The eighty-first article was about a loss that arrives when nobody has been wrong. This one is about the story people tell afterwards, and about a calculation that did not come out the way we expected.
Key Takeaway
A 2020 meta-analysis of 55 experiments reports that "recent research has produced inconsistent evidence for this effect" and that its own moderator was "confounded by year of appearance, such that older studies reported larger effects."[1] On our own arithmetic, if skill explains half the variance in business outcomes, one in eight operators in the worst quarter was still better than average. We expected to show that inference never works. It does, for extreme outcomes, and we report the correction.
The Verdict, Stated First
Five claims, in descending order of confidence.
One. The hypothesis itself is clearly stated and old. A 1978 review defines it as the need to believe the environment is a place where people usually get what they deserve.
Two. Its central experimental effect has a decline problem, and the meta-analysts said so themselves. Older studies reported larger effects, and the proposed moderator was confounded with publication year.
Three. Those meta-analysts then did the right thing, running two fresh experiments to disentangle the confound rather than leaving it in the discussion section.
Four. Blaming is not the only response to an unjust outcome. A separate literature reports that helping is the alternative, and that which one you reach for depends on whether you are focused on yourself or on the other person.
Five. On our own arithmetic the inference from outcome to character is badly wrong for ordinary failure and roughly right for catastrophic failure, which is not the conclusion we set out to reach and is the one the numbers support.
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A for the meta-analytic findings, from an abstract obtained verbatim from an open-access repository, including its own statement of the confound.
Grade A for the definition and for the 1978 review's self-criticism, from an abstract obtained verbatim from a university research record.
Grade A for the description of the 1966 experiment, which reaches us through the 2020 meta-analysis's own account of it rather than from the paper.
Grade B for the self-focus finding, from an abstract obtained verbatim, reporting two studies whose sample sizes we do not have.
Grade A for our own arithmetic, which is exact and which we verified two ways after it produced a result we did not expect.
Our position: a real phenomenon with a documented decline problem, handled unusually well by the people who documented it.
A Note On Method
Everything here is verified to August 2026.
We obtained the 2020 meta-analysis abstract and body passages verbatim from an open-access repository[1], including its description of the original experiment and its statement of its own confound.
We obtained the 1978 review's abstract verbatim from a university research record[2].
We obtained the self-focus paper's abstract verbatim from its publisher[4].
We did not obtain the 1966 paper, the 1980 book, or any of the scale-validation papers, and every statement about them reaches us through other sources' descriptions or through reference lists[3].
All arithmetic is ours. The mathematics is exact and we verified it by closed form and by simulation; the input assumptions are illustrative and no source supplies them.
This article discusses research on attribution and outcomes. It is not psychological, legal or credit advice.
The Hypothesis
The claim, in a review's own words.
A 1978 review states it: "The just world hypothesis states that people have a need to believe that their environment is a just and orderly place where people usually get what they deserve."[2]
Four observations, ours.
The word doing the work is "need." The claim is not that people mistakenly believe the world is fair, but that they are motivated to believe it, which predicts that the belief will bend evidence rather than follow it.
"Usually" is a hedge worth noticing. The hypothesis does not require anyone to think outcomes are perfectly deserved, only that they broadly are, which is a much more common and more defensible-sounding position.
And the prediction that follows is uncomfortable and specific. If you need to believe outcomes are deserved and you see someone suffering undeservedly, one available move is to decide they deserved it, which is the effect the literature has spent sixty years measuring.
It is worth saying plainly that this is a claim about observers, not about victims, and the business application below concerns how owners and advisers judge other people's outcomes.
The 1966 Experiment
The founding study, described by the meta-analysts who later assessed the literature it started.
They record: "In the first experimental demonstration of this phenomenon, Lerner and Simmons asked participants to view a live video feed (in reality a recording) of a confederate completing a learning task and receiving electric shocks for responding incorrectly, to which she reacted with expressions of pain and anguish. They found that participants who believed that the learner would continue to receive painful shocks in a subsequent phase of the experiment evaluated her character less favorably (e.g., as less likable) than did participants who learned that her ordeal had ended."[1]
The citation: Lerner, M. J., and Simmons, C. H. (1966), Observer's reaction to the "innocent victim": Compassion or rejection?, Journal of Personality and Social Psychology, 4, 203–210[3]. We did not obtain it.
Four observations, ours.
The manipulation is elegant and worth understanding. Both groups saw identical suffering; the only difference was whether it was described as over or as continuing.
That isolates the mechanism the hypothesis predicts. Suffering that has ended poses no ongoing threat to a belief in a fair world; suffering that will continue does, and it is the continuing case where derogation appeared.
The deception involved is substantial and would face a different ethics review today. Participants believed they were watching a real person being shocked, which is the same apparatus that made the obedience studies of that decade famous and contested.
And the measured outcome is character evaluation, not behaviour. Participants rated her as less likeable, which is a real finding and a narrower one than "people blame victims."
Why It Is Puzzling
The meta-analysts state the puzzle better than we would.
They write: "the capacity for people to derogate an innocent victim is puzzling because, by any commonly accepted standards or norms, people ought not to devalue someone's character for negative outcomes brought about by chance or factors beyond their" control[1].
Three observations, ours.
The puzzle is what makes the finding worth studying. An effect that everyone would endorse on reflection needs no explanation; one that people would disavow if asked does.
The framing also sets the standard for evaluating the business version. The question is not whether outcomes ever reflect character, but whether we adjust for the part that does not, and that is a question with a numerical answer.
And the phrase "factors beyond their control" is the whole of the arithmetic below, which asks how much of a business outcome that phrase covers.
A 1978 Review Already Flagged Problems
Twelve years in, the literature was already being told it had methodological trouble.
The same 1978 review's abstract closes: "Finally, recurrent conceptual misinterpretations and methodological errors found in the literature are identified."[2]
The citation: Lerner, M. J., and Miller, D. T. (1978), Just world research and the attribution process: looking back and ahead, Psychological Bulletin, 85, 1030–1051[3].
Four observations, ours.
The review's first author is the same person who ran the 1966 experiment, which makes the self-criticism more notable rather than less.
"Recurrent" is the word that matters. Not isolated errors, but a pattern the review considered worth naming in an abstract.
It also means the concerns raised in the 2020 meta-analysis are not a new discovery about an unsuspecting field. This literature has been examining itself for close to fifty years.
And it fits a pattern this series has now seen repeatedly. The people closest to a finding are usually the first to publish its problems, and the popular version is the one that never hears about them.
One further point about the venue, which is not incidental. Psychological Bulletin is the field's review journal, so this was not a critic's note in a minor outlet but the standard place a discipline takes stock of itself.
Which makes the persistence of the simple version harder to excuse. The caveats were published in the most visible possible location and the popular account of the finding has never carried them, which is a pattern rather than an accident.
Inconsistent Evidence
The modern position, stated by the meta-analysis in its opening line.
It records: "Research during the 1960s found that observers could be moved enough by an innocent victim's suffering to derogate their character. However, recent research has produced inconsistent evidence for this effect."[1]
Four observations, ours.
That sentence is doing something specific and honest. It separates the decade the effect was found in from the decades since, which is precisely the shape a decline effect takes.
The word is "inconsistent" rather than "failed to replicate," and the difference matters. Inconsistent means sometimes yes and sometimes no, which is what you would see if a real effect depended on conditions nobody had pinned down.
Which is exactly the hypothesis the meta-analysis then goes on to test, and this is what a healthy literature looks like. Notice the inconsistency, propose a moderator, and test it, rather than defending the original.
And it is the same structure as the seventy-fifth article's dispute, with one difference. Here the people investigating the inconsistency are not the original authors' opponents, and the tone of the resulting paper reflects that.
The 2020 Meta-Analysis
The source.
Dawtry, R. J., Callan, M. J., Harvey, A. J., and Gheorghiu, A. I. (2020), Victims, Vignettes, and Videos: Meta-Analytic and Experimental Evidence That Emotional Impact Enhances the Derogation of Innocent Victims, Personality and Social Psychology Review, 24(3), 233–259, April, DOI 10.1177/1088868320914208[1].
Its design: "We conducted the first meta-analysis (k = 55) of the experimental literature on the victim derogation effect to test the hypothesis that it varies as a function of the emotional impactfulness of the context for observers."[1]
Its first result: "We found that studies which employed more impactful contexts (e.g., that were real and vivid) reported larger derogation effects."[1]
Three observations, ours.
The first meta-analysis, in 2020, of a literature beginning in 1966. Fifty-four years of experiments before anyone pooled them, which is worth sitting with.
k = 55 is a substantial body of experiments, and the moderator tested is a sensible one: whether the suffering felt real to the observer.
And the finding as stated would tidy the literature nicely, which is exactly when a careful reader should look for the catch. The authors found it themselves.
The Confound
The sentence that makes this article worth writing.
The abstract continues: "Emotional impact was, however, confounded by year of appearance, such that older studies reported larger effects and were more impactful."[1]
Four observations, ours.
Read carefully, this says the tidy explanation could not be distinguished from a plain decline. Older studies were both more vivid and more successful, so the moderator and the calendar move together and the meta-analysis alone cannot separate them.
That is a serious problem for the moderator hypothesis and the authors put it in the abstract, which is the least convenient place available to them.
The reason older studies were more vivid is not mysterious, and it points at something outside psychology. Ethics review has made the vivid manipulations of the 1960s largely unavailable, so the shift from live shock apparatus to written vignettes is real and dated.
Which produces a genuinely difficult identification problem. If the field's methods changed at the same time as its results, no amount of pooling old studies will tell you which caused which, and the honest response is to run something new.
Two readings remain open and the meta-analysis cannot choose between them, which is worth stating plainly because both are live.
Either the effect is real and requires vivid conditions, in which case modern vignette studies were simply too weak to find it and the 1960s work stands. Or the effect has declined for the reasons effects usually decline, and vividness is a story fitted to the pattern afterwards.
What They Did About It
The response, which is the part we want to hold up.
The abstract: "To disentangle the role of emotional impact, in two primary experiments we found that more impactful contexts increased the derogation of an innocent victim."[1]
Four observations, ours.
They ran new experiments because the meta-analysis could not settle it, which is the correct move and an uncommon one.
The logic is clean. In a fresh experiment, emotional impact can be manipulated while the year is held constant, which breaks exactly the confound the pooled data could not.
We did not obtain the experiments' details, so we cannot tell you their sample sizes, their effect sizes, or whether they were preregistered, and those would all matter.
But the structure deserves credit regardless. A paper that reports a confound in its own headline finding and then goes and tests around it is doing the job properly, and this series has covered enough literatures that did not.
Blaming Or Helping
A separate finding that changes the practical picture considerably.
A paper on reactions to innocent victims states: "Reactions toward innocent victims can range from harsh derogatory reactions to great effort to alleviate the victims' ill fates."[4]
And concludes: "these findings extend previous studies on just-world theory and show that both blaming and helping can be viable strategies to deal with unjust situations."[4]
Four observations, ours.
This reframes the whole thing. Derogation is not the response to an unjust outcome; it is one of two, and both restore the sense that the world is manageable.
The mechanism is the same in both cases, which is why they can substitute for each other. Helping makes the outcome less unjust by changing it; blaming makes it less unjust by reinterpreting it, and either resolves the discomfort.
That is a considerably more useful finding for a business reader than the derogation effect alone. The question is not whether you will respond to a failure, but which of the two you will reach for, and the next section says what determines it.
And it also explains why the derogation effect might be inconsistent. If two responses are available, which one a study elicits depends on its setup, and a study offering no way to help may find more blaming than one that does.
The Self-Focus Finding
What determines which response you reach for, on this paper's account.
It proposes: "self-focused versus other-focused motives can evoke derogatory versus more benevolent reactions, respectively, toward innocent victims."[4]
And reports: "By manipulating self-focus versus other-focus, we indeed show in two studies that a self-focus enhanced indirect victim blaming and derogation and decreased helping of innocent victims. Furthermore, when participants were focused on another person these effects attenuated."[4]
Four observations, ours.
The variable is manipulable, which is what makes it useful. Focus is a state rather than a trait, and the studies moved it experimentally.
The direction is intuitive once stated and worth making explicit. Blaming protects the observer, because if the failure was deserved then it will not happen to you, and self-focus is exactly the condition under which that protection is wanted.
Two studies with sample sizes we do not have, in a specialised journal, is a modest evidence base and we grade it accordingly.
And a related citation confirmed elsewhere points the same way. A paper titled "We Blame Innocent Victims More Than I Do" reports that self-construal level moderates responses to just-world threats[5], and we obtained only its title.
What Actually Survives
Our reading, stated directly.
Five statements.
The hypothesis is clearly defined and the original experiment was well constructed, holding the suffering constant and varying only whether it would continue.
The derogation effect has a decline problem, with recent evidence described as inconsistent and older studies reporting larger effects.
The proposed explanation is confounded with publication year, and the meta-analysts said so in their own abstract and ran fresh experiments to get around it.
Blaming is one of two responses, the other being helping, and which one appears depends at least partly on whether the observer is focused on themselves.
And none of this tells you how much a bad outcome actually reveals about a person, which is the question a business reader has and which the rest of this article computes.
The Inference Underneath
Turning the psychology into a question with a numerical answer. Ours.
Four observations.
Strip the emotion out and the just-world move is an inference from an outcome to a characteristic. They failed, therefore they were bad at it.
That inference is not automatically wrong. Outcomes do carry information about skill, and a series that spent the seventy-fourth article defending the reality of expertise cannot pretend otherwise.
So the useful question is quantitative. Given how much of a business outcome is skill and how much is everything else, what fraction of failed operators were actually below average?
And that has an exact answer, which we computed and which did not come out the way we expected.
What A Bad Outcome Tells You
Our own arithmetic. Suppose an operator's outcome is their skill plus everything else, and skill explains some fraction of the variance in outcomes. Of the operators who ended in the worst quarter, how many were nonetheless above median in skill?
If skill explains 10 percent of outcome variance: 33.8 percent of the failed were above-median operators.
At 20 percent: 26.9. At 30: 21.5. At 40: 16.7. At 50: 12.5. At 70: 5.2. At 90 percent: 0.3 percent.
Four observations.
If skill explains half the variance in business outcomes, then one in eight of the operators in the worst quarter was better than average at running a business.
At a tenth, it is one in three, and a great many people would put business outcomes closer to that end than to the other.
No source gives us this number for business, and the fractions above are illustrative rather than measured. What the table provides is the shape of the relationship, not a reading of reality.
And the mathematics is exact. We verified it two ways, which turned out to matter.
A Correction We Owe You
This section exists because the arithmetic disagreed with the section we had planned.
We built a further table asking how extreme an outcome has to be before the inference becomes reliable, expecting to show that it never does. That is a satisfying conclusion and it is the one this article was going to reach.
The numbers came back suspiciously clean: at 50 percent skill variance the share of above-median operators was exactly half the percentile, every time. That pattern usually means a bug.
It was not a bug. At exactly 50 percent skill variance the two terms in the conditional distribution are equal, the expression collapses, and the answer is exactly half the percentile as an algebraic identity. We confirmed it against a simulation of four million draws, which returned 12.53 against a predicted 12.50, 5.02 against 5.00, and 0.51 against 0.50.
Which meant our planned conclusion was wrong. The inference does get clean: in the worst one percent of outcomes only half a percent of operators were above median, or one in two hundred. A draft sentence claiming one in a hundred was also wrong, from rounding a figure we had not checked.
Three observations, ours.
The honest finding is more useful than the one we wanted, and the next section states it.
The near miss is worth naming. We came close to publishing a conclusion the arithmetic did not support, and what caught it was checking a result that looked too tidy rather than accepting it.
And this is the sixth self-correction documented in the body of an article in this series, which we record for the same reason as the first. A correction you cannot see is indistinguishable from never having made the error.
An Exact Rule
The identity, stated in a usable form. Ours, exact at 50 percent skill variance.
The share of above-median operators among the failures is half the percentile. So:
The worst 50 percent of outcomes: 25.0 percent were above-median operators, or one in four.
The worst 25: 12.5, or one in eight. The worst 10: 5.0, or one in twenty. The worst 5: 2.5, or one in forty. The worst 2: 1.0, or one in a hundred. The worst 1 percent: 0.5, or one in two hundred.
Three observations.
The rule holds only at exactly 50 percent skill variance, which is a specific assumption and one nobody has measured for business.
It is nonetheless a useful anchor, because it is easy to carry. Halve the percentile, and you have the error rate of the just-world inference under a middling assumption about how much skill matters.
And the relationship is proportional, which is the substantive point. The commoner the outcome, the worse the inference, and ordinary business failure is very common indeed.
Ordinary Failure And Catastrophic Failure
The conclusion the arithmetic actually supports. Ours.
Four observations.
For ordinary failure the just-world inference is badly wrong. A quarter of businesses doing worse than most is a common event, and one in eight of those operators was above average.
For catastrophic failure it is roughly right. An outcome in the worst one percent is unlikely to have been produced by a good operator, and pretending otherwise would be its own kind of dishonesty.
That asymmetry is the practically important part and it cuts against the way the judgement is usually made. People are most confident about the visible collapses and most casual about the quiet underperformance, and the arithmetic says the confidence should run the other way.
And the reason is worth stating in one line. Extreme outcomes are hard to produce by luck alone, while ordinary bad outcomes are produced by luck constantly, which is why the two cases deserve different treatment.
One qualification on the catastrophic case, because stated flatly it is too strong. The arithmetic assumes luck is normally distributed, and business luck plainly is not: a fire, a lawsuit, a lost anchor client or an illness can move an outcome further than any normal distribution predicts.
Fat tails cut in the direction of caution. If genuinely catastrophic luck is more common than the model allows, the worst outcomes contain more competent operators than our table shows, and the one case where the inference looked safe is the case where our assumption is weakest.
The Same Error About Winners
The mirror image, which follows immediately. Ours.
Four observations.
The distribution is symmetric, so every figure above applies in reverse. At 50 percent skill variance, one in eight of the operators in the best quarter was below median in skill.
Which means we are exactly as wrong about the winners as about the losers, and this is the half of the error nobody complains about.
It also connects directly to the fifty-eighth article. Survivorship bias is this same arithmetic seen from the other end, and the reason business advice drawn from successful firms disappoints is contained in the number above.
And it makes one practice indefensible on arithmetic grounds alone. Studying the winners to learn what works is sampling on the outcome, and at these figures a meaningful fraction of the sample is there by luck.
Two consequences worth drawing out, because the symmetry is where most of the practical damage happens. The winners are the ones who get asked, on stages and in case studies and in books, and the losers are not available for interview.
Which compounds the error rather than merely repeating it. Every winner interviewed produces an explanation, and the explanation is generated after the outcome is known, which is the mechanism the fifty-third article documented in detail.
How Much Skill Would Have To Matter
The last computation, and the one that puts a floor under the judgement. Ours.
To be 90 percent confident that a bottom-quarter outcome means a below-median operator, skill would have to explain 56 percent of the variance in business outcomes, which is a correlation of 0.75 between how good you are and how you do.
Four observations.
A correlation of 0.75 between operator quality and business outcome would be extraordinarily high for any social science relationship.
For comparison, this series has covered several literatures where the celebrated correlations were a fraction of that, and none of them involved anything as noisy as a small business in a real economy.
So the standard required for confident judgement is a good deal higher than most people's intuition about how much skill matters, and that gap is the space the just-world inference operates in.
And we would flag the assumption again, because it carries everything. This treats skill and luck as independent and normally distributed, and reality is neither so tidy nor so kind.
What The Arithmetic Does Not Say
Guarding against the misuse this calculation most invites. Ours.
Four observations.
It does not say that skill is unimportant. Every figure in the table assumes skill matters, and the whole exercise is about how much information an outcome carries, not about whether competence exists.
It does not say that anyone should be excused anything. A supplier who is not paid is not paid regardless of why, and a business that fails has failed whatever the cause.
It does not provide the fraction of business outcome variance that skill explains, which is the input everything depends on and which we could not find measured anywhere.
And it does not license the reverse error, which is the one this article could easily encourage. Treating every failure as bad luck is exactly as unfounded as treating every failure as bad management, and the arithmetic gives no support to either blanket position.
Two Questions That Come Apart
The distinction that makes this usable rather than merely interesting. Ours.
Four observations.
There is a question about what happened and a question about who it happened to, and the just-world move answers the second using evidence about the first.
For most business purposes the first question is the one that pays. Why did this firm run out of cash is answerable and useful; was the owner any good is neither, for ordinary outcomes.
The habit worth building is to notice when an explanation has crossed from one to the other. An account that ends in a character trait has stopped being diagnostic, because character is not a thing you can inspect and adjust for.
And that is the practical case against the inference, independent of whether it is fair. It terminates the analysis, and the analysis was the point.
Judging A Competitor
The first application. Ours, and not advice about anyone's business but your own.
Four points.
When a competitor closes, the available explanation is usually about them. They overexpanded, they were badly run, the owner lost interest, and one of those may well be true.
The arithmetic says to hold it loosely for ordinary failures. At a middling assumption, one in eight of the firms in the worst quarter was better run than yours, and you will not be able to tell which from the outside.
The practical cost of getting this wrong is not moral, it is informational. A competitor who failed for reasons outside their control is telling you something about your market, and an explanation that terminates in their character means you stop looking.
And the check is simple and worth running. Ask what would have to have been true for a well-run firm to have had that outcome, and see whether any of it applies to you.
One version of that check is concrete enough to run this week. Write down the three explanations you would give for a competitor's closure, then mark which of them you could verify from outside their business.
In our experience most such lists survive that test badly, and the ones that do not survive are the character explanations. What you can actually observe is usually the market, the pricing and the timing, which are the parts that also apply to you.
Judging Your Own Failure
The second application, and the harder one. Ours.
Four points.
The same arithmetic applies to a bad year of your own, in both directions. An ordinary bad outcome is weak evidence about your competence, and a catastrophic one is stronger evidence.
The failure mode here is the opposite of the one above. People are often harsher on themselves than the arithmetic supports for ordinary setbacks, and more forgiving than it supports for the severe ones.
And the useful discipline is to separate the two questions, which people routinely merge. Was the decision good given what was knowable, and was the outcome good, are different questions with different answers, and only the first is about you.
We would add the boundary this publication always adds. If a business setback is affecting your health, sleep or relationships, that is a matter for a doctor rather than for a series of research articles, and it is a common and treatable thing.
The Business Advice Problem
The third application, and it implicates writing like this one. Ours.
Four observations.
A great deal of business advice takes the form successful firms do X, therefore do X, which is an inference from outcome to characteristic run in the favourable direction.
The arithmetic above says what is wrong with it. At a middling assumption, one in eight of the studied winners is there by luck, and whatever they were doing enters the recommendation alongside everything that worked.
The implicit converse is the part that does damage, and it is rarely stated aloud. If success comes from doing X, then people who failed did not do X, which is the just-world hypothesis wearing a business jacket.
And the honest version is available and less marketable. Compare firms that succeeded with firms that failed doing the same things, which is the design the fifty-eighth article described and which almost nobody runs.
One defence of the genre is fair and we would make it. A practice common among successful firms is at least worth examining, and demanding a controlled comparison before considering anything would leave a reader with nothing.
The reasonable position is about strength rather than existence. Treat it as a hypothesis to check against your own situation rather than as a finding, and the arithmetic above says how much weight the observation can carry on its own, which is not much.
Clients Who Fall Behind
The fourth application, which is close to home for this firm. Ours, and not credit advice.
Four points.
A customer who stops paying gets explained, and the explanations available are mostly about their character. Disorganised, unreliable, not serious, and the just-world literature predicts that judgement will feel more certain than the evidence supports.
The self-focus finding is directly relevant here and is the practical hinge. A supplier worried about their own cash is in exactly the self-focused state the research associates with blaming rather than helping, which is the state most collection conversations happen in.
That is not an argument for leniency and we would not make one. A business extending credit is entitled to be paid, and the arithmetic says nothing about what you should do.
What it says is narrower and still useful. The inference from late payment to bad character is unreliable for ordinary lateness, and a customer whose reason is genuine is a customer worth keeping if the account can be worked out.
Bibliographic Note
The series keeps a count.
One reference list gives the 1966 paper's title as "The observer's reaction to the innocent victim: Compassion or rejection?"[3], adding a definite article and dropping the quotation marks that the standard form places around "innocent victim."
Three observations, ours.
This is minor and would not defeat a search, which is why we record it as a variant rather than an error.
The quotation marks in the original title are doing real work, and dropping them changes the sense slightly. They mark the innocence as the thing in question, which is what the paper is about.
That brings the running count of bibliographic variants across this series to twenty-eight.
It is the mildest of the twenty-eight, and we would not have recorded it in the first ten articles. The count has become a measure of how carefully we are reading rather than of how careless the literature is, and that is worth admitting at this point in the series.
What To Do
Ask how common the outcome was before judging it. On our own arithmetic, at 50 percent skill variance the share of above-median operators among the failures is half the percentile, so one in eight for the worst quarter and one in two hundred for the worst one percent.
Be less confident about ordinary failure and more confident about catastrophic failure, which is the reverse of how the judgement is usually made.
Apply the same discount to the winners. The distribution is symmetric, so one in eight of the top quarter was below median, and advice drawn from winners inherits that.
Separate the decision from the outcome when reviewing your own year. Whether the decision was good given what was knowable is a different question from whether it worked.
Notice when you are self-focused. One paper reports that self-focus increased blaming and reduced helping, and a cash-flow worry is a self-focused state.
Ask what would have to be true for a competent operator to have that outcome, and check whether any of it applies to you. That is the diagnostic value an explanation about character destroys.
Treat "successful firms do X" as sampling on the outcome, and look for the firms that did X and failed before acting on it.
Do not use this arithmetic to excuse anything. It says an outcome is weak evidence about a person, not that performance does not matter, and a business still has to be paid.
The Limits Of This Analysis
Several caveats matter. This article discusses research on attribution and outcomes and is not psychological, legal or credit advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain the 1966 paper, and our description of it comes from the 2020 meta-analysis's account rather than from the study; we did not obtain the 1980 book, the scale-validation papers, or the two primary experiments in the 2020 paper, so we cannot report their sample sizes, effect sizes or preregistration status. The self-focus finding rests on two studies whose sample sizes we do not have, reported in an abstract. All arithmetic is ours. The mathematics is exact and we verified it by closed form and by a four-million-draw simulation, but it rests on assumptions that are ours alone: that outcome equals skill plus independent noise, that both are normally distributed, and that skill can be treated as a single quantity. No source supplies the fraction of business outcome variance attributable to skill, and every figure using 50 percent is illustrative rather than measured; the true figure is unknown to us and may vary enormously by industry, and the whole table moves with it. The exact halving rule holds only at exactly 50 percent and is a coincidence of that value rather than a general law. And this article contains a documented correction: we set out to show that the inference from outcome to character never becomes reliable, the arithmetic showed that it does for extreme outcomes, and we have reported that rather than quietly adjusting the framing.
Frequently Asked Questions
What is the just world hypothesis?
Does the original finding hold up?
How did the meta-analysts handle that confound?
Is blaming the only response to an unfair outcome?
How much does a business failure tell you about the owner?
Does the inference ever become reliable?
Does this apply to successful firms too?
References
- Dawtry, R. J., Callan, M. J., Harvey, A. J., & Gheorghiu, A. I. (2020). Victims, Vignettes, and Videos: Meta-Analytic and Experimental Evidence That Emotional Impact Enhances the Derogation of Innocent Victims. Personality and Social Psychology Review, 24(3), 233–259, April, DOI 10.1177/1088868320914208. Open-access repository copy reproducing the abstract: that research during the 1960s found observers could be moved enough by an innocent victim's suffering to derogate their character, but that recent research has produced inconsistent evidence for this effect; that the authors conducted the first meta-analysis, of 55 studies, of the experimental literature on the victim derogation effect, to test the hypothesis that it varies as a function of the emotional impactfulness of the context for observers; that studies employing more impactful contexts, being real and vivid, reported larger derogation effects; that emotional impact was however confounded by year of appearance, such that older studies reported larger effects and were more impactful; and that to disentangle the role of emotional impact the authors ran two primary experiments in which more impactful contexts increased the derogation of an innocent victim. The same copy reproduces body text describing the original demonstration, in which participants viewed a purportedly live video feed of a confederate receiving electric shocks for incorrect responses, and those who believed the shocks would continue evaluated her character less favourably than those told her ordeal had ended; and stating that the capacity to derogate an innocent victim is puzzling because people ought not to devalue someone's character for negative outcomes brought about by chance or factors beyond their control. Note: an open-access repository copy of a peer-reviewed meta-analysis. Our source for the modern position, for the confound the authors identified in their own work, and for the description of the 1966 experiment, which we did not obtain directly. We did not obtain the two primary experiments' details. pmc.ncbi.nlm.nih.gov
- University research record for Lerner, M. J., & Miller, D. T. (1978), Just world research and the attribution process: Looking back and ahead, Psychological Bulletin, 85, 1030–1051, reproducing the abstract: that the just world hypothesis states people have a need to believe that their environment is a just and orderly place where people usually get what they deserve; that the article reviews the experimental research generated by the hypothesis; that considerable attention is devoted to the experiment by Lerner and Simmons; that in light of the existing empirical findings an elaboration of the initial hypothesis is offered, suggesting that people's need to believe in a just world affects their reaction to the innocent suffering of others; and that recurrent conceptual misinterpretations and methodological errors found in the literature are identified. Note: a university research record. Our source for the definition of the hypothesis and for the fact that the literature's own founder was identifying recurrent methodological errors in it by 1978. We did not obtain the review itself. gsb.stanford.edu
- Reference list carried on a peer-reviewed book chapter on methodological strategies in research validating measures of belief in a just world, confirming Lerner, M. J. (1980), The belief in a just world: A fundamental delusion, New York: Plenum; Lerner, M. J. and Miller, D. T. (1978), Psychological Bulletin, 85, 1030–1051; Lerner, M. J., Miller, D. T., and Holmes, J. G. (1976), Deserving and the emergence of forms of justice, in Advances in Experimental Social Psychology, Vol. 9, 133–162; Lerner, M. J. and Simmons, C. H. (1966), Journal of Personality and Social Psychology, 4, 203–210; Lipkus, I. M. (1991), on the construction and preliminary validation of a global belief in a just world scale; and Zuckerman, M., Gerbasi, K. C., Kravitz, R. I., and Wheeler, L. (1975), on the belief in a just world and reactions to innocent victims. Note: a reference list; citations only. Recorded also as the source of a bibliographic variant: it renders the 1966 title with an added definite article and without the quotation marks the standard form places around "innocent victim". We obtained none of the works named here. link.springer.com
- Publisher record for a peer-reviewed article on reactions toward innocent victims, DOI 10.1007/s11211-015-0249-3, reproducing its abstract: that reactions toward innocent victims can range from harsh derogatory reactions to great effort to alleviate the victims' ill fates; that using insights from just-world theory and perspective taking the paper investigates both negative and positive reactions; that the authors propose self-focused versus other-focused motives can evoke derogatory versus more benevolent reactions respectively; that by manipulating self-focus versus other-focus they show in two studies that a self-focus enhanced indirect victim blaming and derogation and decreased helping of innocent victims; that when participants were focused on another person these effects attenuated; and that taken together the findings show both blaming and helping can be viable strategies to deal with unjust situations. Note: the publisher's record. Our source for the finding that blaming and helping are alternative responses and that focus determines which. Two studies whose sample sizes we do not have; we obtained the abstract and not the paper. link.springer.com
- Academic publisher citation record confirming Callan, M. J., and colleagues (2009), We Blame Innocent Victims More Than I Do: Self-Construal Level Moderates Responses to Just-World Threats, Personality and Social Psychology Bulletin, 35(11), 1528, alongside further citing works in the just-world literature including studies on the role of perpetrator similarity in reactions toward innocent victims. Note: a citation record; titles only. Recorded because the title points the same direction as the self-focus finding above, and we obtained nothing beyond the citation. cambridge.org
This article discusses research on attribution and outcomes and is not psychological, legal or credit advice. The 1966 paper was not obtained and is described through a later meta-analysis's account of it. All arithmetic is the authors' own; the mathematics is exact and was verified by closed form and simulation, but the assumption that skill explains half the variance in business outcomes is illustrative and is supplied by no source. This article contains a documented self-correction.