Scarcity

Barton, Zlatevska and Oppewal pooled 416 effect sizes from 131 studies. Scarcity cues raise purchase intentions by δ = 0.31, and move the inferences behind that more: willingness to pay 0.47, perceived quality 0.47, perceived value 0.37, perceived popularity 0.71 — while attitude towards the product shifts by only 0.15. Lynn's earlier review of the commodity-theory literature put scarcity's effect on perceived value at r = 0.12, with wide variation between studies. The direction is consistent and the size is modest: a scarcity cue is worth a few percent on a page, not a redesign around it.

Which cause you name matters less than the folk version claims. Supply-based scarcity (few were made) moved purchase intentions most at δ = 0.39, time-based (the offer ends) 0.34 and demand-based (many were sold) least at 0.28, and the three were not significantly different (Q(2) = 1.76, p = .414). The paper's abstract still offers a frame-per-product rule — demand for utilitarian products, supply for experiences, time limits for high involvement — and only the first of the three cleared its own test: among utilitarian products demand framing returned δ = 0.41 against −0.02 for supply and 0.02 for time (Q(2) = 12.38, p < .01). The other two rest on marginal omnibus tests (experiences p = .083, high involvement p = .060) in which supply framing beat demand framing but could not be separated from a deadline, and for material goods and low-involvement products the framings did not differ at all. The product mattered more than the frame: low-familiarity brands showed about twice the effect of high-familiarity ones (δ = 0.41 against 0.21) and high-involvement products more than low (0.38 against 0.28). Discount figures from within-subjects designs: they returned δ = 0.48 against 0.11 for mixed designs.

The honest limit is what was measured. Of the 416 effect sizes, 384 came from hypothetical situations; only 24 measured actual behaviour and only 6 came from field studies. The country spread is uncomfortable too: δ = 0.70 in Korea, 0.49 in Canada, 0.28 in the USA, 0.27 in Taiwan, and non-significant in both Germany (0.05) and the Netherlands (0.19). Treat the pooled figures as a direction worth testing on your own interface, not as a lift to forecast.

The cleanest evidence from a live interface is a randomised field experiment on Amazon lightning deals, which display the share of inventory already claimed as a bar. Cui, Zhang and Bassamboo spiked that bar by adding items to carts, raising the claimed figure by an average 16.48 percentage points, and measured 3.42% more cart add-ins over the next hour — a 2.08% increase per 10% rise in the bar. The treated deals were far from stockout, so the mechanism is mostly observational learning from other buyers, and the authors write that they “cannot completely disentangle the herding effect from the scarcity effect.” A progress bar on a deal is as much social proof as scarcity, and it buys single digits.

Involvement in the category decides whether the cue lands at all. Van Herpen, Pieters and Zeelenberg showed participants a shelf holding one fully stocked wine and one barely stocked one, the gap explained as either high demand or limited production. The scarce wine won overall — preference 6.06 against 5.60, and 185 of 276 participants, 67%, chose it — but uninvolved participants showed none of the predicted pattern, and among involved ones the supply-caused version only worked for those with a high need to be unique in creative ways, while demand-caused scarcity worked regardless. As they put it, the scarcity principle “has been observed in various contexts, and thus seems universal, but it is not.” The spoken explanation changed nothing on its own, because the empty facing had already said it. A screen has no empty facing, so the sentence is the only cue there is.

The mechanism usually asserted for deadlines — that urgency degrades deliberation — did not appear in the one carefully incentivised test of it. Sugden, Wang and Zizzo ran price-search tasks against matched, untimed lottery choices: no regret effects, time-limited offers more likely to be accepted when more time was available rather than less, and timed choices consistent with their untimed equivalents. Their own conclusion is that the gain to retailers comes “not by using time constraints to impair the rationality of consumers' decision-making” but from search deterrence — “a restriction of consumers' opportunities to compare available offers” — amplified by ordinary risk aversion, with subjects taking the offer at a standardised probability of 0.590 and 0.645 where a risk-neutral agent would have been indifferent. The CMA's 2022 online choice architecture paper reads the same study as evidence of harm: that buyers who took timed deals “ended up worse off than those who waited” and that “participants' behaviour did not improve with experience.” The first is the search-deterrence result restated; the second sits awkwardly beside the article, which reports no trend across tasks but a highly significant effect of previously seen good offers on willingness to accept. Harm claims are shakier still: Kristofferson and colleagues' finding that limited-quantity promotions induce aggression failed to replicate in O'Donnell and colleagues' PNAS audit, with no effect on perceived threat and none of the promotion itself. A countdown has no documented power to switch off deliberation, so it has nothing to earn beyond telling the truth about a deadline.

What actually ships is largely unverifiable. Mathur and colleagues crawled about 53,000 product pages across roughly 11,000 shopping sites and found 632 low-stock messages, 49 of them with no number at all, 47 high-demand messages of which 38 appeared regardless of which product was shown, and 393 countdown timers. Where deception could be proved, it was: 157 timer instances reset or vanished with the offer still valid, and 17 low-stock messages were fabricated — 16 decrementing on a deterministic schedule, one generating values at random on page load, 8 of the 17 using off-the-shelf Shopify plugins built for the purpose — and the detection method was a lower bound by construction. The EU's Consumer Protection Cooperation Network screened 399 online shops in 2023 and found 148 carrying at least one of three dark patterns, 42 of them fake countdown timers. The CMA's assessment of the long-run cost is the one to keep: false scarcity claims “can damage trust in markets because consumers will learn to ignore these claims, meaning that when a product is truly scarce, the seller will not be able to credibly communicate this information.” One fabricated figure converts a modest, conditional, real effect into legal exposure and a dead signal.