Psychology of RPS

Game Theory: The Mathematical Foundation

In game theory, Rock Paper Scissors is classified as a finite, two-player, zero-sum game with simultaneous moves. John Nash proved that such games always have at least one equilibrium strategy. For RPS, that's playing each throw with exactly ⅓ probability.

This equilibrium has expected payoff zero against every opponent strategy and cannot be exploited. It also does not exploit a biased opponent. A player can depart from it profitably only when the estimated bias is real, persists long enough, and outweighs estimation error.

Why Humans Can't Be Random

Random-sequence research has repeatedly found departures from mathematical randomness, including excess alternation and avoidance of runs. The size and form vary with instructions, timing, symbols, and individuals, so the careful conclusions are:

• Avoid repetition - People switch throws more often than randomness would dictate, because repetition "looks non-random." It doesn't. Your brain is lying to you.
• Overuse alternation - R-P-S-R-P-S feels random but it's a perfectly predictable cycle. You're reciting the name of the game.
• Balance short sequences - After two Rocks, people feel compelled to play something else to "even things out." Each round is independent. The universe does not keep score.

The gambler’s fallacy is the belief that an independent random process must compensate for a recent run. In RPS, the opponent may not be independent, so test that assumption instead of applying the label automatically.

Cognitive Biases in RPS

Win-Stay, Lose-Shift (WSLS)

Outcome-conditioned behavior is a useful model for repeated RPS. In the Wang et al. experiment, players’ probabilities of staying or rotating depended on the preceding result, and those individual conditional responses helped explain population-level cycles.

The result is not “all people use WSLS.” Wang et al. studied 360 university students in 60 six-person groups for 300 incentivized rounds. Dyson et al. studied 31 undergraduates against a balanced computer for 225 rounds and found switching after losses and draws; staying after wins was not statistically significant. Population, incentives, feedback, opponent, and match length all limit generalization.

Key Research

• Wang, Xu & Zhou (2014) — 360 Zhejiang University students, arranged in 60 six-person groups, completed 300 randomly paired and monetarily incentivized rounds. The authors observed population-level cycling and modeled it with win-loss-tie conditional responses.
• Cook et al. (published online 2011) — Compared sighted and blindfolded face-to-face RPS and reported automatic imitation in that experimental setup. The design supports a specific visual-motor effect, not a universal prediction for all live matches.
• Dyson et al. (2016) — 31 undergraduates played 225 rounds against an equal-random computer. Participants switched more after losses and draws; the tendency to stay after wins and the tendency to over-select Rock were not statistically significant.

What RPS Teaches Us About Decision-Making

RPS provides a compact way to compare a normative random strategy with observed sequential choices. It can reveal conditional dependence, alternation, repetition, and imitation under controlled conditions. It cannot diagnose an individual, establish a personality trait, or transfer an effect from one study population to every culture or tournament.

That makes the game useful for teaching probability and experimental design: state the opponent, incentives, information, number of rounds, outcome definition, and uncertainty. A compelling pattern becomes evidence only after a suitable comparison and enough observations.