In 2014, a study out of Zhejiang University in China upended the conventional wisdom about how people actually play Rock Paper Scissors. The game had been analyzed by game theorists for decades, and the standard answer was always the same: play randomly, achieve Nash Equilibrium, deny your opponent any edge. What Zhijian Wang and his colleagues found was that nobody actually does this, and the deviation is consistent enough to be exploited.
The study put 72 students into groups of six. They played rounds of RPS against each other with small financial rewards tied to wins — enough to create real stakes without distorting the behavior. The aggregate data looked clean. Each throw appeared roughly a third of the time, which is exactly what Nash Equilibrium would predict. The researchers almost stopped there.
Instead, they looked at what happened after each outcome. The pattern that emerged was precise enough to be named. Winners reliably repeated their winning throw — a reinforcement response that the brain likely learned from contexts where repeating a successful behavior made sense. Losers reliably shifted to the next throw in the cycle: Rock to Paper, Paper to Scissors, Scissors to Rock. Not randomly. Not to whatever would have beaten them. In a predictable loop.
Wang called this the conditional response, and the paper won a Best of 2014 award from MIT Technology Review — the first time a Chinese researcher in social science had earned the honor. The implication was significant: even in a game designed to be solved by randomness, humans introduce a structural bias that a paying-attention opponent can exploit.
The practical application is simple enough to explain in a sentence but hard to execute in practice: watch what just happened and play the counter to what you'd expect next. If your opponent won with Rock, expect them to throw Rock again — counter with Paper. If they lost with Paper, expect them to shift to Scissors — counter with Rock. Across thirty throws, this reads better than chance. In a best-of-three, you're working with thinner data, but the same principles apply.
The study also explains something that serious players noticed empirically long before anyone published it in a journal: the pattern breaks down when people know about it. The moment you tell a skilled player about win-stay lose-shift, they start fighting their own tendencies, which introduces its own kind of unpredictability. The meta-game that emerges — I know you know I'll repeat, so I'll switch, so you should throw Scissors, so I'll throw Rock — is the same recursion problem that appears in every prediction-based game. At some point you're just picking, and the study becomes less useful.
But most players you'll face in a tournament aren't there yet. For them, the conditional response is running on autopilot, and the Zhejiang study is basically a cheat sheet.