal of Sports Economics in 2023 identified a structural pricing flaw in college football team totals markets. The flaw stems from a mathematical bias called censoring, and it has produced a win rate above 55 percent for over two decades. The opportunities cluster in games where one team's implied score is low, which happens regularly in Week 1 when large favorites meet conference underdogs.

This is not a public-fading angle or a trend report. It is a documented market inefficiency with a clean causal explanation. The following is how it works and how to apply it right now.

What a Team Total Is

A game total (the over/under you see on the main betting board) covers the combined score of both teams. A team total is a separate prop covering only one team's score. If Notre Dame plays Wisconsin and the game total is 46.5, the team total options look like this: Notre Dame team total over/under 31.5, Wisconsin team total over/under 15.

Team totals are available at DraftKings and FanDuel under the team props section for most college football games. Limits run lower than on the main line, which also means the market corrects more slowly.

Where the Bias Comes From

When a book prices a team total, it works backward from the game total and the spread. The implied score for each team follows a straightforward formula:

• Favorite implied score = (game total + spread) / 2
• Underdog implied score = (game total - spread) / 2

For Notre Dame at -16.5 with a 46.5 total: Notre Dame's implied score is (46.5 + 16.5) / 2 = 31.5. Wisconsin's implied score is (46.5 - 16.5) / 2 = 15.0.

The problem is that football scores are bounded at zero. Teams do not score negative points. A model treating scores as symmetrically distributed around the implied mean understates the probability of low-scoring outcomes for underdogs. When the implied score is 15, the distribution of actual scores is not centered at 15 the way a symmetric model assumes. Probability mass from below-zero outcomes (which the mathematical model allows but reality does not) piles up near zero, pulling the real-world score distribution to the right and raising the true expected score above the naive implied price.

This is censoring bias. The floor at zero creates a right-skewed distribution for low-implied-score teams. The actual probability of scoring above the naive implied value is meaningfully higher than 50 percent. If the book sets the team total at that implied value and prices both sides at the standard -110, the over carries positive expected value.

Robert Arscott formalized this in his 2023 paper in the Journal of Sports Economics: "Market Efficiency and Censoring Bias in College Football Gambling." Using data spanning over two decades of college football games, Arscott showed a betting strategy exploiting this bias, using only the publicly available game total and spread, produced a win rate above 55 percent. The paper concluded the college football team totals market is semi-strong inefficient because of this censoring effect. The inefficiency has not been arbitraged away despite being academic public record.

The Math in Practice

For a low-scoring underdog in a moderate-total game, the zero-floor correction is significant. Take Wisconsin at an implied score of 15.0 (Notre Dame -16.5, total 46.5).

A simple truncated normal calculation shows why the naive line is wrong. Assume college football team scoring follows a normal distribution with standard deviation around 12 points, which is consistent with historical CFB scoring variance. For a team implied at 15.0 with that variance:

• The uncorrected (naive) line: 15.0
• The zero-floor-corrected expected score: approximately 17.5
• The probability of scoring above 15.0 (given the correction): approximately 56 percent

That 56 percent figure matches what Arscott's 20-year empirical data found. If the book sets Wisconsin's team total at 15 or 15.5 (the naive implied range), the over wins at roughly that rate. The vig on a standard -110 line requires 52.4 percent to break even. A 56 percent win rate clears that bar by a meaningful margin.

Here is a current Week 1 snapshot:

The threshold for meaningful bias is roughly when the implied score falls below 18. At 15.0, Wisconsin's team total sits in the high-bias zone. UNC's implied score of 21.5 is above that threshold. The censoring correction shrinks quickly as the implied score rises above 18-20, because the zero floor matters less when the expected value is further from zero.

For LSU vs. Clemson and Auburn vs. Baylor, the game totals were not finalized at time of research. If LSU vs. Clemson comes in at a total below 45, Clemson's implied score at -11 would be (45-11)/2 = 17.0, putting it in the high-bias zone.

Why Week 1 Is the Best Window

Two effects compound in the season opener.

The first is the censoring bias above. It exists throughout the season. Week 1 magnifies it because team totals markets are thinner early in the year, with lower limits and fewer sharp bettors tracking them. A bias corrected by October in game 8 stays uncorrected longer in game 1.

The second is holdover bias, documented by Randall Bennett in a 2019 paper in the Atlantic Economic Journal. Bennett studied betting lines for prior-year AP Top 25 teams in their season openers and found prior-year top-10 teams are systematically overvalued at the start of the following season. Bets against them won at rates well above the 52.4 percent break-even, particularly in games against non-Power 5 opponents. The mechanism is behavioral: bettors and market makers anchor to the prior year's rankings and overweight reputation when there is no current game film to reference.

For 2026: Notre Dame and LSU carry top-10 holdover status from 2025. The market prices them too short. Their opponents, Wisconsin and Clemson, have extra value on the spread. Those same opponents carry the censoring bias in their team totals.

The combination: a holdover-biased spread and a censoring-biased team total in the same ticket. You do not need both to act. The censoring bias in team totals stands on its own.

What the Market Gets Wrong (And Why It Persists)

Francisco and Moore's 2019 paper in the Journal of Economics and Finance, "Betting with house money: reverse line movement based strategies in college football totals markets," tested whether following sharp-money signals in CFB game totals is profitable. Their conclusion: it is not. Reverse line movement strategies in CFB totals showed no consistent edge over the 2005-2016 sample.

This matters for two reasons. It confirms the main game-totals market is relatively efficient. Sharp money flowing into game totals is processed quickly. Second, it frames why team totals remain soft: the sharp money disciplining the main line does not flow into team totals with the same force. Lower limits mean sharp bettors hit the game total first, and the team total market lags behind.

The censoring bias persists for the same reason: correcting it requires a mathematical adjustment that no simple trend-following approach captures. The public does not bet team totals based on truncated normal distributions. The correction requires academic insight, not wagering pressure. A 20-year edge does not close on its own when the mechanism is invisible to most participants.

How to Run the Play

Before Week 1 games post to the team props section, pull the game total and spread from the main board. Compute each team's implied score. When the underdog's implied score falls below 18, check the team total line on DraftKings or FanDuel once it posts.

If the team total is set at or below the implied score, the over is the bet. If the team total is set materially above the implied score (two or more points above it), the censoring correction is already partially priced in. Pass or size down.

Bet early in the week when the line first posts. Team totals on lower-profile matchups often sit without adjustment for days. By midweek, books start pricing them more accurately. The edge lives in the gap between line posting and market correction.

Size these as a smaller fraction of a normal bet. Liquidity is thin. Getting $200 into a team total is easier than getting $500. If you hit limits early, spread across two books.

Checking Your Work

Before each game, run through four checks:

1. Compute the implied score for the underdog: (total - spread) / 2.
2. Confirm the team total posts at or within two points of the implied score.
3. Confirm the game total has not moved significantly after the team total posted. If the game total dropped after the team total was set, the implied score is lower than when the book set the line, making the over even more +EV.
4. Check for injuries that would reduce the underdog's scoring floor. A team total over on a team missing its starting quarterback is a different thesis.

The Arscott strategy is mechanical: spread and game total, no qualitative inputs. Adding an injury filter narrows the play count but improves the quality of plays taken.

The 2026 Week 1 Watchlist

Notre Dame vs. Wisconsin at Lambeau Field on September 6 is the clearest current case. Wisconsin's implied score of 15.0 at a 46.5 total puts it in the high-bias zone. Once team totals post on DraftKings and FanDuel, check whether Wisconsin's line is at or below 15.5. If so, it fits the Arscott criteria.

Check LSU vs. Clemson once the game total finalizes. At LSU -11, Clemson's implied score at a 44-point total would be 16.5. At 46, it is 17.5. Either number sits in the high-bias zone.

For TCU vs. UNC at -6.5 and 49.5: UNC's implied score of 21.5 is above the threshold. The bias is weaker here. Monitor but do not prioritize.

Games beyond Week 1 follow the same setup: any CFB game with a total below 42 and a spread above 10 pushes underdogs into sub-18 implied territory. The opportunity is not exclusive to Week 1. It is largest in Week 1 because market depth is lowest and holdover bias adds a second layer.

The Return on Getting It Right

A 55 percent win rate on -110 bets produces a return on investment of about 2.5 percent per game. That is thin on a single bet. Across a CFB season, it compounds.

A bettor placing 40 plays per season at this win rate, sizing at a flat $100 each, nets roughly $100 in pure EV above the break-even threshold. That is not life-changing. What it is: a documented, non-zero edge in a market most people ignore. Combined with holdover-biased underdogs and smart timing, the per-play edge improves further.

The goal of running this play is not to replace your main betting strategy. It is to add a separate, academically documented edge in a thin market where the math works in your favor. Most bettors lose because they take -EV bets repeatedly. Running a +EV play 40 times per season, even one with a modest edge, is the correct direction.

Bottom Line

The censoring bias in college football team totals has been in the academic record since 2023. The underlying mechanism (the zero-score floor distorting price distributions for low-implied-score teams) is straightforward once you see it. The 55 percent win rate over 20 years is not a data artifact. It is a structural feature of a market priced off a formula that does not account for mathematical reality.

Three weeks before the 2026 season opens, the plays are already visible. Wisconsin's implied score of 15 is on the board now. The math is three lines of arithmetic. The rest is execution.