Loss aversion in copy: an effect whose strength is debated

“Don’t miss out on your €120 saving”

“Don’t miss out on your €120 saving” rather than “Save €120” — you will find that advice in almost every book on copywriting. The justification sounds compelling: a loss hurts about twice as much as an equivalent gain feels good. So threaten the loss instead.

The justification has two problems, and both are bigger than they look.

The first: since 2018 researchers have seriously disputed whether the effect exists in this form at all.

The second matters more for copy. Even if it does exist, it was measured on something else entirely: on how people value gambles and trade objects. Not on how they respond to sentences.

This guide separates the two questions. The short answer first: for choices under risk, loss aversion is well supported. For your product page, it is not.

Where the “twice as much” figure comes from

Loss aversion is part of prospect theory (Kahneman & Tversky, 1979), a model of how people value gains and losses. The core in one sentence: the road from zero to minus a hundred feels steeper than the road from zero to plus a hundred.

The famous 1979 paper, however, gives no number at all. It rests on hypothetical bets answered by students and faculty, usually 66 to 95 people per problem. The authors name the limits of that method themselves.

The factor of 2 appears only in 1991 — and it comes from coffee mugs. Tversky and Kahneman gave one group a mug and asked what they would give it up for. A second group could choose freely between mug and money.

From this follows a ratio “slightly greater than two” — the authors’ own phrase. A bet with real money gave 2.5 to 1. Of that too they write that the convergence should be interpreted with caution.

There is now an overall picture. Brown, Imai, Vieider and Camerer (2024) collected every available estimate: 607 values from 150 papers.

The much-quoted 2.25 comes from a 1992 paper with 25 graduate students and unincentivised lottery choices; on this collection it is pitched too high.

The decisive line, though, is the last one. Everything lies between 0.04 and 23.46. The factor is not a constant of nature but an average across very different situations.

For online retail another figure matters anyway. Brown and colleagues point to an earlier review by Neumann and Böckenholt (2014) covering consumer brand choice alone: 109 estimates from 33 studies. There the factor is 1.49, or 1.73 in the enhanced model. The pattern is clear — the closer the task sits to a real purchase, the smaller the factor comes out. That work is reported here at second hand.

The dispute since 2018

In 2018 a paper appeared with a provocative title: The loss of loss aversion. In it Gal and Rucker argue that many findings taken as evidence can be explained without loss aversion — by inertia, by a preference for the accustomed state, by plain effort.

At the core sits a confound in the design. In the classic experiments “loss” coincides with “doing nothing” and “gain” with “acting”. Whoever keeps the mug is also staying passive. Two things are measured at once, and the result is credited to one of them.

Gal and Rucker separated the two:

Same stimuli, same population, effect gone.

A second finding in the same paper is more awkward still. To obtain fibre-optic service respondents paid $50.43, to retain the same service only $43.75 — although 85 percent of them expressly described the second situation as avoiding a loss. They called it a loss and did not act on it.

The rebuttal came two years later, and it carries weight. Mrkva, Johnson, Gächter and Herrmann (2020) analysed five samples totalling 17,720 respondents. Their core argument: the objection hits the trading tasks, not the gambles. At least ten papers offer a choice between two lotteries, neither of which is the accustomed state — and people are loss averse there too.

For a German-speaking market their first study is the most revealing. 360 car buyers in Austria, Germany and Switzerland, two interviews two weeks apart, one task with real money. For bets of €6 the factor was 1.12 — effectively one. For the same people’s trading tasks, between 1.8 and 2.7.

The same people, two types of task, values far apart. That is not measurement error; that is the finding.

And it shows for whom the effect runs stronger. Older respondents are more loss averse, respondents with product knowledge less so — and strikingly precisely: someone who knows about fuel economy reacts less on fuel economy and unchanged on seat comfort. So for regular customers the lever is smaller than for newcomers.

The dispute resolves once you sort by subject matter. Gal and Rucker’s finding hits the endowment effect — the observation that people demand more to give up an object than they would pay to acquire it. There it hits hard: Mrkva and colleagues themselves concede that this effect can be explained without loss aversion, for instance by an aversion to bad deals or by plain feelings of ownership. For choices under risk, they say, that expressly does not apply.

So the position is this: loss aversion in gambles and risky choices — well supported, on 607 estimates and 17,720 respondents. The endowment effect — shaky, as the other side itself concedes.

Two different things under one name

Now to the actual point. This is where practice most often slips.

Research on loss aversion measures how people value gambles and trades. The rule “phrase it as a loss” is a claim about the effect of sentences. That the one supports the other has never been shown.

The mechanism that governs wording belongs to a different field: attribute framing. It means the same fact described once positively and once negatively — “90% lean” against “10% fat”. That is exactly what ad copy does.

Piñón and Gambara (2005) analysed 51 papers with 13,343 participants and averaged the fields separately.

What does d = 0.26 mean? The d measures how far two groups sit apart, expressed in their own spread. By the usual rule of thumb, 0.2 counts as a small effect, 0.5 as medium and 0.8 as large. So 0.26 sits at the low end of “small”: the two groups almost overlap, and the difference shows up only in the average across many people.

And that average is deceptive. Of the 30 attribute-framing effect sizes, 27 percent point the other way, and the values run from −0.60 to 2.03. Across the four psychology papers the mean is −0.25 — there the positive wording worked better.

That justifies thinking about a single phrase. It does not justify rebuilding a pricing page around threats.

One review asks exactly our question. O’Keefe and Jensen (2007) summarised 93 studies with more than 20,000 participants on health messaging. The result: in not a single application area were loss-framed appeals significantly more persuasive than gain-framed ones. For disease-prevention messages the gain-framed ones were in fact significantly superior.

Two caveats belong with it: this work is reported here at second hand, after Gal and Rucker, and health communication is not ad copy. But it points the same way as d = 0.26 — and in one respect further: loss framing is not a lever, and sometimes it is the worse choice.

Free trials: two explanations that look alike

The free trial is the case where loss aversion is most plausible. Someone who has worked with a tool for four weeks is reluctant to give it up.

But this is exactly where Gal and Rucker’s objection bites. That someone stays may be loss aversion. It may equally mean: the data are configured, the routines are learned, and switching is work. In the numbers the two look identical.

The distinction still has consequences, because it calls for different measures:

Doing both is the safe option. Testing one against the other is the more instructive one.

Practical cases

What we deliberately do not recommend

• The factor of 2 as a quantity to calculate with. It is not a constant. Carrying a textbook number into a spreadsheet transfers a precision that does not exist.
• Sustained loss framing. Copy that threatens throughout produces resistance; see reactance.
• Invented losses. “Your basket will be deleted in 10 minutes”, when it will not.
• Loss aversion as a justification for cancellation hurdles. That people dislike giving up what they have does not justify a hard-to-find cancellation path. That is sludge.

Ethics check: nudge or sludge?

Limits and criticism