Social Proof: Cascades, Endorsement and Online Consensus

Social proof means treating the number of people who already chose something as evidence that choosing it is sensible. A displayed count moves decisions even when nobody knows who those people were or what they knew. The response rises with the logarithm of the count, so the first handful of visible choosers does nearly all the work and the thousandth changes almost nothing.

One question decides everything here. Following a crowd can be perfectly sensible, and whether it is sensible in any given case comes down to that one question: did the people in the crowd look at the thing, or did they look at each other? The question is answerable when the crowd itself is visible. A displayed count removes every fact needed to answer that question and leaves behind only size, the one property that feels most like evidence. So the crowd is not what a decision actually runs on. A number standing in for the crowd is.

The number arrives in more shapes than most readers separate: a raw tally from strangers, one known person standing beside the thing, the mood of a visible group, or what the most comparable people happen to hold. Four mechanisms, four different things supplying the signal, four different corrections. Merge any two and the correction applied does nothing.

What is social proof, and what exactly is being treated as evidence?

Social proofTreating the number who did something as evidence that it is worth doing. is the habit of reading a quantity of prior choosers as a statement about quality. The term is set out by Cialdini in Influence in 1984, and it names something older that Asch had already measured under laboratory conditions: in Opinions and Social Pressure in Scientific American in 1955, Asch showed that a person will give an answer they can see is wrong when enough others have given it first. Asch's people could see the group. The modern version cannot, and that turns out to matter enormously.

Away from money first. Somebody is walking down an unfamiliar street at eight in the evening looking for somewhere to eat. Two places sit side by side. One has four people inside, the other has forty. The second gets chosen, and chosen without a thought. Almost nothing was actually known at the moment of choosing. Nothing was tasted. Nothing was read. None of the forty was asked why they were there. A quantity was available, and the quantity was converted into a belief about the food.

Notice the swap that just happened: a fact about how many people are present became a claim about what is on the plate. Sometimes that swap works. Forty people who each ate there before and came back are forty independent verdicts, and that is real information. Sometimes it fails completely. Forty people who each walked past, saw thirty-nine others and joined are one verdict copied forty times, and the number says only how visible the place was an hour ago.

The distinction matters for one reason. The number is identical in both situations, so looking at the number cannot tell the two apart. Forty is forty. Everything that would separate a genuine tally of independent looks from an echo of one early look lives outside the count, and the count is all that is shown.

Why does the same fifteen points cost ten times as many people every time?

Everyone accepts that a bigger count moves a choice more. Almost nobody has the shape right. The response is not proportional to the count and it does not keep climbing at a steady rate. The response is logarithmicRising by equal steps for each tenfold increase, so large counts add little.. Each tenfold increase in the count adds a fixed amount, and the counts in between add almost nothing.

The illustrative scale used for every figure below starts at a floor of 20.0 per cent, the share who choose the option with no count shown at all, and then adds 15.0 points for every tenfold increase in one plus the count. The floor and the step together give four readings simple enough to check mentally.

Read the middle column and the arithmetic is trivial. One plus nine is ten, one plus ninety-nine is a hundred, one plus nine hundred and ninety-nine is a thousand, so the three readings sit one, two and three tenfold steps above the floor. Each step adds exactly 15.0 points on a scale invented for teaching rather than measured from anything.

The useful part is not the arithmetic, it is the bill. Going from nothing shown to nine shown costs nine people and buys 15.0 points. Going from ninety-nine shown to nine hundred and ninety-nine shown costs nine hundred people and buys the same 15.0 points. The ninth person and the nine hundredth person are each worth one step, and one of them arrives with eight others behind them while the other arrives with eight hundred and ninety-nine.

The practical finding is easy to state. A displayed countA number of prior choosers shown to somebody deciding. of ninety-nine and one of nine hundred and ninety-nine sit one step apart, the same distance as nothing at all and nine. Past a few dozen, replacing a number with any other large number barely changes what happens. Most people assume a thousand is worth roughly ten times a hundred.

Social Proof vs Bandwagon Effect: which of the two is actually an inference?

Social proof and the bandwagon effect get merged constantly, and the merge is what costs the correction. Both start from the same fact, that many people chose the thing. The two do completely different things next.

Social proof draws a conclusion. Many chose it, therefore it is probably good, therefore I will take it. There is a reasoning step in there, and a reasoning step is something that can be shown to be unsound. The conclusion depended on the many having examined the thing, so pointing out that they never did makes the whole chain give way.

The bandwagon effectWanting a thing more because it is popular, rather than believing it better., named by Leibenstein in the Quarterly Journal of Economics in 1950, draws no conclusion at all. The effect says only that wanting a thing rises with how many others have it. The popularity is not evidence of quality; the popularity is the thing being wanted. Leibenstein was writing about demand curves that shift because other people bought, not about anybody being fooled.

Only one of the two contains a step that could be wrong, and that is why the same correction cannot serve both. Telling somebody under social proof that the count carries no information takes away their reason. Telling somebody under the bandwagon effect the same thing tells them something they never disputed. The bandwagon buyer wanted the thing many people have, and the remark has just confirmed that many people have it.

A household example makes the split obvious. Two neighbours buy the same phone. The first bought it because the queue outside the shop convinced her it must be the better handset. The second bought it because everybody at her college has one. Show both a review proving the handset is ordinary: the first has lost her reason, and the second has lost nothing.

Celebrity Effect: why does it need no expertise at all to work?

The celebrity effectA choice shifted by a known person's association, without any expertise involved. is the shift produced by a known person being placed alongside a thing. The effect is worth separating carefully from an expert endorsement. The two look similar on a screen and work in opposite ways.

An expert view is a claim. Somebody who has studied the subject says the thing has a property, and that statement can be examined, tested and found wrong. An expert can be disagreed with, and the disagreement has content because there is a proposition sitting between the two parties.

An association is not a claim. A known person appearing beside a thing asserts nothing about it. There is no proposition, so there is nothing to test, nothing to disagree with and no way for it to turn out false. The celebrity effect is powerful precisely because it makes no statement. Nothing that makes no statement can be checked, contradicted or shown to have been mistaken. Expertise is not merely absent from it; expertise is irrelevant to how it works.

The Palash decision log carries an instance. On 4 January Meera Sundaram opens a holding of Rs 12,00,000/- across four positions of Rs 3,00,000/- each. On 19 February a television segment names Suvarna Chemicals Limited, and that evening she adds Rs 1,00,000/- to it, taking its cost to Rs 4,00,000/- and the total to Rs 13,00,000/-. Nothing about Suvarna Chemicals Limited changed that day except that it had been named on a screen.

The cohort shows the same pull at scale. Of the 96 buys recorded across the eight quarters, 41 followed a media mention within three days, a share of 42.7 per cent. In any given week only about 11.0 per cent of the eligible list was mentioned anywhere at all. So mentions are rare and purchases cluster behind them. The clustering is a statement about where attention went and not a statement about what was worth holding.

Community Sentiment: whose mood does a visible consensus actually measure?

Community sentimentThe visible mood of a group, measuring who is posting rather than who holds. is what can be read off a group whose members can be seen expressing a view. Community sentiment seems to come with reasons attached rather than being a bare number, and that makes it feel like the most informative of the four mechanisms. In fact it is the most treacherous, and the reason is a sampling problem rather than anything about the reasons themselves.

A visible mood is produced by the people who chose to make themselves visible. People make themselves visible when they have acted and when they feel strongly, in either direction. People who considered the thing and did nothing produce no visible record whatsoever, and neither do people who acted quietly and have not thought about it since. So a visible consensus measures the population that speaks. The population that speaks is selected on exactly the two variables the reading was meant to establish, having acted and feeling strongly.

The Palash log has a clean illustration of the gap and it has nothing to do with any forum. Over the eight quarters, 9 investors complained in writing and 14 left the practice without saying anything at all. Reading only the written record gives two conclusions: that 9 people were unhappy, and that every unhappy person put it in writing. Both conclusions are wrong. The visible share was 9 of 23, or 39.1 per cent, and no amount of careful reading of the nine letters could have recovered the fourteen who wrote nothing.

Turned around, that gives the rule. A visible mood is a good measure of the mood of visible people and says nothing reliable about anybody else. A visible mood is still worth reading. The sentence it supports is a sentence about who is posting rather than a sentence about who is holding.

Peer Effect: what do the holdings around a person say about their own?

A peer effectA choice shifted by what comparable people nearby are doing. is the shift produced by what comparable people are known to hold. Hong, Kubik and Stein measured this in the Journal of Finance in 2004, finding that participation moves with the people around a person rather than being decided in isolation. Unlike a stranger's count, this signal comes with a face attached, and the face is exactly what makes it feel like better evidence than it is.

Two things are worth separating. The first is that a comparable person's holding is a fact about that person, not about the holding. A colleague at the next desk bought the Nilgiri mid-cap scheme. The purchase establishes one thing: the colleague bought it. Nothing follows about what they examined, what they were told, what else they hold, what they earn, what they owe or how long they intend to keep it. The face makes it feel like a report. A face on a data point does not attach reasons to it, and a bare count is in exactly the same condition.

The second point is sharper, and it runs the opposite way to intuition: the closer the match that makes a peer's holding feel relevant, the more of the same circumstances that peer already shares. Same employer, same income band, same stage of life is also one payroll, one sector and one set of conditions that can turn at once. The similarity that makes the signal feel informative is the same similarity that makes copying it concentrate both holdings in the same place.

The everyday version is a street of shops that all sell school uniforms because the shop at the corner did well selling school uniforms. Each shopkeeper watched a neighbour, and each neighbour was the best available evidence. Then the school changes its supplier, and the whole street finds out at the same moment that they had one piece of information between them and had counted it eleven times.

What does a displayed count strip away?

A count is not a summary of the people it counts. A count is a deletion of them. Every property that would allow the count to be graded is removed in the act of producing it, and the one property that survives is the one that feels most like evidence.

Grading a count of sixty would take five things. Who they are. Sixty specialists and sixty passers-by are not the same. The information each of them had when they chose. Whether any of them examined the thing rather than the number. Whether they still hold it. A count of people who chose is not a count of people who kept. And whether they are sixty distinct people at all, rather than fewer people counted more than once. Not one of those five survives into the number, and the number is the entire message.

When is a count of prior choosers genuine evidence?

None of this says a count is worthless. A count of prior choosers is genuine evidence under one condition, stated plainly: each counted person examined the thing rather than the count. Where that holds, the number really is an aggregation of independent looks, and aggregating independent looks is one of the strongest things arithmetic can do.

The condition is checkable and not a matter of degree, and that is why it is worth stating as a condition rather than as a caution. Sixty people who each read a document and then decided is sixty looks. Sixty people who each saw a running number and joined is one look and fifty-nine copies. The two produce the same number and different amounts of information, and the difference is not subtle once the question is asked.

The question is what came first for the people being counted. Where the process made each person examine the thing before the tally could rise, the size of the tally means something. Where the tally was visible while people decided, its size measures how visible it was. And where which of the two happened cannot be established, the number cannot be graded. Most displayed counts are in exactly that state.

How does somebody deciding for other people read a displayed count?

Everything above works the same way for a person deciding alone and for a professional deciding for others, but the professional carries one extra duty: writing down what the count was doing in the decision. Devika Rao at Palash Advisory Services Private Limited does not get to record that an investor bought something and leave it there. A record with no reason in it cannot be reviewed later by anybody.

The working question is narrow. When a number appears in a conversation, ask what it would have to be for the answer to change. If somebody says a scheme has thousands of holders and would still be interested at nine hundred and at ninety, the count was never doing the work. If the answer flips somewhere in the middle, the count is carrying the decision, and that is the moment to ask what those holders looked at.

A lender does the same thing routinely and nobody finds it strange: an application saying other banks have already lent to this borrower gets treated as a fact about the borrower's obligations, never as evidence that lending is a good idea. The other banks made their own assessments, and copying their conclusion without seeing their working would be a way of holding the same exposure with none of the analysis. A displayed count read the way a lender reads an existing loan puts the right question in place by habit.

One caution remains. A count is not evidence about the thing, and that cuts both ways: it is not a reason to choose and it is not a reason to refuse. A count is a number about people, and the only honest thing to do with it is set it aside and look at the thing itself.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, the Nilgiri mid-cap scheme and Suvarna Chemicals Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.