The behavioral literature on repeated choice has a stubborn finding: people abandon strategies after losses far more readily than they abandon them after equivalent gains. Kahneman and Tversky's prospect theory predicted this asymmetry in 1979, and the intervening decades of replication have done little to soften it. What remains contested is the mechanism. Does loss aversion operate as a stable trait that biases every decision, or does it function as a situational trigger that reshapes preferences only when a specific decision node is reached? The distinction matters for anyone who studies how consumers navigate repeated purchasing decisions under uncertainty — including the small but instructive population of people who buy liquid flavour concentrates for vaping.
The Decision Node Problem
Consider what a flavour shop customer actually does across a sequence of purchases. Unlike a single-transaction buyer, the repeat customer accumulates a history: bottles finished, bottles abandoned, batches mixed, coils burned out. Each new order is a trial in the loose sense that behavioral economists use the term — a discrete choice made against a backdrop of prior outcomes. The interesting question is what happens at trial fourteen, or trial twenty, when the accumulated record contains both wins and losses.
Standard expected-utility models predict that a rational buyer updates on the marginal information: if a flavour line has produced three good bottles and one bad one, the posterior should reflect that ratio, and the decision to repurchase should follow the posterior. What the data tends to show instead is that the bad bottle exerts disproportionate pull. A single disappointing concentrate — muted at low wattage, harsh at high wattage, or simply not matching the description — can terminate a product line that had otherwise performed well.
Why the Fourteenth Trial Is Different
Early trials are dominated by exploration. The buyer has no strong priors, so each purchase is cheap information. By the early teens, though, a reference point has formed. The buyer now has an expectation about what a given brand or flavour profile should deliver, and that expectation becomes the benchmark against which outcomes are judged. This is where loss aversion does its work. A bottle that underperforms the reference point registers as a loss, not merely as a less-than-ideal gain, and losses of this kind are weighted roughly twice as heavily as equivalent gains in most published estimates.
The practical consequence is that the fourteenth purchase decision is not driven by the odds of getting a good bottle. It is driven by the salience of the last bad one.
Variable-Ratio Reinforcement and the Flavour Search
There is a second behavioral layer worth naming. B.F. Skinner's work on variable-ratio schedules showed that behavior maintained by unpredictable rewards is remarkably resistant to extinction. A pigeon pecking for food delivered on an unpredictable schedule will peck far longer than one rewarded on a fixed schedule, even after rewards stop entirely. The parallel to flavour hunting is not exact, but it is close enough to be useful.
Flavour concentrates are, in practice, a variable-ratio reward structure. The buyer cannot reliably predict from a product description whether a given concentrate will land. Reviews help, but taste is idiosyncratic, and the same bottle that delights one vaper reads as chemical or flat to another. So the search behavior — trying new lines, ordering samplers, revisiting old favorites — persists well past the point where a purely rational cost-benefit calculation would suggest stopping.
What makes this more than a curiosity is the interaction with loss aversion. Variable-ratio schedules sustain the search; loss aversion determines which specific products survive the search. A buyer may keep searching indefinitely while systematically pruning any line that produced a single salient disappointment. The result is a search process that looks persistent from the outside and ruthlessly narrow from the inside.
Competitive Play and the Social Amplification
Flavour selection is not purely private. Online communities, review threads, and mixing forums turn individual trials into shared data, and this changes the decision environment in ways that behavioral research on competitive contexts helps explain. When choices are observable, losses carry a social cost in addition to a personal one. A buyer who publicly endorsed a concentrate that turned out poorly has lost not just the purchase price but a bit of standing.
Research on risk-taking in competitive settings — the work on tournament behavior and reference-dependent preferences in contests — suggests that people become more conservative after public losses and more aggressive after public wins. Applied to flavour communities, this predicts a kind of herding around safe recommendations and a corresponding reluctance to champion anything unfamiliar. The visible result is a review ecosystem where a handful of products accumulate outsized endorsement and the long tail stays perpetually under-tested.
The Asymmetry in What Gets Reported
There is a reporting bias that compounds this. Buyers are more motivated to write about a bad experience than a merely adequate one, and the language of disappointment is more vivid than the language of satisfaction. A forum thread titled with a warning about a specific concentrate will attract more engagement than a thread noting that a product was fine. This means the shared information pool that subsequent buyers draw on is itself skewed toward loss-salient cases, which reinforces the individual-level loss aversion rather than correcting it.
What a Forward-Looking Buyer Can Actually Do
The honest answer is that loss aversion is not something a buyer can simply decide to switch off. It is a robust feature of how people evaluate outcomes against reference points, and the evidence that it can be trained away is thin. What can be adjusted is the structure of the decision.
One approach is to make the reference point explicit rather than implicit. A buyer who tracks outcomes in a simple log — product, batch, subjective rating, whether it was repurchased — converts a fuzzy sense of "this brand has burned me before" into a record that can be examined. The log does not eliminate loss aversion, but it makes the asymmetry visible, and visibility is a precondition for any correction.
A second approach is to separate the exploration budget from the repurchase budget. If a fixed portion of spending is designated for genuinely new trials, those trials are framed as the cost of information rather than as potential losses against a reference point. This is a reframing move, and reframing is one of the few interventions with decent empirical support in the prospect theory literature.
A third is to treat the fourteenth trial as a deliberate decision node rather than a habitual one. The behavioral default is to let the most recent salient outcome drive the choice. Making the node explicit — pausing, reviewing the log, asking whether the odds have actually changed — introduces a small amount of friction that can interrupt the automatic pull of the last bad bottle.
The broader point is that the flavour shop is a small, tractable laboratory for a general phenomenon. The same asymmetry that shapes which concentrates survive on a shelf shapes which strategies survive in markets, which hypotheses survive in labs, and which candidates survive in hiring. The mechanism is not mysterious, and it is not irrational in any deep sense — it is simply the way organisms weight losses against gains when the future is uncertain. Recognizing it in a low-stakes domain like flavour selection is good practice for recognizing it in higher-stakes ones, where the cost of letting a single bad trial end a productive line of inquiry is considerably larger.