Authority, trust badges and expert quotes: what is supported

The badge next to the pay button

Nearly every checkout carries one. A padlock, a tick, a logo with the word “verified”. The justification offered in conference talks and agency decks is regularly the authority principle, and the evidence cited is Milgram.

That is the most overstretched piece of evidence in this entire project. Milgram showed something very strong — just not this.

What the classic studies measure

Milgram (1963) led 40 men to believe they were administering electric shocks of increasing strength to another person while an experimenter told them to continue. Twenty-six of them went to 450 volts. Bickman (1974) tested the everyday form on the street. His paper is called “The Social Power of a Uniform”, and that title is exactly what misleads: what worked was not the uniform but one particular uniform.

In Brooklyn, 153 passers-by were told by a young man to perform some trivial act — pick up a paper bag, give a stranger at a parking meter a dime, move to the other side of a bus stop. He wore either a sports jacket and tie, a milkman’s uniform or a guard’s uniform.

Between milkman and civilian the difference was not significant; only the guard stood significantly above both. Counting only the cases without a follow-up justification, the milkman at 14 percent falls even below the civilian at 19 — the guard reaches 38. Bickman draws the conclusion himself: the power lay in the type of uniform and not in the mere presence of a uniform.

For trust seals this is the most awkward line in the table. The guard’s uniform, Bickman records, superficially resembled a policeman’s — different badge, no gun. What worked there was a signal of official enforcement authority. The occupational uniform, the pure badge of belonging, did nothing. A seal in the checkout is structurally the second thing, not the first.

Burger (2009) repeated the procedure 45 years later. Seventy percent continued past the 150-volt mark, against 82.5 percent in Milgram; the difference is not significant.

Three qualifications belong with that. Burger stopped every session at 150 volts — that it would have continued to 450 is an inference, not a measurement, and he says so himself. The sample was tightly pre-filtered: a third screened out by telephone, 38 percent of the remainder by a clinical psychologist, plus everyone who had taken more than two psychology courses. And Burger reconstructed Experiment 5 from Milgram’s 1974 book, not the 1963 version cited here, in which the victim does not protest at all until 300 volts.

Doliński and colleagues (2017) ran the same shortened procedure in Poland in 2015, with 80 participants and for the first time in Central Europe. Ninety percent pressed the tenth button, against 85 percent in Milgram’s comparable condition. They too stop at 150 volts, and they too chose a different version from Burger: Experiment 2 rather than Experiment 5.

That leaves three papers testing three related but not identical setups, two of which measure only the first third of the scale. For the direction of the finding that is a good position: two countries, two independent teams, consistently high figures. What is measured is exactly that: obedience to a present authority, up to the 150-volt mark. How far people go after that, nobody has observed since the 1970s.

What they do not measure

What was measured is compliance with an instruction given by a present person in an institutional role, usually against the inner resistance of the person complying, with an immediate consequence for behaviour.

A trust badge instructs nobody. It is present the way a piece of furniture is, it holds no role in a conversation, and it demands nothing. The transfer is not a small step away from the finding but a change of subject.

That is not to say there is nothing on trust seals. There is research — it simply does not come from Milgram.

What the research on trust seals itself finds

Two papers test exactly what Milgram does not. Both are sobering.

Kim, Ferrin and Rao (2008) surveyed 468 online shoppers in two rounds: once before the purchase, once weeks later after the goods had arrived. The actual purchase is therefore among the quantities measured, not just the intention. The attitude towards trust seals was related to none of it.

A correlation runs from −1 to +1; 0 means the two quantities have nothing to do with each other. All three values sit practically there.

In the full model a single significant path appears, from the seal to perceived risk (−0.216). We do not present it here as an effect. A path resting on a zero correlation only emerges once other variables are partialled out — that is a suppression effect, and it carries no recommendation.

On top of that comes a finding that shifts the question: 73.7 percent of respondents did not know that websites are endorsed by third parties at all — in a sample where over 90 percent had bought online in the previous year.

Head and Hassanein (2002) arrive at the same place in Canada. Of 223 respondents, 52 percent recognised such seals at all; of those, 45 percent said seals influenced their purchasing decision. Neither share differs statistically from half (p > 0.05), and the authors explicitly treat both hypotheses as unsupported. Worked through, that leaves under a quarter of the sample for whom a seal means anything by their own account.

What both papers find instead agrees: it is not the seal but the traceability of the vendor. In Kim and colleagues, perceived privacy protection (0.494), information quality (0.369) and reputation (0.165) explain trust. In Head and Hassanein, respondents name in open answers reputation, a phone number that reaches someone, and a page that looks professional — one of them says he calls the number to check whether it is even in service. Another suspects the company “just paid the regulator money to put the seal on their website”.

Where the high figures come from

Much better numbers circulate in marketing material. López Jiménez and colleagues (2021) are a traceable case: there, seals raise online sales by more than 66 percent for the majority of companies. What was surveyed, however, was not buyers but 127 retailers who already carry a seal — mostly their marketing or IT managers, online and without any identity check, as the authors concede themselves. The figure is an estimate by adopters about their own decision. We therefore list the work as a market source and not as evidence. Its own overview of some 25 seal experiments comes out mixed to negative, incidentally.

Source credibility: the fitting concept

When it comes to expert quotes, testimonials and the effect of who is speaking, the relevant literature is not obedience research but research on source credibility. Pornpitakpan (2004) surveyed five decades of it.

Her finding is clearer than one would expect in this field: almost all studies of the main effect find the credible source superior, for attitudes as well as for actual compliance. Papers with the opposite result, or none, Pornpitakpan expressly calls “few”.

What is inconsistent is something else, and it is the more interesting part: the interactions. There, credibility turns into a liability in places.

When the less credible source wins

Pornpitakpan names three situations in which the low-credibility source is superior:

• Customers already hold a positive attitude towards the brand.
• The product has been on the market long enough that they have tried it — then their own behaviour serves as the evidence, not the sender.
• The claim runs against the source’s own interests.

For practice this means the expert appearance belongs at the start of a relationship, not in communication with regulars. Putting a specialist in front of satisfied existing customers argues against their own experience.

Evidence beats the sender

Hunt (1972) and McCroskey (1969, 1970) arrive at the same point: where the message contained supporting evidence, high- and low-credibility senders were equally persuasive. Pornpitakpan concludes that the presence of evidence can override the credibility effect. Conversely, unfamiliar evidence raises the persuasiveness of the low-credibility source, not of the high one.

That is the soberest reading of the whole field: a sender label stands in for an argument as long as there is no argument.

For everyday products nobody found anything

Friedman and Friedman (1979) separate by product type. Expert voices worked for products carrying high financial, performance or physical risk, such as household durables. Celebrities worked for products with high psychological or social risk, such as jewellery. For packaged everyday goods neither did anything. For low-risk products the typical consumer was the most effective sender, not the specialist.

Freiden (1984) also found that expert voices produced better quality ratings than celebrities but were less influential than the chief executive or a typical customer.

Celebrity endorsers: figures, but with sharp limits

Amos, Holmes and Strutton (2008) summarised the effect of celebrity endorsers quantitatively: 32 papers, 266 effect sizes, 185 of them significant. The work ranks the sender effects like this:

Two things stand out. The largest value is a negative one. Bad news about the person affects advertising effectiveness more strongly than any positive attribute does. And trustworthiness beats expertise — the same finding Pornpitakpan reports under the heading “better liked than right”.

Performance in their own field, the very reason someone is famous at all, correlates at practically zero and with the wrong sign.

These figures need reading more cautiously than they look, however. The authors themselves put a note under their table saying the mean r values are unweighted and should be viewed with caution; they ran the test on weighted ranks. And the most striking value rests on the thinnest basis: negative information accounts for only 3 percent of the 266 effect sizes, about eight. The authors expressly call for more research on exactly that.

Practical cases

What we deliberately do not recommend

• Invented or self-awarded badges. A seal designed by your own marketing department asserts a review that never happened.
• Milgram as justification. Supporting the authority principle with obedience experiments transfers a finding onto a different subject.
• Titles without competence. A credential from an unrelated field is a signal that works against you on closer inspection.
• Actors in white coats without disclosure. That is not a design question.

Ethics check: nudge or sludge?

Limits and criticism