Two decisions cost recreational players more money than any others. The first is calling a bet because an opponent looked nervous. The second is taking a long price on a weak team because the payout is large and the story is appealing. Decades of research on deception detection put human accuracy at reading a liar near 53%, which is the number that should govern the first decision and almost never does.

The Arithmetic Behind a Bluff

A bluff is a bet made with a hand that cannot win a showdown. Its value comes from the price it offers the opponent. A bet of $50 into a $100 pot asks the opponent to risk $50 to win $150, so a fold is correct only some of the time, and the bluffer profits when the fold happens more often than the price justifies.

None of that requires reading a face. The frequency of profitable bluffing is set by pot size, bet size and the range of hands an opponent is likely to hold. Psychology enters through the opponent’s response, since a fold is a human decision made under pressure with incomplete information.

Two Audiences for the Same Bet

A bluff has one audience, the player who has to act next, and it succeeds or fails within seconds. Betting on an underdog has no audience at all. The price was set before the bettor arrived and will not react to anything the bettor believes, because the bookmaker has already priced the crowd’s opinion into the number. The emotional content of the decision has nowhere to go, which is the structural reason one habit can be trained and the other mostly cannot.

Underdog Prices and Longshot Preference

Betting markets have shown the same pattern for decades. Bettors overvalue longshots and undervalue favorites, a tendency first flagged as an economic puzzle in 1949 and confirmed since in horse racing, football, basketball and tennis markets.

Long-run returns on favorites lose roughly 5% of money staked, while returns on longshots lose closer to 40%. Explanations divide between a genuine preference for risk, difficulty telling small probabilities apart, and the market power of a bookmaker who knows demand for long prices is insensitive to the odds.

Habits Carried Over From the Table

The habits formed at a card table follow a player everywhere. Someone who learns to fold when the price is wrong applies the same discipline to a point spread, and someone who calls because an opponent scratched their nose will take a bad number for the same reason.

People who play poker online lose the visual information entirely, so attention goes onto bet sizing and timing. A bettor looking at a price with no player in front of them works from exactly the same inputs.

Detection Rates in Controlled Studies

The evidence on reading other people is unkind. A meta-analysis of 253 studies on detecting deception found overall accuracy at 53%, barely above a coin flip, and the result held for students, psychologists, judges, job interviewers and police officers alike.

53% sets the ceiling on what any table read can be worth. Anyone convinced they detect bluffs 70% of the time is measuring confidence instead of accuracy.

Training and professional exposure did not lift accuracy in those studies at all, which is the part experienced players resist hardest. Years at a table produce familiarity with opponents and a large library of remembered hands, neither of which improves the underlying signal.

Later work identified part of the reason. People trying to decide if someone else is lying lean on their own behavior when they lie, a cue that contains no information about anyone else, while under-using statistical cues that do predict deception. Fidget when bluffing and fidgeting starts to look like a bluff in everyone else, which makes the read a description of the reader.

Arm Motion and the Failure of the Poker Face

One line of research did find a usable signal, and it came from the hands. In a 2013 study published in Psychological Science, untrained observers watched 2-second clips of professional players making bets at the World Series of Poker and were asked to judge hand quality. Clips of faces and upper bodies produced nothing. Clips of arms and hands let observers judge accurately, and the instruction to look to the arms came directly out of the results.

Players holding strong cards pushed chips forward smoothly. Bluffers moved with a small amount of awkwardness that observers picked up without being told what to look for. The effect appeared in short clips of one specific action, so it says nothing about the guesswork that fills the time between hands.

Sentiment in Underdog Markets

The pull toward the weaker side has been studied outside betting entirely. Work on why people root for the underdog points to perceived effort, identification with the disadvantaged party, and a payoff structure where a loss is expected and a win feels larger than it is.

The same work found a limit. When the stakes became personal, support for the underdog weakened and sometimes reversed. Sentiment that weakens as soon as the outcome matters personally is a poor foundation for a wager placed hours before kickoff.

Upset Frequencies in Real Data

In the men’s college basketball tournament since 1985, games won by a team at least 5 seed lines below its opponent have averaged 8.5 per year across the whole bracket, with the highest counts of 14 in 2021 and 2023 and a low of 3 in 2007. Analysis of March Madness upsets shows they concentrate in particular seed matchups.

A 12-seed beats a 5-seed about 35.6% of the time, frequent enough to be routine. An 11-seed has taken 62 wins from 6-seeds, close to 39% of those meetings.

Further down the bracket the rates fall away. 13-seeds have 33 wins from 160 first-round games and 14-seeds have 23 from 160. A 15-seed beats a 2-seed 6.9% of the time since 1985, a figure that rises to 13.6% counting only since 2013. A 16-seed has beaten a 1-seed twice in 160 attempts, or 1.25%. Every one of those teams is an underdog, and pricing a 35% event and a 1% event the same way loses money on both.

The Cost of Both Habits

The two errors compound in the same direction. Trusting a 53% read costs money one call at a time, and pricing sentiment as probability costs 40% of everything staked on long prices across a season. Neither loss announces itself, because both feel like insight and the money leaves in amounts small enough to be explained away. The correction is a fold in one case and a skipped bet in the other, which is as cheap as either mistake will ever be to fix, and the price only rises with the number of weekends spent ignoring it.