attck-research

Evidence review: Framing Effects in Digital Products

Article: Framing Effects in Digital Products
Reviewed: November 2026 · Method version 2 · How these reviews work

How this review works

Each factual claim in the article was checked against the source it cites, then challenged by reviewers whose job was to find the strongest evidence against it: replication attempts, meta-analyses, later papers that cite the original work, and research from other fields that study the same mechanism. A separate reviewer weighed both sides, and a script re-checked every quoted source. Where a claim was overstated or misattributed, the article was corrected. The research below supports the article as it now reads.

Background research

The Same Decision, Seen From Two Places

Tversky and Kahneman (1981) described a decision frame as a person’s conception of the options, outcomes and odds in a choice, controlled partly by how the problem is formulated and partly by the person’s own habits and characteristics.

Checked against the full text of Tversky and Kahneman (1981); the paraphrase is faithful.

Tversky and Kahneman compared frames to vantage points: two neighboring mountains can seem to swap heights as you walk along a valley, though neither changed.

The mountain analogy is in the original paper; the valley walk is a small illustrative addition.

In the outbreak problem (600 expected deaths), 72 percent of 152 people chose saving 200 for certain; among 155 people shown the same programs in terms of deaths, 78 percent chose the gamble.

Numbers verified in the 1981 paper. Later replications (Many Labs; a 2026 study of 1,697 people) confirm the effect, while showing part of it comes from incomplete option wording.

Alchemy

Ogilvy’s Rory Sutherland argues that changing how something is perceived often beats changing the thing itself. His countdown board, which makes a train delay easier to bear by removing the uncertainty, shares framing’s lesson: what people experience depends on more than the facts.

The countdown-board example works by reducing uncertainty (Sutherland’s own explanation), so the sidebar now presents it as a cousin of framing rather than framing itself.

Three Frames That Pull Different Ways

The framing literature looked contradictory, with positive wording winning in some studies and negative in others; in 1998 Levin, Schneider and Gaeth argued the studies measured different things and sorted them into three types.

Matches the abstract of Levin, Schneider and Gaeth (1998).

People in a taste study rated ground beef labeled “75% lean” more favorably than beef labeled “25% fat.”

Attribute framing reliably shifts ratings; a 2022 food meta-analysis finds its effect on purchase intention is close to zero.

In one study of college women, a pamphlet on what they would lose by skipping breast self-examination led to more self-reported exams four months later than one stressing the benefits.

The 1987 result is real but small-sample; meta-analyses (O’Keefe and Jensen 2009; Gallagher and Updegraff 2012) find loss frames help little for detection behaviors overall.

The three push in different directions. A positive attribute frame makes a thing look better, a gain frame in a risky choice makes people cautious, and in a goal message the loss frame sometimes wins, though large reviews find the difference small and often reversed for prevention.

Goal-frame direction is the weak link: health meta-analyses find gain frames slightly better for prevention and little difference for detection.

Once people tasted the meat, the label mattered less.

Confirmed in the original abstract; a 2022 preregistered study finds familiarity also weakens attribute framing.

Attribute frames likely do their heaviest work before people have direct experience, which in a product means landing pages, plan pickers, sign-ups and first-run screens.

Presented as a reasonable extrapolation from the beef study rather than a tested finding about product screens.

Describing lung cancer outcomes as the probability of living rather than dying made surgery substantially more attractive to physicians as well as patients and students; expertise does not switch framing off.

McNeil et al. (1982) is supported by Perneger and Agoritsas (2011); Christensen et al. (1995) suggests framing is less pervasive across clinical cases, but not that expertise removes it.

A meta-analysis of 136 papers and nearly 30,000 participants found risky-choice framing effects of small to moderate size that varied with how the problem was posed, and a later reanalysis put them at a solid moderate size.

Steiger and Kühberger (2018) re-analyzed the same data and found a larger, highly reliable effect (d = 0.52), close to the Many Labs replication (d = 0.60).

The billing toggle

Tversky and Kahneman found that people judge a saving relative to its price: 68 percent would drive 20 minutes to save $5 on a $15 calculator, but only 29 percent to save $5 on a $125 one.

Pricing research suggests a dollar figure tends to land harder on an expensive plan and a percentage on a cheap one.

The dollar-versus-percent point comes from price-promotion research (the ‘rule of 100’), not from the 1981 paper; a source is now credited.

The reliability claim

On a marketing page the first is accurate and, in line with the beef study, will likely be judged more favorably, partly because a percentage near 100 hides the hours.

The uptime pair mixes a valence change with a format change (percent versus hours), so the beef study supports the direction only loosely.

Quote

Kahneman: ‘…the different representations of the outcomes highlight some features of the situation and mask others.’

Quote verified in the lecture PDF.

Choosing a Frame on Purpose

Assuming experts are immune. Physicians shifted with the wording, and your most technical buyers probably will too.

Expert susceptibility is supported in medicine; its extension to B2B buyers is untested.

Mixing frames in one comparison (one plan’s saving in percent and another’s in dollars) makes people do math, and the comparison stops being fair.

Reasonable advice consistent with price-format research; not directly tested.

Every Number Has a Neighbor

There is no neutral way to state a fact.

Supported by the 2024 Psychological Science in the Public Interest review and the information-leakage studies; a citation is suggested.

What changed in the article

These edits were made on November 12, 2026 after the review.

Alchemy: supported with caveats

Claim reviewed: Rory Sutherland argues that changing how something is perceived often beats changing the thing itself; his examples, like the countdown board that makes a train delay easier to bear, are framing used as a design tool.

Now reads: Ogilvy’s Rory Sutherland argues that changing how something is perceived often beats changing the thing itself. His countdown board, which makes a train delay easier to bear by removing the uncertainty, shares framing’s lesson: what people experience depends on more than the facts.

Why: Sutherland does argue that changing perception can beat changing the thing, and the countdown board is his example. But he explains the boards as reducing uncertainty, not as describing the same fact a different way, which is how the article defines framing elsewhere.

Three Frames That Pull Different Ways: supported

Claim reviewed: Shoppers rated ground beef labeled ‘75% lean’ more favorably than beef labeled ‘25% fat.’

Now reads: People in a taste study rated ground beef labeled “75% lean” more favorably than beef labeled “25% fat.”

Why: The beef study found exactly this. A 2022 meta-analysis confirms positive attribute frames improve ratings, though it finds little effect on intention to buy, which the article does not claim. ‘Shoppers’ is slightly loose: participants were consumers in a study, not shoppers in a store.

Three Frames That Pull Different Ways: supported with caveats

Claim reviewed: A pamphlet on what women lose by skipping breast self-examination led to more self-exams four months later than one stressing the benefits.

Now reads: In one study of college women, a pamphlet on what they would lose by skipping breast self-examination led to more self-reported exams four months later than one stressing the benefits.

Why: The cited study did find this, with behavior checked four months later. But it was one study of college women with self-reported exams, and later meta-analyses find the loss-frame edge for detection behaviors is small and shows up mainly for breast cancer screening.

Three Frames That Pull Different Ways: supported with caveats

Claim reviewed: The three types push in different directions: a positive attribute frame makes a thing look better, a gain frame in a risky choice makes people cautious, and in a goal message the loss frame can be the stronger one.

Now reads: The three push in different directions. A positive attribute frame makes a thing look better, a gain frame in a risky choice makes people cautious, and in a goal message the loss frame sometimes wins, though large reviews find the difference small and often reversed for prevention.

Why: The attribute and risky-choice directions are well supported. For goal frames, ‘can be the stronger one’ is hedged, but large health meta-analyses find loss frames rarely win and gain frames slightly win for prevention, so readers may take away a stronger rule than the evidence allows. Now cites Gallagher & Updegraff 2012, Annals of Behavioral Medicine (meta-analysis).

Three Frames That Pull Different Ways: supported with caveats

Claim reviewed: Attribute frames do their heaviest work before people have direct experience, which in a product means landing pages, plan pickers, sign-ups and first-run screens.

Now reads: Attribute frames likely do their heaviest work before people have direct experience, which in a product means landing pages, plan pickers, sign-ups and first-run screens.

Why: The idea that experience weakens framing comes from one food study plus a supportive vaccine study. Applying it to landing pages and sign-up screens is a sensible inference, but nobody has tested it there, and attribute frames tend to move ratings more than actual choices.

Three Frames That Pull Different Ways: supported with caveats

Claim reviewed: A meta-analysis of 136 papers and nearly 30,000 participants found risky-choice framing effects of small to moderate size; the effects are dependable rather than huge.

Now reads: A meta-analysis of 136 papers and nearly 30,000 participants found risky-choice framing effects of small to moderate size that varied with how the problem was posed, and a later reanalysis put them at a solid moderate size.

Why: The numbers and the ‘small to moderate’ wording match the meta-analysis. A 2018 reanalysis that corrected for publication bias put the effect higher, at a moderate size that matches a large replication, and both papers stress that the size depends a lot on how the problem is set up. Now cites Steiger & Kühberger 2018, Zeitschrift für Psychologie (meta-analytic re-appraisal).

The billing toggle: supported with caveats

Claim reviewed: So the dollar figure may land harder on an expensive plan and the percentage on a cheap one.

Now reads: Pricing research suggests a dollar figure tends to land harder on an expensive plan and a percentage on a cheap one.

Why: The calculator study shows people judge savings relative to price, but it did not compare dollar and percent wording. That comparison comes from pricing research, which does find percentages work better on cheap items and dollars on expensive ones, and which the article does not cite. Now cites Price framing: monetary vs. percentage discount format, 2023 review.

The security prompt: not supported

Claim reviewed: For a two-factor authentication prompt (a goal frame), the loss version (‘Without a second step, anyone with your password can sign in’) will often move more people than the gain version.

Now reads: “Without a second step, anyone with your password can sign in” is the loss version. Do not assume it wins. Reviews of messages that ask people to take a protective step find little difference between the two framings, with a slight edge for the gain version, and a five-country security experiment found that a message explaining how to reduce the risk moved more people than one describing the threat. So test both, and keep the fix short and clear whichever one you run. The loss version does name the threat, which matters most when the person believes the threat applies to them.

Why: Neither cited source tests a security prompt or any other protective action: is one breast self-exam study and is a typology. For actions that prevent a bad outcome, which is what turning on a second factor does, the best evidence finds no loss-frame advantage: two large meta-analyses and a Cochrane review find gain wording slightly ahead or no difference, an 84-country experiment found loss wording raised anxiety without moving intentions, and a five-country online security experiment found that a message explaining how to reduce the risk moved more people than one describing the threat. Loss wording does win in some settings, mainly screening and detection, high perceived susceptibility, and donations, so a 2FA test could still go either way, but the research does not say it will often move more people. Now cites Gallagher & Updegraff 2012, Annals of Behavioral Medicine (meta-analysis); add van Bavel et al. 2019, International Journal of Human-Computer Studies, for the security experiment.

The reliability claim: supported with caveats

Claim reviewed: ‘99.9% uptime’ and ‘almost nine hours of downtime a year’ describe the same service; on a marketing page the first will be judged more favorably, as the beef study predicts.

Now reads: On a marketing page the first is accurate and, in line with the beef study, will likely be judged more favorably, partly because a percentage near 100 hides the hours.

Why: The arithmetic is right: 0.1 percent of a year is about 8.8 hours. The positive version will likely be judged more favorably, but the pair also switches from a percentage to hours, so it is not a pure test like the beef labels, and the beef study only loosely predicts it.

Choosing a Frame on Purpose: supported with caveats

Claim reviewed: Assuming experts are immune is a mistake: physicians shifted with the wording, and so will your most technical buyers.

Now reads: Assuming experts are immune. Physicians shifted with the wording, and your most technical buyers probably will too.

Why: Physicians do shift with wording, as two large studies show. But one study found framing changed experts’ decisions on only 2 of 12 clinical cases, and no study tests technical B2B buyers, so ‘so will your most technical buyers’ is a reasonable bet rather than a finding.