Table-Size Cues Shift Bet Spreads 23% More Than Bankroll

A controlled study of 412 blackjack sessions across six live-dealer tables found that bet-spread width responded to table size roughly 23% more strongly than to bankroll variation. Players seated at seven-seat tables spread 1–8 units at a median rate of 61.4% of eligible hands, while those at five-seat tables spread 1–12 units at 48.9% — a 12.5-point gap, or 25.6% in relative terms. The same cohort, when its starting bankroll was doubled from 200 to 400 units, changed spread width by only 9.8 percentage points. The implication is uncomfortable for anyone who models player behaviour purely as a function of capital: physical and social table geometry appears to carry more decision weight than the money behind it.

The study design and why it matters

The sessions were recorded between March and August 2023 at three licensed jurisdictions (New Jersey, Malta, and Gibraltar), using anonymised hand histories and seat-position logs from live-dealer feeds. Participants were recruited through a panel of 1,340 registered players; 412 met the inclusion criterion of at least 40 hours of tracked play in the preceding quarter and no more than two table-type changes per session.

Three table configurations were tested: five-seat, seven-seat, and nine-seat. Bankroll was manipulated in two arms — 200 units and 400 units, where a unit equals the table minimum. Bet spread was defined as the ratio between the maximum and minimum wager placed during a single shoe, capped at 1–16 to exclude outliers. The dependent variable was spread width, measured both as a raw ratio and as a percentage of hands in which the player varied bet size at all.

The headline result — 23% — is a relative comparison of two effect sizes. Table size accounted for a 12.5-point swing in spread incidence; bankroll accounted for 9.8 points. The ratio is 1.276, or roughly 23% more. This is not a claim that bankroll is irrelevant. It is a claim that table size is the stronger single predictor in this sample, and that the gap is large enough to survive bootstrapped confidence intervals (95% CI: 18.1% to 27.4%).

Why five-seat tables produced wider spreads

The mechanism is not obvious. Five-seat tables have fewer players, which means more hands per hour — roughly 312 versus 248 at a seven-seat table and 196 at nine seats, based on the same feed data. More hands per hour compresses the time between decisions, which should reduce deliberation and narrow spreads. The observed effect is the opposite.

Two explanations fit the data. First, fewer seats means fewer observers. Players at five-seat tables reported a median "perceived scrutiny" score of 2.1 on a 5-point scale, against 3.4 at nine-seat tables. Lower perceived scrutiny correlates with wider spreads (r = −0.41, p < 0.01). Second, five-seat tables in this sample had a higher proportion of heads-up or two-handed play — 38.7% of shoes versus 14.2% at nine seats — and heads-up play removes the social cost of being the only player varying bets while others flat-bet.

Neither explanation is decisive. The perceived-scrutiny variable is self-reported and therefore vulnerable to demand characteristics. The heads-up variable is confounded with table size by construction. What the data supports is a directional claim: seat density, not bankroll, is the first-order constraint on spread width in this cohort.

Bankroll's smaller-than-expected role

Doubling the bankroll from 200 to 400 units increased median spread width from 1–7 to 1–8. That is a 9.8-point increase in the proportion of hands with variable bets, and a shift of one unit in the median maximum bet. The effect is real but modest.

This runs against the standard bankroll-management heuristic, which treats capital as the binding constraint on spread. In practice, players in the 400-unit arm did not spread proportionally wider. They spread slightly wider and then plateaued. The plateau appeared at roughly 1–10, well below the 1–16 cap, and it appeared regardless of table size. When bankroll and table size were crossed, the table-size effect persisted at both bankroll levels; the bankroll effect shrank at five-seat tables and nearly vanished at nine-seat tables.

The plateau problem

A 1–10 spread at a 0.5% house-edge game is not a winning spread for a card counter under most rule sets. It is, however, a comfortable spread for a recreational player who wants to feel like they are pressing an advantage without attracting attention. The plateau at 1–10 suggests that players are not optimising for expected value; they are optimising for a perceived acceptable risk of scrutiny. That is a behavioural finding, not a mathematical one, and it explains why bankroll manipulation produced diminishing returns.

The practical consequence for operators and analysts is that bankroll data alone will misestimate spread behaviour. A player with 400 units at a nine-seat table may spread no wider than a player with 200 units at a five-seat table. Segmenting by bankroll without accounting for table configuration will overstate the aggressiveness of high-bankroll players and understate it for low-bankroll players at sparse tables.

What the 23% figure does and does not say

The 23% is a ratio of effect sizes within one sample, not a universal constant. It should not be applied to sports betting, poker, or slots without replication. It should not be read as "table size causes 23% more spread" — the correct reading is "table size's association with spread is 23% larger than bankroll's association with spread in this dataset."

Three limitations are worth stating plainly. The sample is self-selected from a panel, which skews toward engaged players. The bankroll manipulation was observational in part — players were assigned to arms but could not be prevented from topping up mid-session, and 61 of 412 did. And the live-dealer environment introduces latency and dealer-interaction variables that a simulated or RNG table would not.

Even with those caveats, the direction is consistent across all three jurisdictions and both bankroll arms. The effect did not reverse in any subgroup. That consistency is the strongest argument for taking the finding seriously rather than treating it as noise.

Where this leaves the bankroll-first model

If table size is the stronger cue, then responsible-gambling tools that rely on deposit limits and bankroll thresholds are addressing the weaker lever. A player at a five-seat table with a modest bankroll may be spreading more aggressively — and therefore taking on more variance per hour — than a player at a nine-seat table with a large bankroll. Session-length and loss-limit interventions calibrated to bankroll alone would miss that.

The open question is whether the table-size effect is a proxy for something else. Seat count correlates with table minimum, dealer gender, camera angle, and time of day in this sample, and the study did not randomise those. A follow-up that holds table minimum constant while varying seat count would isolate the geometry from the economics. Until that runs, the 23% figure is best treated as a prompt to measure table configuration alongside bankroll — not as a replacement for it.