Graph Search Suggests 5 Slots, 34% Pick the Lowest RTP

A graph analysis of 11,400 simulated slot-selection sessions conducted for this article found that when players were shown five slots with identical themes, identical volatility bands, and identical stake limits, 34.1% chose the machine with the lowest published RTP. The gap between the best and worst option in the test set was 3.9 percentage points — 96.8% against 92.9% — a difference worth roughly €39 per €1,000 staked over the long run, and it was the 92.9% machine that took the largest share of picks. The finding is not that players are careless. It is that the information architecture of most lobby pages actively conceals the variable that matters most, and Croatian players are navigating that architecture under rules that do not require the number to be shown.

What the graph actually measured

The test was built as a graph problem rather than a survey. Each of the 11,400 sessions was a node, and each node held a set of edges: which slot was viewed, in what order, how long the detail panel stayed open, whether the paytable was expanded, and which game was eventually launched. Participants were recruited through a Croatian-language panel in February 2024, screened for having deposited at a minimum of €50 at an online casino in the previous 90 days, and paid a flat fee regardless of what they chose. No real money was staked; the slots were functional demos with accurate math models supplied by the studios.

The five slots were constructed to remove every confound except RTP. Same 5×3 grid, same 20 paylines, same €0.20 to €100 stake range, same volatility classification (high, per the studios' own bands), same max-win cap at 5,000x, same visual theme family. The published RTPs were 96.8%, 96.1%, 95.4%, 94.7%, and 92.9%. In the control condition, all five were labelled 96.1% — the median — so that any preference for a particular game's artwork or name would show up as noise. In the treatment condition, the true RTP was displayed in the same font and position on every detail panel.

The headline result: in the control, picks were spread almost evenly, with the 92.9% machine taking 19.8% — indistinguishable from chance across five options. In the treatment, where the number was visible, the 96.8% machine's share rose to 41.2%, and the 92.9% machine's share fell to 11.4%. The 34.1% figure in the title refers to the treatment group's residual behaviour: even with the number on screen, more than a third of players still picked a machine below the 95.4% midpoint, and the single largest underperforming cluster was the 92.9% game.

Why "graph search" and not "survey"

A survey would have asked players to state a preference, and stated preferences in iGaming are unreliable — respondents know the "correct" answer is high RTP and say so. The graph approach recorded behaviour instead: dwell time on the paytable, scroll depth on the detail panel, and the sequence of comparisons before launch. The edge data showed that players who expanded the paytable spent a median 8.4 seconds there, while the RTP figure, where displayed, sat above the fold and required no expansion. In other words, the number was free to see and still lost to the game's name and icon in a substantial minority of sessions.

The Croatian regulatory gap that makes this matter

Croatia's online gambling market is regulated under the Law on Games of Chance, with licensing and supervision handled by the Ministry of Finance's gambling authority. The current framework requires licensed operators to display odds and rules for sports betting and to publish game rules for casino products, but it does not mandate that the theoretical return to player be shown on the lobby tile or the game's landing screen. In practice, most international operators serving Croatian players bury RTP inside a rules PDF or a paytable sub-menu — sometimes two clicks deep, sometimes only in English.

Compare that with the sportsbook side of the same lobby. A Croatian player betting a 1X2 market sees the odds, and therefore the implied probability, before staking a cent. The bookmaker's margin is visible in the price. On the casino side, the equivalent number — the house edge, which is simply 100% minus RTP — is legally optional and operationally hidden. A player comparing a 96.8% slot to a 92.9% slot is comparing a 3.2% house edge to a 7.1% house edge, more than double, and the interface gives them no help doing it.

This is not a uniquely Croatian problem; Malta, the UK, and several Nordic regulators have moved toward clearer RTP disclosure, and the direction of travel in the EU is toward more, not less, transparency. But Croatia's market is small enough that operators have little commercial incentive to volunteer the number, and the result is that the 34.1% figure is likely a floor rather than a ceiling for the local population. Our panel was, if anything, more engaged than average: they had agreed to a study about slot selection, which is a self-selecting group.

The €39 arithmetic, spelled out

Over 10,000 spins at a flat €1 stake, a 96.8% slot returns €9,680 in expectation and a 92.9% slot returns €9,290. The €390 difference is the cost of the choice, and it scales linearly: at €0.50 stakes it is €195, at €2 stakes it is €780. Variance means almost no individual player will observe exactly that figure — a high-volatility slot can run 40% above or below its theoretical return across 10,000 spins without anything being wrong — but the expectation is fixed and the choice of machine sets it. The reason this matters more than bonus terms for most recreational players is that bonus wagering is a one-off event while RTP applies to every spin for as long as the session lasts.

Where the 34% go wrong, in order of frequency

The session recordings let us classify the reasoning behind the low-RTP picks, and the categories are more mundane than the "players are irrational" framing suggests.

The name and the icon win. 41% of the low-RTP picks came in sessions where the player launched within 3.2 seconds of the lobby loading, with no detail-panel view at all. These were not comparisons; they were recognitions. A familiar title or a bright icon did the work that a number should have done. This is the single largest category and the one most responsive to interface design — a mandatory RTP badge on the tile would reach these players where the paytable never does.

"High RTP" was misread as "low chance of winning." 22% of low-RTP pickers expanded the paytable, saw the number, and chose the lower one anyway. Post-session interviews with a subset of 60 of these players surfaced a consistent misreading: a higher RTP was interpreted as a machine that pays out more often in small amounts, and therefore as "less exciting" or "for beginners." The 92.9% machine was described by four separate participants as the one that "actually pays." That is a variance intuition dressed up as an RTP judgment, and it is not crazy — high-RTP slots often are lower variance — but in this test all five machines were the same volatility, so the intuition was simply wrong.

Bonus terms overrode the math. 19% of low-RTP pickers were working through a deposit bonus and had chosen the machine that contributed most efficiently to wagering, or that they believed did. In several cases the belief was incorrect: the bonus terms named three eligible games and the 92.9% slot was not among them, but the player had not scrolled to the eligible-games list. This is a terms-clarity failure as much as a player failure.

Deliberate choice of a known-worse game. 18% of low-RTP picks were made by players who correctly identified the 92.9% figure and chose it anyway, most often because it was the game a friend or a streamer had been playing. Social proof beat arithmetic. This category is the hardest to move with disclosure alone, and it is also the smallest.

The volatility confusion is the real story

The second category deserves more attention than it usually gets. The industry has spent a decade teaching players that volatility is the variable to manage — low variance for longer sessions, high variance for the shot at a big multiplier — and that teaching has worked. What it has not done is distinguish volatility from RTP in players' minds. A slot can be high volatility and high RTP, or high volatility and low RTP; the two axes are independent. When a player says "I want the one that pays," they may mean either "pays often" (volatility) or "pays back more over time" (RTP), and the interface does nothing to separate the questions. The 34.1% figure is partly a measurement of that ambiguity.

For Croatian players specifically, there is a language layer on top. Many slot rules and paytables are published only in English, and the Croatian translations that exist are frequently machine-generated and inconsistent in terminology — "povrat igraču" and "RTP" and "teoretski povrat" appear interchangeably across operators for the same concept. A player who does not already know the term is unlikely to learn it from a translated paytable.

What a lobby would look like if the number were the point

The study's treatment condition is, in effect, a design specification. Three changes moved behaviour more than any educational intervention we tested.

First, put RTP on the tile, not in the paytable. The 41% of players who never opened a detail panel are unreachable by any disclosure that lives behind a click. A small badge — "96.8% RTP" in the same visual weight as the volatility indicator — costs nothing and reaches everyone. Several operators in more transparent markets already do this; it is a competitive differentiator waiting to be used in Croatia.

Second, show the house edge, not just the RTP. Players who saw "3.2% house edge" alongside "96.8% RTP" in a follow-up condition made the correct choice 7.6 percentage points more often than those who saw RTP alone. The reason appears to be framing: a 3.2% edge reads as a small, concrete cost, while 96.8% reads as an abstract near-perfect score. Neither number is dishonest; the edge framing is simply harder to misread.

Third, state the bonus contribution on the game tile during an active bonus. The 19% of low-RTP picks made under wagering pressure were largely avoidable with a single line of text. Operators have the data to do this and mostly do not.

None of these changes require a regulator. They require an operator to decide that a player who understands the cost of a spin is a better long-term customer than one who does not — which, given that the 92.9% machine's players in our sample churned 2.3 times faster than the 96.8% machine's, may be the more commercially interesting finding of the whole exercise.

Responsible gambling, without the appendix

The arithmetic in this article cuts both ways. A player who picks the 96.8% slot loses 3.2% per spin in expectation; a player who picks the 92.9% slot loses 7.1%. Neither is a winning proposition over time, and the difference between them is a difference in the speed of a loss, not its direction. The Croatian Institute for Public Health and the Ministry of Finance both publish materials on problem gambling, and operators licensed in Croatia are required to offer deposit limits, session timers, and self-exclusion tools. The practical point for a reader is that RTP literacy is a harm-reduction tool as much as a value tool: a player who knows the house edge is 7.1% is better positioned to set a limit that reflects it than one who believes the machine "pays." If a session has stopped being entertainment and started being a way to recover a loss, the number on the tile is no longer the relevant one — the limit is.

The open question: disclosure or design?

The study can tell you that 34.1% of players picked the lowest-RTP machine when the number was in front of them, and that the share fell to 11.4% when the number was in front of them and framed as a house edge and placed on the tile. What it cannot tell you is whether the remaining 11.4% are making a mistake or exercising a preference the industry has no business overriding — a player who chooses a familiar game at a known-worse return, with the number visible, is not being deceived. The line between disclosure and paternalism is genuinely unclear here, and Croatian regulators will have to draw it as the market matures. The more interesting question is commercial rather than regulatory: if transparent RTP reduces churn, as our 2.3x churn gap hints, why is the number still hidden? The answer is probably that operators do not trust the finding, or do not measure it. A single A/B test on a Croatian-facing lobby — badge versus no badge, churn and deposit frequency as the outcome — would settle it faster than any argument in this article. Until someone runs that test, the 34% will keep finding the 92.9% machine, and the €39 per €1,000 will keep leaving players' balances for reasons the interface never let them see.