Nepal’s IQ Is Not 42

A forensic audit of the four studies behind Richard Lynn and David Becker’s national estimate for Nepal shows how unserious and unscientific their research is.

TL;DR

• Nepalese IQ of ~42 isn’t measured — it’s (badly) assembled. Lynn and Becker’s “42.79” for Nepal isn’t from any national survey; it’s the simple average of nine numbers pulled from four studies, none of which set out to measure Nepalese intelligence.
• Two of the four are nutrition trials where they used the wrong people. In Buckley and Christian, the actual intelligence test was given to the children; Lynn and Becker ignored those results and used the mothers’ background Raven scores instead.
• It’s one tiny, deprived corner of the country. All four sources come from three adjacent Terai border districts (Sarlahi, Bara, Rautahat) — rural, nutritionally stressed, 40–55% of kids stunted, adults averaging about one year of schooling — with no urban, hill, mountain, or educated sample anywhere.
• The arithmetic is broken on its face. A score (13.45) from one study was copied into another where it doesn’t exist; the same input converts to two similar IQ scores to be ~12 points apart; the Christian Flynn-correction math doesn’t reproduce their own printed numbers; and an unweighted average lets one low-scoring trial count four times.
• Even the authors don’t believe it. They call 42.79 “very implausible,” admit no urban comparison and no external check exists — and the figure has swung from ~78 to 42.79 to ~73 across different compilations, which no real population does.

Introduction

In The Intelligence of Nations (2019), Richard Lynn and David Becker assign Nepal a national IQ of 42.79 — effectively about 43, one of the lowest figures in the entire book.

Taken literally, a national average that low would sit far below the score range used to flag an individual for an intellectual-disability evaluation (a real diagnosis needs more than a test score). And taken literally is exactly the problem. As the physician Kiran Raj Pandey has pointed out, the same data imply that almost the entire population of Nepal could not finish primary school — a country with essentially no teachers, shopkeepers, engineers, or doctors. The claim is frankly ridiculous. So the only interesting question is where the number came from.

So I looked. The 42.79 is not drawn from any survey designed to measure Nepalese intelligence. There is no national sample, no representative cross-section, no urban data point of any kind. It is the simple average of nine numbers pulled from four studies — and not one of those studies set out to estimate the intelligence of Nepal.

Two of them are child-nutrition trials, and the “IQ” figures Lynn and Becker take from them are not the children’s scores at all — they are the mothers’ background test scores, recorded as one demographic detail among many. The other two are rural development surveys of farm and household heads in two adjacent districts, where the adults tested averaged barely a year of schooling.

Every sample is rural. Every source comes from one small strip of the southern Terai plains, pressed against the Indian border.

What follows is a study-by-study audit. Each section opens with the book’s own words — the full entry for that study, exactly as Lynn and Becker wrote it — before turning to what the source actually was and what is wrong with how it was used. The aim is to let the primary documents do the work.

How IQ=42 was built

Here is the whole machine on one page — the components Lynn and Becker average in the Nepal entry:

 SOURCE STUDY             | WHAT LYNN & BECKER TOOK     | n
 _________________________|_____________________________|___
 Buckley et al. 2013      | Mothers' Raven, vitamin-A   | 1
                          | arm only (1 of 2 arms)      |
 Christian et al. 2010    | Mothers' Raven, 4 supple-   | 4
                          | mentation groups            |
 Jamison & Lockheed 1985  | Household heads' Raven      | 1
 Jamison & Moock 1984     | Male farm heads' Raven,     | 3
                          | 3 crop subsamples           |
 _________________________|_____________________________|___

The 9 "corrected IQ" values, averaged with equal weight:
   46.88  40.87  39.34  38.90  40.87  39.49  43.61  51.20  43.96
   ___________________________________________________________
   Unweighted national "IQ" of Nepal = 42.79   (weighted: 42.99)

Average those nine numbers and you get 42.79 to the decimal. Lynn and Becker note that weighting the samples barely changes it, nudging it only to 42.99 — which tells you the problem is not the weighting scheme but the pool of sources itself.

Two features of the construction are worth holding onto before we look at the studies.

First, it is a simple, unweighted average. A single subsample of crop farmers surveyed in 1984 is given exactly the same say as a large maternal cohort studied in 2010. In any serious meta-analysis you weight by sample size and quality. Here, nine numbers from very different sources are just added up and divided by nine.

Second, every number is already heavily processed. Lynn and Becker take a raw Raven’s Matrices score from each sample, convert it onto the Advanced Progressive Matrices scale, read off an “uncorrected IQ,” then apply a Flynn-effect correction. The studies themselves report none of these quantities. And the data span roughly three decades of fieldwork — from 1977–78 to 2007–09 — averaged as if they were nine readings of one instrument at one moment.

Study 1 — Buckley et al. (2013)

Here is the entry in full, in Lynn and Becker’s own words:

A study from Buckley et al. (2013, Table 2) measured a CPMraw score of Nepalese Mothers of 17.78. No age for mothers was given. The testing of mothers took place 1994 to 1997 during early pregnancy. CIA (2017, Index: Mother’s mean age at first birth) gives a mean age of mothers at first birth in Nepal of 20.80y, which was used by us at the mean age of the sample. The given raw score is equivalent to -1.39 on the APM-scale, which is at the 0.05th GBRP and equivalent to an uncor. IQ of 47.72, corrected by -0.84 for FE to 46.88. Only the sample with Vitamin-A supplementation was used by us while the placebo sample was excluded due to lower test scores. However, the prevalence of Vitamin-A deficiency in this sample might cause an underestimation of IQ.

Now take it apart.

Buckley et al. (BMJ Open, 2013) was not an intelligence study. It followed 390 children aged 10–13 in rural Sarlahi, in Nepal’s southern plains, to test whether vitamin-A supplementation given to their mothers years earlier had improved the children’s later cognition and motor skills. It hadn’t.

The children came from just 12 wards out of 270 in the original trial area, in what the paper itself calls a chronically undernourished population — about 55% of the children were stunted (abnormally short for their age, the standard marker of long-term undernutrition, defined as a height-for-age deficiency more than two standard deviations below the international reference).

What’s wrong with how it was used:

• Lynn used the parents, not the children. The study’s actual cognitive measurement was the children’s score on the Universal Nonverbal Intelligence Test (UNIT) — a test built to be fair across language and culture, administered by hand gesture. Lynn and Becker ignore it and reach instead for the mothers’ Raven score, which appears in the paper not as a result but as one line in a table of family background. (More on what “background variable” means below.)
• The date and the age are simply wrong. Lynn and Becker say the mothers were tested “during early pregnancy.” They weren’t: the maternal Raven was administered at the follow-up visit, when the children were already 10–13 and the mothers were in their thirties. Because the paper gave no maternal ages, Lynn and Becker plugged in Nepal’s national mean age at first birth, 20.8 years, from a CIA almanac — a demographically nonsensical proxy for age at testing. The wrong number was then fed straight into the score conversion.
• They kept one half of a randomized trial. Lynn and Becker used only the vitamin-A arm and discarded the placebo arm “due to lower test scores.” But in a randomized trial the two arms are random halves of the same population — that is the whole point of randomization — so there is no valid reason to report one and bin the other. Selecting the arm that scored higher does not clean up the estimate; it produces a hand-picked, nonrepresentative datapoint (and, if anything, a flattering one — and it still came out at ~47).
• It is a tiny, deprived slice. 390 children, 12 of 270 wards, one plains district, a nutritionally stressed trial cohort. Nothing about it was built to stand in for a nation.

Study 2 — Christian et al. (2010)

Again, the entry in full:

Four samples from the rural Sarlahi District consisting of mothers with different nutritional supplementation and a control were tested on SPM by Christian et al. (2010, Tab.1). The samples had mean ages of 31.60y, 32.00y, 31.20y and 32.00y. The raw scores obtained by these samples were similar to each other and therefore, all four samples were used by us. The first sample scored on SPM 15.90 the second 15.70, the third sample 16.60 and the fourth sample 13.45. Converted to the APM-scale, these scores are equivalent to -2.23, -2.33, -1.85 and -3.61, and were all below the 0.01st GBR-P. Uncor. IQs are 42.70, 42.26, 44.23 and 41.84. All had to be reduced by 3.36 for FE-correction to cor. IQs of 40.87, 39.34, 38.90 and 40.87.

Christian et al. (JAMA, 2010) followed 676 children aged 7–9 whose mothers had been in a prenatal micronutrient trial, across 30 of the 48 Village Development Committees (the local administrative units below the district level) in Sarlahi. The hard-to-reach hill areas, the transient border zones, and the far edges of the district were deliberately excluded.

As in Buckley, the real cognitive outcome was the children’s scores. The mothers’ Raven score is a background variable.

What’s wrong with how it was used:

• One of the four numbers is not in the study. I pulled Christian’s Table 1 directly. The real maternal Raven means are 15.7, 16.6, 16.6, and 15.9. Lynn and Becker list them as 15.90, 15.70, 16.60, and 13.45. Three match (reordered); the fourth does not exist anywhere in the paper:

Original Table from Christian et. al

CHRISTIAN ET AL. 2010 -- MATERNAL RAVEN SCORE (TABLE 1)
_______________________________________________________

Supplementation group        Actual mean
   --------------------------    -----------
   Control                          15.7
   Iron / Folic Acid                16.6
   Iron / Folic Acid / Zinc         16.6
   Multiple Micronutrients          15.9
   The four numbers Lynn & Becker enter:
   15.90    15.70    16.60    13.45
                              ^^^^^
   No value of 13.45 appears anywhere in Christian et al.
   13.45 is the household-head Raven mean from a DIFFERENT
   study (Jamison & Lockheed, below). Lynn and Becker appear
   to have copied a row across studies -- their error, not
   the original authors'.

• A minor mistake, but their own conversion arithmetic doesn’t add up. Lynn and Becker print uncorrected IQs of 42.70, 42.26, 44.23, and 41.84, then say each was reduced by 3.36 for the Flynn correction. But row-by-row subtraction (42.70 − 3.36 = 39.34, and so on) makes the fourth corrected value 38.48 — while they print 40.87. The corrected numbers they publish cannot be produced from the uncorrected numbers they publish using the correction they describe. The conversion chain is internally inconsistent, which is a deeper problem than any single bad entry.
• It is a severely deprived rural subset. Maternal literacy across the four groups was 12% to 21% — roughly 35 to 45 points below Nepal’s national female literacy rate. Child stunting ran 40% to 50% (figures the JAMA table reports directly). By the study’s own measurements, this is a disadvantaged rural trial population, and Lynn mined its parents’ background scores rather than the children’s actual test results.

Study 3 — Jamison & Lockheed (1985)

The entry in full:

Raven’s-raw scores for heads of households from the districts of Bara and Rautahat were presented by Jamison and Lockheed (1985, Table 4). The full sample includes heads of households from villages with and without schools, but no differences in intelligence between both sub-samples were detected. The full sample had a mean age of 41.71y and scored 13.45. The source reported a maximum score of 36. No information about the specific used Raven’s test was given but a source from the same first author reported an intelligence measurement on a similar population from the same country, named the CPM (Jamison and Moock, 1984). The score of 13.45 would be 13.14 on the SPM and -3.61 on the APM, far below the 0.01st GBR-P and equivalent to an uncor. IQ of 36.55, to which 2.94 had to be added for FE-correction to 39.49.

Jamison and Lockheed was a World Bank study of what drives children’s school participation, built on 795 rural farm households across six panchayats (village-level administrative units) in two Terai districts, Bara and Rautahat. The cognition figure is the household heads’ Raven score; among them, mean schooling was 1.39 years.

What’s wrong with how it was used:

• It isn’t an intelligence study, and the test version is unknown. Converting a raw Raven score into an IQ requires knowing exactly which Raven form was used and which norms apply — and the book concedes that information “was not given.” Lynn and Becker borrow a version by analogy to a different paper, then convert the score to a two-decimal IQ as though it had been measured. The source reports only a score on a 0–36 scale.
• The same number becomes two different IQs. This is the cleanest mathematical red flag in the whole entry. Jamison & Lockheed’s 13.45 is converted to an APM value of −3.61 and an uncorrected IQ of 36.55. A few lines later, Jamison & Moock’s 13.45 is converted to the same APM value, −3.61 — but an uncorrected IQ of 48.26. Identical input on the same test, a nearly 12-point difference in output. A conversion that gives two different answers for the same number is not measuring anything.
• It isn’t independent of Study 4. This survey and Jamison & Moock come from the same World Bank research project (RPO 671–49, “Education and Rural Development in Nepal and Thailand”), the same household survey, the same two districts. They are not two separate looks at Nepal.

Study 4 — Jamison & Moock (1984)

And the last entry in full:

Three rural samples were separated by Jamison and Moock (1984, Table 2) according the kind of crops they grow. No significant differences in CPM-raw scores were reported. The three samples obtained raw scores of 12.98, 13.45 and 13.12, converted to -3.89, -3.61 and -3.80 on the APM-scale. These scored are also far below the 0.01st GBR-P and equivalent to uncor. IQs of 40.67, 48.26 and 41.02. Adding 2.94 to these scores for FE-correction resulted in cor. IQs of 43.61, 51.20 and 43.96.

Jamison and Moock (World Development, 1984) was an agricultural-economics paper — Farmer Education and Farm Efficiency in Nepal — about how schooling and cognitive skill relate to farm productivity. After excluding landless and non-crop households, the analytic sample was entirely male farm heads in Bara and Rautahat:

ADULT SAMPLES LYNN & BECKER TREATED AS "NEPAL"
___________________________________________________________________

Sample                          | Schooling   | Education status
   --------------------------------|-------------|--------------------
   Jamison & Moock (farm heads)    | 1.19 years  | 544 of 683 never
                                   |             | attended school
   Jamison & Lockheed (hh* heads)  | 1.39 years  | ~20% literate
   Christian mothers (4 groups)    | --          | 12-21% literate
   ___________________________________________________________________
   Nepal, 2011 census, for reference:  ~17% urban; female literacy ~57%

* - household head

What’s wrong with how it was used:

• It’s a farm-productivity study, not an IQ survey. The Raven test appears as the one available proxy for reasoning ability in an analysis of crop efficiency among male smallholders.
• The schooling level is not “Nepal” — it is rock bottom. Mean schooling in this sample was 1.19 years, and 544 of the 683 heads — about 80% — had never attended school at all. That is not the national average; it is the least-schooled rural male farming stratum in the country. Urban and educated Nepalis of the era had vastly more schooling. Reading these men’s Raven scores as “Nepal” substitutes the very bottom of the distribution for the whole of it.
• Same unknown-version problem. The paper describes a 36-item Raven test; Lynn and Becker run it through a CPM/SPM/APM conversion chain the source never supplies.
• An internal outlier. The three crop subsamples scored 12.98, 13.45, and 13.12 — converted to corrected IQs of 43.61, 51.20, and 43.96. A raw difference of a third of a point produces a seven-point IQ swing, and the outlier sits on the same 13.45 value flagged above. Whatever it is, it is not a stable conversion.

The patterns that run through all four

Audited one at a time, each study fails in its own way. Lined up together, they share the same defects — and several only become visible once you see the whole construction.

1. These are not “average Nepalis” — they are a specific, deprived population segment. Every sample is rural, with no urban, middle-class, or educated comparison anywhere. The adults averaged barely one year of schooling; the mothers were 12–21% literate. They are subsistence and smallholder farmers and nutrition-trial households, in a population the studies themselves describe as undernourished, with 40–55% of children stunted. The Moock sample is male only. This is the demographic with the lowest measured outcomes in the country, sampled four times.

2. It is one corner of the country, not the country. All four sources come from three adjacent southern Terai/Madhesh districts — Sarlahi, Rautahat, and Bara — a plains border belt with major Maithili-, Bhojpuri-, and Bajjika-speaking populations. There is no hill sample, no mountain sample, no Kathmandu Valley, none of the dozens of other peoples who make up Nepal. “Nepal’s IQ” is, in practice, three rural border districts.

3. The samples aren’t independent, and the lowest ones are counted multiple times. The two Jamison papers come from the same World Bank project, the same survey, the same district pair — not two separate replications. And because the average is unweighted, overlapping data dominate: the four Christian values all come from a single maternal trial and all sit in the lowest cluster (~39–41), so that one population is effectively counted four times. Collapse it to a single value and Nepal’s figure rises by more than a point. The arithmetic doesn’t just lean on a narrow stratum; it double-, triple-, and quadruple-counts it.

4. Lynn used the parents, not the children. In both nutrition trials, proper nonverbal cognitive tests were administered — to the children, on instruments deliberately designed to be fair across language and culture. Lynn and Becker walked past those results and extracted the mothers’ background Raven scores instead.

5. The test rewards schooling and practice, then gets read as raw intelligence. Raven’s Matrices are nonverbal and intended to reduce language loading — Jamison and Moock themselves call the test “relatively language and culture independent” and note it was “all that we have available.” But that does not make it immune to schooling, test familiarity, visual-symbolic practice, or administration context. Here it was given to adults with about one year of schooling, sometimes through on-the-spot Nepali-to-Bhojpuri translation, then converted into a national IQ with a precision the original studies never claimed.

6. The Flynn corrections cut both ways and defeat their own purpose. The older samples are adjusted upward by 2.94, the newer ones downward by as much as 3.36 — then averaged together. Worse, the Flynn correction exists to account for generational gains in nutrition, education, and health. Applying it to populations the studies explicitly define as undernourished and barely schooled defeats the purpose of the adjustment.

7. Low variance is not validity. The book reassures the reader that “the standard deviation across the different studies is only 4.10,” as if tight agreement proved accuracy. It doesn’t. Low variance is exactly what we would expect if the same narrow stratum — rural, low-schooling, nutritionally stressed, southern-plains households — were sampled repeatedly.

8. The only sanity check is circular. The book judges 42.79 implausible because it is far below the figure it gives for neighbouring India, 76.24 — itself a Lynn estimate produced by the same method. And the people actually sampled belong to the Terai population that are near the Indian border: Jamison and Moock note that the Terai runs along the edge of Bihar and Uttar Pradesh and that inhabitants on both sides “share many cultural and economic characteristics.” This doesn’t prove equal test scores across the border, but it makes a 34-point cognitive cliff running along that political line deeply implausible.

9. There was no external check, and the authors say so. Lynn and Becker had no scholastic-achievement or national test data to validate the figure against, and concede they “can neither obtain confirmation nor rejection of the psychometric IQ.” The number stands on these four studies alone.

The number that won’t sit still

Here is perhaps the most telling fact of all. The “42” is not a stable finding even across successive national-IQ compilations.

As critics including the Kathmandu Post’s Kiran Raj Pandey have documented, the estimate assigned to Nepal has changed from roughly 78 in earlier compilations, down to 42.79 in the 2019 book, and back up toward 73 in a later revision — a 35-point collapse and a 30-point recovery across successive versions of the same data.

No real population does that. National cognitive ability does not evaporate and rematerialize in a few years. What changed was not Nepal but which scraps of data were being averaged. The 42.79 is the artifact of one particular set of source choices — the four rural studies above — and nothing more.

Conclusion — What an honest entry would have said

The strangest thing about the Nepal section is that Lynn and Becker very nearly write the rebuttal themselves. Here are their own closing words:

All these samples scored extremely low, both in terms of global relations and to the geographical neighbourhood. At first there was a suspicion that not all sets of Raven’s Matrices were used, but this was not stated in the sources. Indeed, Jamison and Moock (1984, Table 2) reported a range of scores which corresponded to the full CPM and these gave results which are not much different to the results from the other measurements in Nepal. Eventually, the rural origin of all samples may explain the results but an urban sample for comparison was not available.

The unweighted national IQ of Nepal is 42.79, which is very implausible, but the standard deviation across the different studies is only 4.10. The score also remained stable after weightings at 42.99. Data to calculate a SAS-IQ were not available, thus we can neither obtain confirmation nor rejection of the psychometric IQ. Even if all used samples are from rural areas we would expect a national IQ for Nepal not so far below the national IQ of its neighbourhood country India (76.24).

Read that passage again. They call the result “very implausible.” They acknowledge that no urban sample was available for comparison. They concede they cannot confirm or reject the number. Elsewhere in the entry they flag the missing maternal ages and the unknown Raven version, and note that every sample is rural.

Each of those admissions is a reason to report no estimate — to say that the available data cannot support a national figure for Nepal.

Instead, the caveats are noted and overridden, and a two-decimal number enters the table, the rankings, and every chart and argument downstream that treats “national IQ” as something real and measured.

It is not a measurement. It is four rural development studies — about vitamin A, micronutrients, farm productivity, and who sends their children to school — averaged together after their purpose, their populations, and their context were stripped away. Strip a study of all three and you can turn almost any number into almost any conclusion.

Find all my essays on Richard Lynn’s work here.

If you find value in my work and want to help me continue disproving lies around race and IQ, please consider supporting.

Researched and written with the assistance of Opus 4.8 and GPT-5.5

View Sources Here

References

Buckley, Gillian J., et al. “Cognitive and Motor Skills in School-Aged Children Following Maternal Vitamin A Supplementation during Pregnancy in Rural Nepal: A Follow-up of a Placebo-Controlled, Randomised Cohort.” BMJ Open, vol. 3, no. 5, 2013, e002000. https://doi.org/10.1136/bmjopen-2012-002000.

Christian, Parul, et al. “Prenatal Micronutrient Supplementation and Intellectual and Motor Function in Early School-Aged Children in Nepal.” JAMA, vol. 304, no. 24, 2010, pp. 2716–2723. https://doi.org/10.1001/jama.2010.1861.

Jamison, Dean T., and Marlaine E. Lockheed. Participation in Schooling: Determinants and Learning Outcomes in Nepal. World Bank, Education and Training Series, Discussion Paper EDT9, 1985. The version Lynn and Becker cite. A revised journal version appeared as Economic Development and Cultural Change, vol. 35, no. 2, 1987, pp. 279–306, [https://doi.org/10.1086/451586.]

Jamison, Dean T., and Peter R. Moock. “Farmer Education and Farm Efficiency in Nepal: The Role of Schooling, Extension Services, and Cognitive Skills.” World Development, vol. 12, no. 1, 1984, pp. 67–86. https://doi.org/10.1016/0305-750X(84)90036-6.

Lynn, Richard, and David Becker. The Intelligence of Nations. Ulster Institute for Social Research, 2019. ISBN 978–0–9930001–6–4.

Lynn, Richard, and Tatu Vanhanen. IQ and the Wealth of Nations. Praeger, 2002. ISBN 978–0–275–97510–4.

Pandey, Kiran Raj. “Average National IQ Pseudoscience.” The Kathmandu Post, 11 Feb. 2025, https://kathmandupost.com/columns/2025/02/11/average-national-iq-pseudoscience.

Central Bureau of Statistics, Government of Nepal. National Population and Housing Census 2011. National Planning Commission Secretariat, 2012.

Nepal Tourism Board. “Madhesh Province.” https://ntb.gov.np/en/madhesh-province.