Fourteen Percent Come Back Someone Else

The ninety-sixth article found a rule whose numbers had no author. This one finds a claim with a real scientific basis that was converted into something the basis does not support, and then a second problem that would remain even if the first were solved.

The Verdict, Stated First

Five claims, in descending order of confidence.

One. Lateralization is real and is a property of functions rather than of people, on the paper's own conclusion, which is a distinction the business version erases.

Two. A study of 1,011 individuals found no whole-brain left or right phenotype, which is the specific claim the workplace usage depends on.

Three. On our own arithmetic, a dichotomous type is unstable at the boundary, reassigning about one person in seven at a reliability better than most instruments achieve.

Four. On our own arithmetic, a four-part typology loses nearly half its assignments on retest, if the dimensions behave independently.

Five. And on our own arithmetic, converting a score into a type discards information worth about thirty-six percent of a sample, even when the underlying trait is real and predictive.

The last three hold for any dichotomous instrument, ours, which is why this article is not really about brain hemispheres.

Two of those five come from a paper and three come from arithmetic, ours, and that split is deliberate. The arithmetic would survive the paper being overturned, and it is the part with a direct cost attached.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A for the study's abstract and conclusion, obtained verbatim from the journal and confirmed against three independent records.

Grade A for our own arithmetic, which uses standard statistical results rather than anything we invented, and which does not depend on the study at all.

Grade C for the state of the wider literature, since we obtained one paper and did not survey what else has been published on the question.

Grade D for the history of the claim, which we have not traced and about which we say nothing.

That third grade is a real constraint, ours. One study, however large, is one study, and we give it a section of its own rather than a line in a footnote.

We would rather be explicit about this than quietly confident, ours. A single paper is the thinnest evidential base this series has rested a factual claim on, and the only reason we proceeded is that the commercially useful half of the article does not rest on it.

The alternative would have been to do a full literature survey, ours, and we did not have grounds to claim one. Naming the limit is cheaper and more honest than pretending to a breadth we did not achieve.

A Note On Method

Everything here is verified to August 2026.

We obtained the study's abstract verbatim from the journal's own page[1] and confirmed it identically against a national medical index[2] and an open archive[5].

We obtained the conclusion sentence from the journal's printable full text[3], confirmed against an academic index's summary[4].

We did not obtain the paper's full text beyond its abstract and conclusion passages, so we cannot describe its analyses or its figures.

We surveyed no other literature on hemispheric lateralization, which is a substantial limitation and is the reason for our grade above.

All arithmetic is ours. The statistical results used are standard; the business applications and every figure derived from them are our own constructions.

This article discusses research on brain organisation and on measurement, and is not hiring, psychometric or medical advice.

The Claim As Used

What the idea does in an office, before we look at what supports it.

The abstract describes the conjecture directly: "It has been conjectured that individuals may be left-brain dominant or right-brain dominant based on personality and cognitive style."[1]

Four observations, ours.

In business it becomes an assignment rule. The right-brained people get the design work and the client-facing role; the left-brained people get the spreadsheet and the process.

It also becomes an excuse, in both directions. A person who has decided they are right-brained has an account of why the numbers are somebody else's job.

And it is durable because it is flattering and unfalsifiable as used. Whichever half you are told you are, the half describes something you value about yourself.

Two features make it stick better than most, ours. Neither half is the bad one, unlike most workplace measurements, so nobody has an incentive to contest their assignment.

And it explains a difficulty rather than asking you to fix one, which is a rare and comfortable thing for a workplace framework to do.

The word doing the work is "dominant." Nobody disputes that the hemispheres differ; the claim is that a person is characterised by one of them.

The Study

The source.

Nielsen, J. A., Zielinski, B. A., Ferguson, M. A., Lainhart, J. E., and Anderson, J. S. (2013), An Evaluation of the Left-Brain vs. Right-Brain Hypothesis with Resting State Functional Connectivity Magnetic Resonance Imaging, PLoS ONE, 8(8), e71275, published 14 August 2013, DOI 10.1371/journal.pone.0071275, PMID 23967180[1][2].

Four observations, ours.

The title names the hypothesis directly, which is unusual and useful. This is a paper written to test the popular claim rather than one whose findings were later applied to it.

The sample is 1,011 individuals aged between 7 and 29[1], which is large for neuroimaging and is the reason this study is worth an article.

The data were publicly available resting state scans[1], so the analysis is reproducible in principle by anybody with the same archive.

And the journal is open access, meaning the paper is readable by any business owner who wants to check what we say about it, which is not true of most sources in this series.

We would encourage that, ours, and note what it costs us. An open-access source makes our summary checkable in a way most of our sources are not, and we would rather be checkable than comfortable.

What It Measured

The method, in the abstract's own words, because the design decides what the result can mean.

The abstract states: "We evaluated whether strongly lateralized connections covaried within the same individuals."[1]

And on the measurement: "For each subject, functional lateralization was measured for each pair of 7266 regions covering the gray matter at 5-mm resolution as a difference in correlation before and after inverting images across the midsagittal plane. The difference in gray matter density between homotopic coordinates was used as a regressor to reduce the effect of structural asymmetries on functional lateralization."[1]

Four observations, ours.

The key word is "covaried." The question is not whether any connection is lateralized, which is known, but whether a person who is strongly left-lateralized in one network is also strongly left-lateralized in others.

That is exactly the right test of the popular claim. A left-brained person, if such a thing existed, would show correlated lateralization across networks, because that is what having a type would mean.

The regressor for grey matter density is a design detail worth noting. They attempted to remove structural asymmetry so that what remained was functional, which addresses an obvious confound.

And we did not obtain the analyses themselves, so we describe the design as stated and take no view on how well it was executed.

One thing about the design we would want to see and did not, ours. Comparing 7,266 regions pairwise implies an enormous number of statistical tests, and how that was handled matters a great deal to what the result means.

We flag it as a gap in our reading rather than as a criticism of the paper. A study of this size will have addressed it somewhere in a full text we did not obtain.

Local, Not Global

The finding, quoted directly.

The paper concludes that "lateralization of brain connections appears to be a local rather than global property of brain networks, and the data are not consistent with a whole-brain phenotype of greater 'left-brained' or greater 'right-brained' network strength across individuals."[3][4]

It adds: "Small increases in lateralization with age were seen, but no differences in gender were observed."[3]

Four observations, ours.

The distinction between local and global is the entire finding and is easy to miss. Lateralization exists; it just does not aggregate into a person-level trait.

An analogy, ours. Your right hand is dominant for writing and your left eye may be dominant for aiming, and neither makes you a right-sided or left-sided person.

The analogy holds in the way that matters, ours. Handedness and eye dominance are each real, measurable and useful, and they do not covary strongly enough to define a sided person.

Which is precisely the structure the study reports finding in the brain. Real lateralization, per network, that does not aggregate, and the popular claim is the aggregation.

The absence of a gender difference is worth recording, since the popular version of this claim frequently carries a gendered implication, and the study reports none.

And the phrasing is properly cautious. "Appears to be" and "not consistent with" are the language of a bounded result, not a refutation, and we read it that way.

What Is Actually True

Because the finding is easy to overstate in the other direction, and we will not. Ours.

The abstract opens by affirming it: "Lateralized brain regions subserve functions such as language and visuospatial processing."[1]

Four observations.

Hemispheric specialisation is real and is not in dispute here. The paper's first sentence asserts it.

So the correct statement is narrow. Particular functions are lateralized in most people, and people are not.

Which means the popular claim is not a fabrication in the way the ninety-fifth article's chart was. It is a real finding generalised past what it supports, which is the ninety-third and ninety-sixth articles' structure appearing a third time.

And that makes it harder to dislodge, ours. Every defence of it can point at something true, and the defence never has to engage with the step that fails.

Which suggests where to aim an objection, ours. Grant the true part immediately and ask only about the inference, because an argument that appears to deny hemispheric specialisation will lose to somebody who has read one popular article about aphasia.

The Deeper Problem

The turn this article takes, and the reason it is worth eight thousand words. Ours.

Four observations.

Suppose the neuroscience had come out the other way, and there really were a whole-brain lateralization phenotype. The business practice would still be broken, for reasons that have nothing to do with brains.

Because the practice does not use a measurement. It uses a category, assigning each person to one of two boxes.

And a category built by cutting a continuous measurement has properties that can be computed exactly, without knowing anything about what is being measured.

The rest of this article computes them, ours, and the results apply to every dichotomous instrument a firm uses, whatever it claims to measure.

One consequence of that generality is worth stating up front, ours. We are not attacking any particular commercial product, having examined none, and a reader should not read the arithmetic as being about whichever instrument their firm happens to use.

What A Type Requires

The precondition, stated before any arithmetic. Ours.

Four observations.

A typology asserts that people come in kinds. For that to be true of a measured trait, the distribution must be bimodal: two clusters with a gap between them.

If the trait is unimodal, with most people in the middle, then cutting it produces two groups that are artifacts of where you cut rather than kinds of people.

The test is available and rarely run. Plot the distribution of scores and look for two humps, which any firm with its own instrument data can do in an afternoon.

We cannot test any particular instrument here and do not claim to. What we can compute is what follows if the trait is unimodal, which is the ordinary case for measured human characteristics.

And there is a reason to expect unimodality that needs no data, ours. Most human traits are the sum of many small influences, and sums of many small influences pile up in the middle rather than at two ends.

A bimodal trait would require something unusual, such as a single switch-like cause. That is possible and it is not the default, so the burden sits with whoever asserts the type.

How Many Come Back Different

Our own arithmetic, using a standard result rather than anything we invented.

For two measurements of the same normally distributed trait with a test-retest correlation of r, categorised by which side of the midpoint they fall on, the probability of landing in a different category the second time is the arc cosine of r divided by pi.

At r of 0.95: 10.1 percent change type. At 0.90: 14.4 percent. At 0.85: 17.7 percent. At 0.80: 20.5 percent. At 0.70: 25.3 percent. At 0.50: 33.3 percent.

Four observations.

A test-retest reliability of 0.90 is excellent and better than many published instruments achieve. At that level, one person in seven comes back a different type.

The trait itself has not changed. The instability is entirely a property of drawing a line through a continuum, and it would occur with a perfectly valid measurement.

Note also what the figure does not depend on. It requires no assumption about whether the trait predicts anything, only that it is continuous, roughly normal, and imperfectly measured.

And the direction is worth stating plainly. Better instruments reduce this and cannot eliminate it, because the boundary is a line and people sit near lines.

Two sanity checks on that figure, ours. At a reliability of 1.0 it goes to zero, since arc cosine of one is zero, which is what a correct formula should do.

And at a reliability of zero it gives 50 percent, which is a coin flip, and is also right. The formula behaves sensibly at both ends, which is why we trust it in the middle.

One more property worth noticing, ours. The relationship is not linear. Going from a reliability of 0.95 to 0.90 costs about four points of stability, and going from 0.60 to 0.50 costs about four as well.

So improvements at the top of the range buy less than they appear to. A provider who raises reliability from 0.85 to 0.90 has moved reassignment from about 18 percent to about 14, which is real and is nowhere near enough to make a category safe.

Four Letters Compound It

Our own arithmetic, and this is where it becomes serious for a firm.

Most workplace typologies use more than one dimension. A four-part type requires a person to stay on the same side of four cuts to keep their assignment.

Treating the four as independent, the chance of keeping the full type at a per-dimension reliability of 0.95 is 65.3 percent. At 0.90: 53.8 percent. At 0.85: 46.0 percent. At 0.80: 40.0 percent. At 0.70: 31.1 percent.

Four observations.

At an excellent per-dimension reliability of 0.90, only 54 percent keep their four-letter type, so nearly half of a workforce would come back as something else.

The independence assumption is ours and is a real limitation, which we flag here rather than only in the limits note. If the dimensions correlate positively, the figure improves; if they are largely independent, as such instruments usually claim, it stands.

The compounding is the mechanism and it is unavoidable. Four chances to cross a line are four chances, and adding dimensions to a typology makes it less stable rather than more precise.

And that inverts how these instruments are usually sold. More dimensions are presented as more nuance, and arithmetically they are more opportunities for reassignment.

One qualification a defender is entitled to, ours. Changing one letter of four is a smaller change than changing a single binary type, so the 46 percent figure and the 14 percent figure are not measuring the same severity.

That is fair and it does not rescue the practice. A firm that seats people by full type, or matches them by it, is using the whole string, and for that use a single changed letter is a different person.

What Dichotomising Costs

Our own arithmetic, on the assumption most favourable to the instrument.

Suppose the underlying trait genuinely predicts job performance at some correlation. Splitting people into two types instead of using their score throws information away, and the amount is a standard result.

Dichotomising a normal variable at its midpoint attenuates its correlation with anything else by a factor of the square root of two over pi, which is 0.798.

So a trait truly correlating 0.30 with performance correlates 0.239 once converted to a type. A trait at 0.40 becomes 0.319. The proportion lost is 20 percent at every level.

Four observations.

The typology is strictly worse than the score it is built from, even granting that the underlying trait is real and predictive.

Nothing is gained in exchange. The score was already available, since the type was computed from it, so the category is a deletion rather than a summary.

The loss applies to every use. Predicting performance, matching people to roles, or composing a team all get worse by the same factor.

And the factor is fixed rather than negotiable, ours. It does not depend on the trait, the population or the instrument, which is what makes it a useful thing for an owner to carry around.

And this is the argument we would put to somebody who believes in their instrument. If it works, stop dichotomising it and it will work better, which costs nothing and requires no change of belief.

We think that framing is worth more than the criticism, ours. It converts an argument about validity, which nobody wins, into an operational change nobody has a reason to resist.

Thirty-Six Percent Of Your Data

The same loss expressed differently, because the second form is more vivid. Our own arithmetic.

To recover the precision lost by dichotomising, you would need a larger sample. The equivalent loss is one minus two over pi, which is 36 percent.

Four observations.

So converting scores to types is equivalent to discarding roughly a third of your people before doing the analysis.

A firm that surveyed a hundred staff and then typed them has the statistical power of a firm that surveyed sixty-four and kept the numbers.

Nobody would accept the second framing. Throwing away thirty-six people's responses would be indefensible, and it is the same operation.

We would put that to a provider as the test question, ours. Ask whether they would recommend discarding a third of a client's respondents at random, and if not, ask what distinguishes that from the categorisation they supply.

And this holds regardless of the instrument's quality. A perfect measurement dichotomised loses exactly as much as a poor one, because the loss comes from the cut and not from the measurement.

Two places this bites in a small firm, ours. Any staff survey reported as percentages agreeing rather than as mean scores has made the same trade.

And any customer measure reported as a proportion above a threshold has too, which is the ninetieth-something article's territory and the same arithmetic.

The Hiring Arithmetic

Our own arithmetic on an invented scenario. Every figure is ours.

Forty candidates, ten to be hired, and suppose a real trait correlates 0.30 with subsequent performance, which would be a respectable figure for any selection instrument.

Used as a score, the correlation is 0.30. Used as a type, it is 0.239.

And on top of that, at a per-dimension reliability of 0.90, about 14 percent of candidates would have been assigned the other type on a different day. Of forty candidates, that is roughly six people on the wrong side of the line for no reason other than when they were tested.

Four observations.

The two problems stack, since a weaker signal is applied to a partly arbitrary assignment.

Six of forty is not a rounding error. It is more than half the number of positions being filled, and those six are not random with respect to the decision, since they are precisely the people closest to the boundary.

The figures are invented and the structure is not. Any firm using a two-box instrument to screen has this arithmetic, whatever numbers it would substitute.

And there is a legal dimension we are not qualified to address and will name anyway. A selection procedure with this instability is difficult to defend as job-related, and a firm should take advice rather than ours on that.

We say that carefully, ours. We are not making a legal claim and have examined no jurisdiction's requirements; we are observing that a procedure reassigning one candidate in seven by the day of testing invites a question a firm should be ready for.

The People At The Boundary

Who bears the cost, which the aggregate figures hide. Ours.

Four observations.

The reassignment is not spread evenly across a workforce. Somebody two standard deviations from the cut essentially never flips; somebody near the line flips often.

So the instrument is most confident about the people it least needed to classify, and least reliable about exactly the people whose classification is doing real work.

That is the reverse of what a firm wants from a measurement. Precision is most valuable near the decision threshold, and here it is worst there.

And there is a fairness point in that, ours. The people harmed by the instability are the ones the instrument treats as marginal, and they receive a definite label describing a genuinely uncertain reading.

One arithmetic note on how sharp that concentration is, ours. Somebody sitting two standard deviations from the cut has a vanishing chance of flipping at any reasonable reliability, so essentially the whole reassignment rate is drawn from a narrow band around the line.

The remedy here is available and almost never used, ours. Report a band rather than a side, so that somebody near the cut is told they are near the cut.

The Team Composition Argument

The most common workplace use, examined on its own terms. Ours.

Four observations.

The argument runs that a balanced team needs both kinds, so a manager should deliberately mix types when forming a group.

Even granting the types, the argument requires a claim nobody makes explicitly: that mixed groups outperform matched ones. That is an empirical question about group performance and has nothing to do with brain organisation.

And the arithmetic makes the practice worse than doing nothing. If 14 percent of assignments are unstable, a manager balancing a team of eight has roughly one member misplaced for no reason connected to the person.

The alternative costs nothing, ours. Compose teams on what people have actually done, which is observable, recorded, and specific to the work in front of you.

Two objections to that we would grant, ours. Past work is a poor guide for someone new, and a manager forming a team from strangers has less to go on.

Even then the honest answer beats the instrument. A short trial on real work tells you more in a fortnight than a questionnaire does in an hour, and it measures the thing you actually care about.

Almost All Of It Is Self-Report

A feature of the workplace instruments that the neuroimaging study does not share. Ours.

Four observations.

The study we examined used scans. Every commercial instrument we are aware of uses a questionnaire the person fills in about themselves.

That difference is not a small one. A self-report measures how a person describes themselves today, which is affected by mood, by recent events, and by what they think the instrument is for.

Which supplies an obvious source of the instability our arithmetic quantifies, ours. The reliability figure is not a defect of the questions so much as a fact about self-description, and better questions can only do so much.

And it introduces a second problem the arithmetic does not cover. A candidate who wants the job answers differently from an employee doing a team exercise, so the same instrument measures different things in different settings.

Which means published reliability figures may not transfer to your use, ours. A reliability measured on volunteers with nothing at stake is an upper bound for a hiring context, and our 14 percent figure would be optimistic there.

Where You Put The Cut

A further degree of freedom, and it is rarely disclosed. Ours.

Four observations.

Our arithmetic assumed the cut sits at the midpoint, which is the most favourable case. A cut placed elsewhere gives different figures.

And the cut has to be placed somewhere. Whoever built the instrument chose a threshold, and that choice determines how many people land in each type.

The choice is usually invisible to the user. A report saying you are one type rather than another does not report how close you were, and the distance is the only thing that tells you how much to believe it.

Which suggests a question worth putting to any provider, ours. Ask where the threshold sits and how it was chosen, and whether the reported type comes with a distance from it.

Two answers would satisfy us, ours, and both are respectable. The threshold was set at the sample median, which is defensible and arbitrary, or it was set where a genuine gap appeared in the distribution, which would be a real finding worth seeing.

An answer we would not accept is that it was set to produce useful-sized groups. That is a design decision about convenience, and it should not be reported as a property of the people.

What An Honest Report Would Say

Because criticism is cheap unless you can describe what would satisfy you. Ours.

Four features.

It would give the score, not the label, along with where the person sits in the distribution.

It would state the test-retest reliability and the resulting reassignment rate, so the reader knows how much of the result is measurement and how much is the day.

It would mark people near the threshold as unclassified rather than assigned, which is what the arithmetic supports and what no commercial report we are aware of does.

And it would be far less satisfying to receive, which is the recurring finding of the last three articles. A report that says you scored 54 with a reassignment probability of 14 percent tells you more and gives you nothing to repeat.

We would put the general form of that plainly, ours. Every property that makes a measurement honest reduces what it can be used for socially, and instruments are bought by people who need the social use.

Two Failures, Not One

Separating the strands, because they are usually merged and have different remedies. Ours.

Four observations.

The first failure is empirical: the specific claim that people have a whole-brain dominance was tested and the data were not consistent with it.

The second is structural: turning any continuous measurement into two boxes costs precision and stability, whatever the measurement is.

They have different fixes. The first is fixed by dropping the claim; the second is fixed by keeping the score, and a firm that only does the first will carry the second into whatever instrument it adopts next.

Which is why we think the second matters more commercially. Businesses replace one typology with another regularly, and the arithmetic follows them across.

And it explains a pattern any owner will recognise, ours. Each new instrument arrives promising to fix what was wrong with the last one, and none of them addresses the operation that causes the trouble.

What Actually Survives

Our reading, stated directly.

Five statements.

Hemispheric specialisation is real, and the paper's own first sentence affirms it.

Lateralization is local rather than global, on that paper's conclusion from 1,011 individuals, and the data were not consistent with a whole-brain phenotype.

On our own arithmetic, a dichotomous type reassigns about 14 percent of people on retest at a reliability of 0.90, purely from cutting a continuum.

On our own arithmetic, a four-part type loses about 46 percent of assignments, if the dimensions are independent.

And on our own arithmetic, dichotomising discards information worth about 36 percent of a sample, whether or not the underlying trait is any good.

Those five are what we would defend, ours, and the last three carry no dependence at all on the neuroscience, which is the structural point of this article.

Not An Argument Against Measuring People

The obvious misreading, addressed directly. Ours.

Four observations.

Nothing here says traits cannot be measured or that measurement is useless. The arithmetic assumes throughout that the underlying trait is real and predictive.

The objection is to one specific operation: converting a measurement into a category, which is done for convenience and costs precision.

And the remedy is unusually cheap. Keep the score, which the instrument already produced, and use it as a score.

A firm that does that has a better instrument for free, with no change to what it administers and no argument about validity required.

And it has a second benefit worth naming, ours. Scores can be tracked over time and categories mostly cannot, since a person who moves from 48 to 56 has visibly moved, where one who crosses a threshold appears to have become someone else.

Why Types Persist

The mechanism, ours, and it differs from the previous few articles.

Four observations.

A category is actionable and a score is not. You can seat the right-brained people together; you cannot seat a 62 next to a 58.

Categories also survive being repeated, which scores do not. A four-letter type is a thing a person tells a colleague about themselves.

And they satisfy a real need. A manager with twenty reports genuinely cannot hold twenty score profiles in mind, and the typology solves that at a cost nobody has costed.

We would note the ninety-fifth article's finding arriving here again. The properties that make an instrument usable are the ones that make it unreliable, and the trade is real rather than a failure of rigour.

Which is why we would not tell a manager simply to stop, ours. The need the typology serves is genuine, and an objection that offers nothing in its place will be ignored, correctly.

What A Label Does Afterwards

A cost the arithmetic does not capture. Ours, and offered as reasoning rather than as evidence.

Four observations.

A score is understood as a reading. A type is understood as an identity, and people describe themselves by it years later.

The label then shapes what a person attempts. Somebody told they are not the numbers person has been given a reason to decline work that would have developed them.

And it shapes what a manager offers, which compounds. The assignment becomes self-confirming, since the person typed as creative accumulates creative work and the evidence appears to mount.

That compounding has a specific cost to a firm, ours. It narrows the range of work each person has done, which reduces cover when somebody leaves and makes the business more fragile than it needs to be.

We flag that this is our reasoning and not a finding we obtained, and a reader should weigh it accordingly. It is the most speculative section in this article and we mark it as such.

Your Own Instruments

The practical application. Ours, and not hiring or psychometric advice.

Four steps.

Ask the provider for the test-retest reliability, per dimension, and for the interval over which it was measured. A reputable instrument will have this and will supply it.

Compute the reassignment rate yourself from that figure, using the arc cosine relation, which takes one line in a spreadsheet.

Ask whether the score is available, since it almost always is, and use it instead of the category wherever a decision is being made.

And plot your own distribution if you have enough staff data. If there is one hump rather than two, the types are a cut through a continuum and you now know it about your own people.

One note on how much data that takes, ours. Forty or fifty scores will not settle the shape of a distribution, and the eighty-eighth article's caution about small samples applies with full force here.

One Study Is One Study

The caveat we owe the neuroscience half of this article, given its own section. Ours.

Four observations.

We obtained one paper and surveyed no others, so we cannot tell you whether it is representative of the field or an outlier.

It is a large sample and an open dataset, which counts for something. It is not a replication, and we do not know whether one exists.

What we did do is confirm the paper exists as described, ours, across four independent records[1][2][5][6]. That establishes the citation and says nothing about the field.

The paper itself is appropriately bounded, which we noted above. "Appears to be" and "not consistent with" are not claims of refutation, and we have not upgraded them.

And this is why we built the article on arithmetic that does not need the study. If the neuroscience were overturned tomorrow, the typology arithmetic would be unaffected, and that is the part a firm can act on.

Bibliographic Note

The series keeps a count, and this article produced two, both minor.

An academic index gives the paper as PLoS ONE, volume 8, with no issue and no article number[4], where the journal's own record is 8(8), e71275[1].

The same index's summary renders the phrase "right-brains" where the paper reads "right-brained"[3][4].

Three observations, ours.

The missing article number matters more than it looks, since an electronic-only journal has no page range, and the article number is the entire locator.

The second is a transcription slip and we record it only because it appears in the sentence carrying the paper's central finding, which is where errors are most likely to propagate.

That brings the running count of bibliographic variants across this series to forty-nine.

Both are unusually mild by this series' standards, ours, which is what an open-access paper with a DOI and four indexed records produces. Good bibliographic hygiene is visible in the variant count, and this article's is the cleanest since the eighties.

What To Do

Stop calling people left-brained or right-brained. The study found lateralization to be a local property of networks, with the data not consistent with a whole-brain phenotype.

Keep the distinction the paper actually draws. Functions are lateralized; people are not, and the first sentence of the abstract affirms the first half.

Use scores rather than types wherever you make a decision. On our own arithmetic, the category correlates 20 percent less with anything than the score it was cut from.

Ask your instrument's provider for test-retest reliability per dimension. Then compute the reassignment rate, which is the arc cosine of that figure divided by pi.

Expect roughly one person in seven to change on a single dimension at a reliability of 0.90, and about 46 percent to change at least one letter of a four-part type.

Treat people near the boundary as unclassified. That is where the instrument is least reliable and where its output is doing the most work.

Plot your own distribution before believing in types. One hump means a cut through a continuum, and your own data can tell you.

And avoid announcing a label to the person. A score is heard as a reading and a type is heard as an identity, which is our reasoning rather than a finding.

The Limits Of This Analysis

Several caveats matter. This article discusses research on brain organisation and on measurement and is not hiring, psychometric, legal or medical advice; a firm using selection instruments should take proper advice. Everything is verified to August 2026. We obtained one paper on hemispheric lateralization and surveyed no other literature on the question, which is a substantial limitation: we cannot say whether this study is representative, whether it has been replicated, or whether contrary findings exist. We obtained its abstract and conclusion passages and not the full text, so we cannot describe its analyses, its figures, or how it handled the many statistical comparisons its design implies across 7,266 regions. The paper's own language is appropriately bounded, saying lateralization appears to be local and that the data are not consistent with a whole-brain phenotype, and we have not upgraded those to a refutation. All arithmetic is ours. The statistical results used are standard rather than invented, but their application here rests on assumptions: that the underlying trait is continuous and approximately normally distributed, which we cannot verify for any particular instrument; that the cut is at the midpoint, where a different cut point gives different figures; and, for the four-letter calculation, that the four dimensions are independent, which is our assumption and would improve the result if they correlate. The forty candidates, the ten positions and the correlation of 0.30 are invented by us. Our section on what a label does to a person afterwards is our own reasoning and rests on no evidence we obtained, and is the most speculative material in this article. And we have made no claim about the validity of any specific commercial instrument, having examined none.

Related Reading

Behavioural Finance The Bottom Eighty Percent Pay The Rent Another rule that turns a continuum into two boxes and then acts on the boxes. No. 98 Ask The Winner Who Was First Earlier research on first-mover advantage asked surviving firms whether they had pioneered. On our own arithmetic, that one design choice can... Behavioural Finance The Interval Nobody Quotes What happens when a finding is generalised past the domain it was measured in.

Frequently Asked Questions

Are some people left-brained and others right-brained?

On the study we obtained, no. It concluded that lateralization of brain connections appears to be a local rather than global property of brain networks, and that the data were not consistent with a whole-brain phenotype of greater left-brained or right-brained network strength across individuals, from scans of 1,011 people.

So hemispheric specialisation is a myth?

No, and the paper's own first sentence affirms it: lateralized brain regions subserve functions such as language and visuospatial processing. The correct statement is that particular functions are lateralized in most people, and people are not.

How unstable is a two-box personality type?

On our own arithmetic, at a test-retest reliability of 0.90, which is better than many published instruments achieve, about 14 percent of people land in the other category on retest. At 0.80 it is about 20 percent. The trait has not changed; the instability comes from drawing a line through a continuum.

What about four-letter types?

Treating the dimensions as independent, which is our assumption, only about 54 percent keep their full four-letter type at a per-dimension reliability of 0.90. Nearly half would come back as something else. Adding dimensions makes a typology less stable, not more precise.

Why is using a score better than using a type?

Dichotomising a normal variable at its midpoint attenuates its correlation with anything else by a factor of about 0.798, so you lose 20 percent of whatever predictive power the trait had. The equivalent loss of precision is about 36 percent of your sample, and the score was already available.

Does this mean personality measurement is worthless?

No. Our arithmetic assumes throughout that the underlying trait is real and predictive. The objection is to one specific operation, converting a measurement into a category, and the remedy is to keep the score the instrument already produced.

Who is most affected by the instability?

People near the boundary. Somebody far from the cut essentially never flips, so the instrument is most confident about those it least needed to classify and least reliable about exactly the people whose classification is doing real work.

References

1. Journal record for Nielsen, J. A., Zielinski, B. A., Ferguson, M. A., Lainhart, J. E., & Anderson, J. S. (2013), An Evaluation of the Left-Brain vs. Right-Brain Hypothesis with Resting State Functional Connectivity Magnetic Resonance Imaging, PLoS ONE, 8(8), e71275, published 14 August 2013, DOI 10.1371/journal.pone.0071275, reproducing the abstract: that lateralized brain regions subserve functions such as language and visuospatial processing; that it has been conjectured that individuals may be left-brain dominant or right-brain dominant based on personality and cognitive style, but that neuroimaging data has not provided clear evidence whether such phenotypic differences in the strength of left-dominant or right-dominant networks exist; that the authors evaluated whether strongly lateralized connections covaried within the same individuals; that data were analyzed from publicly available resting state scans for 1011 individuals between the ages of 7 and 29; that for each subject, functional lateralization was measured for each pair of 7266 regions covering the gray matter at 5-mm resolution as a difference in correlation before and after inverting images across the midsagittal plane; and that the difference in gray matter density between homotopic coordinates was used as a regressor to reduce the effect of structural asymmetries on functional lateralization. Note: the journal's own record for an open-access paper, and our source for every figure and quotation from the abstract. We obtained the abstract and conclusion passages, not the full text. journals.plos.org
2. National medical index record for the same paper, giving PLoS One, 2013 Aug 14;8(8):e71275, DOI 10.1371/journal.pone.0071275, PMID 23967180, PMCID PMC3743825, with the first author at the Interdepartmental Program in Neuroscience, University of Utah, and reproducing the abstract identically to the journal record. Note: an independent index, used to confirm the abstract verbatim and to establish the bibliographic identifiers. pubmed.ncbi.nlm.nih.gov
3. Printable full-text version of the paper hosted by the journal, carrying the conclusion that lateralization of brain connections appears to be a local rather than global property of brain networks and that the data are not consistent with a whole-brain phenotype of greater left-brained or greater right-brained network strength across individuals; and that small increases in lateralization with age were seen, but no differences in gender were observed. Also carries the funding statement and open-access licence. Note: the journal's own printable text, and our source for the conclusion sentence and the age and gender findings. We did not read the full paper. journals.plos.org
4. Academic indexing service record for the same paper, carrying a summary stating that lateralization of brain connections appears to be a local rather than global property of brain networks and that the data are not consistent with a whole-brain phenotype of greater left-brained or greater right-brained network strength across individuals. Note: an indexing service, used to confirm the conclusion independently of the journal. Recorded also as the source of two bibliographic variants: it gives the paper as volume 8 with no issue or article number, and its summary renders the phrase as right-brains where the paper reads right-brained. semanticscholar.org
5. Open archive copy of the same paper, reproducing the abstract identically and listing the authors' departmental affiliations across neuroscience, paediatrics and neurology, bioengineering, and a laboratory for brain imaging and behavior. Note: an open archive, used as a third independent confirmation of the abstract. pmc.ncbi.nlm.nih.gov
6. Bibliographic record for the same paper in an astrophysics data system's general index, listing the five authors and the 2013 PLoS ONE publication. Note: a bibliographic index, recorded only as a fourth independent confirmation that the paper and its authorship are as described. No content was drawn from it. ui.adsabs.harvard.edu

This article discusses research on brain organisation and on measurement and is not hiring, psychometric, legal or medical advice. One paper was obtained on hemispheric lateralization and no other literature on the question was surveyed. Its full text was not obtained, only the abstract and conclusion passages. All arithmetic is the authors' own, using standard statistical results under assumptions stated in the limits section, and the section on what a label does to a person rests on no evidence obtained.