Three explanations get offered for why a game with no story, no levels and no ending is still hard to put down: variable-ratio reinforcement, the near miss, and the unfinished task. They are usually presented as a set of equally settled facts. They are not.
One of the three has been measured directly, in laboratories, with instruments. A second turns out to rest on a famous study that measured something else. The third went through a meta-analysis of fifty-nine publications in 2025, and only half of it survived.
Sorting them by evidence rather than by how often they are repeated beats another list of "psychological triggers". It also changes which one is actually about your afternoon.
Key Takeaways
- The near-miss effect is the best-evidenced of the three. When the player picked the icon, near misses were rated as less pleasant than plain misses and simultaneously more motivating, and they recruited circuitry that also responds to actual wins.
- Outcomes engineered to look like wins while being net losses produce arousal similar to real wins.
- Variable-ratio reinforcement is a real phenomenon, but the paper usually cited for it does not use a variable-ratio schedule, and the extension to browser games is an analogy rather than a measurement.
- A 2025 meta-analysis found no memory advantage for unfinished tasks: the interrupted-to-completed recall ratio came out at 0.99, with a within-subject effect size (Cohen's dz) of 0.15.
- What did survive that meta-analysis is the urge to resume, which is a different claim and a better one.
Variable-ratio reinforcement: what Skinner actually showed
Not a variable-ratio schedule, for a start.
In 1948 B. F. Skinner published "'Superstition' in the pigeon" in the Journal of Experimental Psychology. He fed hungry birds at fixed intervals regardless of what they were doing. Six of eight developed elaborate, repeatable rituals: turning counter-clockwise, thrusting the head into a corner, a pendulum swing of the head. The food was arriving on a clock, not in response to anything, and the birds behaved as though it were.
Notice what the schedule was. The food came at a fixed interval, and it was not contingent on the bird at all. That is not a variable-ratio schedule. The term belongs to the catalogue of reinforcement schedules Skinner compiled with Charles Ferster in 1957, and this famous experiment is not an example of one.
What the study does show is more useful than the label. Reinforcement that has nothing to do with your behaviour will still shape it, and the resulting habit is extraordinarily hard to stop. When Skinner cut off the food for the bird that had learned a side-to-side hop, the bird carried on: "More than 10,000 responses were recorded before 'extinction' had reached the point at which few if any responses were made during a 10 or 15 min interval."
Ten thousand repetitions of something that had never actually caused anything. Worth separating the two halves of that: "superstition" is Skinner's interpretation of what the birds were doing, and interpretations of animal behaviour get argued over. The extinction count is not an interpretation. It is a count, and it does not depend on what you think the pigeon believed.
What variable-ratio reinforcement does not explain
The leap from a pigeon in a box to a person on a browser game is longer than most articles admit.
Skinner's demonstration concerns a hungry animal and a food reward. A reflex test has no reward in that sense at all: nothing is withheld, and the only thing that varies is your own performance. Calling a fluctuating score "variable-ratio reinforcement" is an analogy, and analogies are not measurements.
The same goes for the confident version people repeat about scores, that beating your own previous best is one of the most reliable motivators there is. Plausible, widely asserted, and not what any of the studies below measured.
Reinforcement schedules have been studied in gambling research, on machines that really do dispense money unpredictably. The findings that survive from that field turn out to be about something more specific than the schedule. A timed test is the contrasting case: a loop with no reward in it whatsoever, which is exactly what makes the reinforcement account hard to apply to one.
Why a near miss keeps you playing
Because almost winning registers, in part, as winning.
Luke Clark and colleagues, writing in Neuron in 2009, ran two experiments on a simple two-reel slot task. The first collected ratings from forty undergraduate volunteers. On the trials where participants had chosen the play icon themselves, near misses were rated as "significantly less pleasant" than full misses and, from the same event, "significantly more motivating". Worse to experience and a stronger pull, at once.
The second experiment scanned a separate and much smaller group, fifteen people retained after one withdrawal and one exclusion, and showed why that is not a contradiction in the machinery. Near misses "recruited striatal and insula circuitry that also responded to monetary wins": the striatum and the insula are among the regions that light up for an actual payout. A near miss is registered partly as a win by the reward system while being reported as a disappointment by the person having it.
Agency mattered, and the authors tested it directly. Control and outcome interacted on how much participants wanted to keep playing, at F(2,78) = 6.50, p = 0.002. That is where a sense of control appears to act.
A related finding from the same field is worth putting beside it. Mike Dixon and colleagues, in Addiction in 2010, looked at multi-line slot machines where a spin can pay out less than it cost while still playing the full win animation. Measuring skin conductance and heart rate in forty novice players, they reported that skin conductance responses to these losses disguised as wins "were similar for wins and LDWs", with both significantly larger than for real losses. The celebration is doing the work, not the money.
One caveat, and by the standard used on Skinner above it has to be stated. Both were run on gambling tasks with real money in them, so carrying the result across to a scoreboard with nothing at stake is the same class of inference that docked variable-ratio reinforcement above: a well-measured mechanism, an unmeasured transfer.
Do unfinished games really nag at you?
Less than the internet believes.
The claim traces to Bluma Zeigarnik in 1927, who reported that people recalled interrupted tasks better than completed ones. It is now quoted constantly as settled, usually to explain cliffhangers, progress bars and half-finished runs.
In 2025 Rafael Ghibellini and Beat Meier published a meta-analysis of the literature in Humanities and Social Sciences Communications, covering fifty-nine publications. The memory result did not hold: the ratio of interrupted to completed recall came out at 0.99, with a within-subject effect size (Cohen's dz) of 0.15. Their conclusion was that there is "no memory advantage for unfinished tasks" and that the effect "lacks universal validity". They suggest the 1927 original may have depended on conditions common in that era and rare now, experimenter authority among them.
Something did survive, and it is the more interesting half. The related Ovsiankina effect, the tendency to go back and finish an interrupted task when given the chance, showed a 67 percent resumption rate. Two thirds of people, offered the chance to return to something they had been pulled away from, took it.
Applied to a game that ended badly, that is a more modest and more accurate claim: the run you abandoned is probably not haunting your memory, but you will very likely start another one.
What actually holds up, ranked by evidence
| Mechanism |
What is claimed |
What was actually measured |
Verdict |
| Near miss |
Almost winning makes you play on |
Ratings from forty people, plus fMRI in a separate fifteen, on a slot task, 2009 |
Well supported, on a money task |
| Losses disguised as wins |
Win sights and sounds work without a win |
Skin conductance in forty players, 2010 |
Well supported, on a money task, abstract only |
| Variable-ratio reinforcement |
Unpredictable reward explains game pull |
A fixed-interval, non-contingent pigeon study, 1948 |
Real effect, wrong evidence usually cited |
| Zeigarnik effect (memory) |
Unfinished tasks stay in your head |
Meta-analysis of fifty-nine publications, 2025 |
Not supported |
| Ovsiankina effect (resumption) |
You go back and finish it |
The same meta-analysis, 67 percent resumption |
Supported |
Best supported are the two gambling findings. Both were measured with physiological instruments, one of them with brain imaging as well, and both point the same way. If a design gives you the sights and sounds of success without the substance of it, that is doing something real to you.
Two caveats attach in fairness. Both were measured with money on the table, so the mechanism is well established and its reach into unpaid games is not. And the slot-machine arousal paper sits behind a paywall, so it is graded here on its abstract rather than its methods.
Variable-ratio reinforcement sits awkwardly in the middle. Skinner's extinction data are solid, but they come from a study that did not use the schedule the phrase names, and the extension to a scoreboard with no reward in it is an inference either way.
Weakest is the memory version of the Zeigarnik effect, which the 2025 meta-analysis found no evidence for. The resumption version is better attested and is the one to use.
Which leaves the opening question with a blunter answer than the usual articles give. None of the three mechanisms has been measured on a game with nothing at stake. Two are solidly established on tasks with money in them, one is a real phenomenon attached to the wrong famous study, and one has been substantially withdrawn. The pull is real. Its explanation is borrowed.
Not covered here, because it is a different mechanism with its own literature: the daily format, streaks and loss aversion, which the daily puzzle explainer takes on separately and which is also what the quizzes run on. Nor is the pull of an unfinished solution, a different animal again and the mechanism behind the codes and ciphers.
Frequently asked questions
What is variable-ratio reinforcement in plain terms?
It is the schedule where a reward arrives after an unpredictable number of attempts rather than a fixed number, named in the catalogue of reinforcement schedules Skinner compiled with Charles Ferster in 1957. Worth knowing: the study people usually cite for it, Skinner's 1948 pigeon paper, does not use one. That study fed birds on a fixed interval regardless of behaviour, and its striking result is a different one, a bird performing a useless learned response for more than 10,000 repetitions after the food stopped.
Is the near-miss effect real?
Yes, and it is the best-evidenced item on this list. Clark and colleagues found in 2009 that near misses the player had chosen were rated as significantly less pleasant than full misses while being significantly more motivating, and that they activated the striatum and the insula, brain regions that also respond to a real payout. Whether the player or the computer had picked the icon interacted with the outcome on the desire to keep playing, which points to a sense of control being part of the mechanism. Both findings come from a gambling task with real money in it.
What is a loss disguised as a win?
An outcome on a multi-line slot machine that returns less than the amount staked while still playing the full winning animation, so a net loss is presented as a success. Dixon and colleagues measured skin conductance and heart rate in forty novice players in 2010, and reported that skin conductance responses to these events were similar to responses to genuine wins and significantly larger than responses to plain losses. It is the clearest demonstration that the feedback, not the payout, is doing the work.
Has the Zeigarnik effect been debunked?
The memory half of it has been seriously weakened. A 2025 meta-analysis of fifty-nine publications by Ghibellini and Meier found an interrupted-to-completed recall ratio of 0.99 and concluded the effect "lacks universal validity". The authors suggest the 1927 original may have depended on conditions common then and rare now, such as experimenter authority. The resumption tendency, at 67 percent, held up much better.
Does knowing all this make a game less compelling?
Probably not. The near-miss result held in a laboratory where participants knew they were watching a simulated slot machine, and the arousal responses to losses disguised as wins were measured in players who could see the credit counter. Understanding a mechanism and being immune to it are different things. What knowing it buys you is the ability to notice when a design is celebrating something that did not happen.
For the pull of a puzzle that has one right answer rather than a score, the codes and ciphers are the other end of the spectrum entirely.
Sources
- B. F. Skinner, "'Superstition' in the pigeon", Journal of Experimental Psychology 38 (1948), 168-172. Full text in Christopher Green's Classics in the History of Psychology archive at York University; the extinction figure, the fixed-interval schedule and the counts of birds are all from this text - retrieved 10 August 2026.
- Luke Clark, Andrew J. Lawrence, Frances Astley-Jones and Nicola Gray, "Gambling near-misses enhance motivation to gamble and recruit win-related brain circuitry", Neuron 61 (2009), 481-490. Open access via PubMed Central. The two sample sizes are from Experimental Procedures (forty behavioural volunteers; seventeen scanned, fifteen retained after a withdrawal and a movement exclusion); the near-miss versus full-miss comparisons and the control-by-outcome interaction are from Results - retrieved 10 August 2026.
- Mike J. Dixon, Kevin A. Harrigan, Rajwant Sandhu, Karen Collins and Jonathan A. Fugelsang, "Losses disguised as wins in modern multi-line video slot machines", Addiction 105 (2010), 1819-1824. Citation and abstract read via PubMed; the publisher's full text is paywalled - retrieved 10 August 2026.
- Rafael Ghibellini and Beat Meier, "Interruption, recall and resumption: a meta-analysis of the Zeigarnik and Ovsiankina effects", Humanities and Social Sciences Communications 12 (2025), article 962. Open access - retrieved 10 August 2026.
- Charles B. Ferster and B. F. Skinner, Schedules of Reinforcement (1957), is named above as the origin of the term variable-ratio. Nothing here is quoted from it and it was not consulted; every Skinner claim in this article comes from the 1948 paper listed first.
