Would Eugenics Work? Simulating Positive Eugenics Targeting IQ
Employing quantitative genetics and the UN's cohort component method we estimate the impact of a positive eugenics policy.
TL;DR
A positive eugenics policy ($20,000 annual child benefit per standard deviation of parental IQ above population mean) produces 2-18 IQ point gains over a century, depending on policy effectiveness
GDP per capita increases 22% to 6.5-fold across scenarios, with the high-effectiveness scenario reaching $242,559
All scenarios experience catastrophic population decline given humanity's demographic crisis. With the no-policy control losing 70% of its population.
Base fertility rates (what occurs if policy is abolished) collapse to 0.66-1.14 children per woman by year 100, creating permanent policy dependency
Policy cost equals 3.2% of GDP annually if implemented in the US; equivalent to the US military budget.
Innovation index starts from Poland/Greece/Ireland levels to Switzerland/USA/Sweden.
Introduction
In our previous article we calculated the effect of population decline and dysgenics on innovation.
Naturally, such inquiry also raised another interesting question by Joseph Bronski: if eugenics had been employed decades earlier, how much more innovative progress would we have had?
Would we have developed LLMs and the internet in the 1960s? Would the singularity have already occurred by now? When would we have landed on the moon?
However and alas, the effect of eugenics and the effectiveness of its policies is complex. There are many varying types of policies one could implement, each with their pros and cons.
For example, broadly speaking, negative eugenics would reduce the overall population of a country through reduction of fertility. Likewise, positive eugenics could fail, given that there is no known tested policy intervention that has successfully elevated the fertility of any population beyond rounding error, let alone a target demographic. Furthermore, any policy successful at increasing the IQ of a country would proportionally decrease its fertility, given its correlation with national IQ.
To help ameliorate the complexities of such hypothesized effects, we employ a series of simulations.
We generate our own novel population projections, modifying the fertility, IQ, innovation and other statistics of a million persons, proportional to the effectiveness of a eugenics program.
The effect sizes of such policies are determined by what we know about fertility altering pro-natal policies. We use this in addition to the employment of the UN's cohort component method and quantitative genetics on an individual level to generate a hypothetical positive eugenics policy outcome.
**Warning: the method ahead is for the nerdy. If you're more interested in the eugenics policy, its effectiveness, outcomes and other key assumptions, we recommend skipping to the section "Method: The Eugenics Policy".**
Method: Main Loop
From top to bottom, this is how our simulation works:
First, we generate the initial population of 1 million persons. For convenience, we use the exact same age and sex distribution as the United States as of 2023—this was provided by the UN.
Next, we need to assign IQs to these individuals. To simplify our simulation, we do not assume any sex differences in intelligence or differences by age. To accomplish this, we give everyone a genotypic IQ and an environmental IQ, then sum these to obtain their resultant phenotypic IQ.
Let \\(P\\) be the phenotype; it is the sum of a genetic component \\(G\\) and an environmental component \\(E\\).
The total phenotypic variance in the population \\(V_P\\) is partitioned into the variance of the genetic values \\(V_G\\) and the variance of the environmental effects \\(V_E\\), assuming no covariance between genetic and environmental factors.
Heritability \\(h^2\\) is defined as the proportion of the total phenotypic variance that is attributable to genetic variance. For our simulation, \\(h^2\\) is key and is assumed based on the literature:
Given the defined total phenotypic variance \\(V_P\\) and heritability \\(h^2\\), the genetic and environmental variances are determined as follows:
We assume, according to standard IQ scaling, that the standard deviation of IQ is 15. Variance equals the standard deviation squared, so in this case:
Genetic Value (G): Sampled from a normal distribution with a mean equal to the population's starting mean phenotype \\(\mu_P\\) and a variance of \\(V_G\\).
Environmental Value (E): Sampled from a normal distribution with a mean of 0 and a variance of \\(V_E\\).
This results in the following initial code:
Next, now that we have the initial population, we need to begin aging the population by one year. This involves the following steps:
Coupling our simulated persons for assortative mating.
Generating the births resulting from these couples for that year.
Estimating who will die and at what age. This includes child mortality for the newborns.
Assortative Mating
To simulate assortative mating, we generate a "desired partner phenotype" for each fertile female. This desired phenotype is correlated with the female's own phenotype, controlled by the mating correlation parameter, which is set to \\(r=0.4\\).
This process ensures that females with higher phenotypic values will, on average, desire partners with higher phenotypic values, and vice versa.
The correlation value is sourced from a meta-analysis of twin studies.
Let \\(P_f\\) be the phenotype of an individual female. We first standardize this phenotype relative to the mean \\(\mu_{P,f}\\) and standard deviation \\(\sigma_P\\) of the fertile female population.
Next, we generate the standardized desired partner phenotype, \\(P_{d, \text{std}}\\). This new variable is constructed to correlate with \\(P_{f, \text{std}}\\) according to our assortative mating parameter. This is achieved by combining the female's standardized phenotype with a random component \\(\epsilon\\), where \\(\epsilon\\) is drawn from a standard normal distribution \\(\mathcal{N}(0, 1)\\).
Finally, this standardized desired phenotype is converted back to the original phenotypic scale to yield the final desired partner phenotype, \\(P_d\\).
All this is computed in the code section starting from population_fertile_females.
This next part is important.
These are only the *desired* partners, not the actual male partners that exist in our population. We conduct matching under two key constraints in our simulation:
Age-Restricted Mating: A female of a specific age can only be matched with males who are a fixed number of years older (three years in this simulation). This mimics realistic age-pairing patterns. The algorithm iterates through each fertile female age group (15-49) to find eligible partners.
Monogamous Pairing: Each male can only be part of one couple per year. Once a male is matched, he is removed from the pool of available partners for that year's mating cycle.
The matching itself is a deterministic, rank-order process designed to satisfy the females' preferences as closely as possible given the available males. For each age-based group:
The number of couples to be formed is determined by the size of the smaller group—either the females of that age or the eligible older males. A random sample of this size is drawn from both groups. Usually there are fewer females than males; however, so all females are paired.
The sampled females are then sorted in ascending order based on their desired partner phenotype.
The sampled males are also sorted in ascending order based on their own actual phenotype.
Couples are formed by pairing the individuals at the same rank in each sorted list. In other words, the lists are bound together. The female desiring the lowest-phenotype partner is matched with the available male who has the lowest phenotype; the female desiring the second-lowest is matched with the male who has the second-lowest, and so on.
Following from our previous section, this is coded in R as:
To verify this method worked, we can visualize the outcome of our algorithm below. It appears to show a correlation of 0.4, and when we verify this in R, we indeed achieve this value.
Births
To simulate which couples produce births, we need the Age-Specific Fertility Rate (ASFR). This is typically represented as the number of births per thousand women for that specific age.
As we did with our population structure, we can rely on the UN to provide the template for ASFR; again we chose the United States in 2023.
In our simulations, we use this value divided by one thousand, so we have the number of births per woman, instead of the number of births per thousand women.
This is crucial because we then treat the ASFR as a probability: the probability that a woman in a given year produces a child.
Using this approach, for all the couples that we assigned above, for that given year, we sample the couples where the probability of being sampled equals the ASFR of the woman. The remaining couples, post-sample, will all generate newborns.
Offspring Genetics, Phenotype and Sex
The genetic and phenotypic IQs of newborns are determined by a combination of parental genetics, random chance from meiosis (genetic variation between siblings), and environmental factors.
The inheritance process for each newborn is modeled in three main steps: determining the genetic value, adding environmental influence, and finally calculating the resulting phenotype.
A child's genetic value, \\(G_c\\), is derived from the genetic values of their parents, \\(G_f\\) (female) and \\(G_m\\) (male). The process begins by calculating the mid-parent genetic value, \\(G_{\text{mid}}\\), which is the simple average of the parents' genetic values.
This mid-parent value represents the expected genetic value of the offspring. However, sexual reproduction involves random chance in which alleles are passed on, a process known as Mendelian segregation. We model this by adding a random segregation component, \\(M_{\text{seg}}\\), drawn from a normal distribution with a mean of 0.
The variance for this segregation is half the total genetic variance \\(V_G/2\\) because the child inherits a random half of each parent's genes, and the variance of the mean of these two halves is \\(V_G/2\\). This term captures the genetic lottery that makes siblings different from one another.
The final genetic value of the child is the sum of the mid-parent value and the segregation component:
Just as with the initial population, each newborn is assigned an environmental value, \\(E_c\\), which is drawn independently from a normal distribution with a mean of 0 and a variance of \\(V_E\\).
This component represents all the non-genetic factors that contribute to an individual's phenotype. The child's final phenotypic IQ, \\(P_c\\), is the sum of their genetic and environmental values.
The sex of each newborn is assigned randomly, governed by the sex ratio at birth (SRB). The SRB gives the number of male births per 100 female births. The probability of a newborn being male, \\(P(\text{male})\\), is calculated as:
Each newborn is assigned "Male" or "Female" based on this probability.
The births, offspring traits and the eugenics policy (which we will discuss later) are modelled in R as follows:
Aging, Death, and Child Mortality
After accounting for new births, the entire population must be aged forward one year, and the chances of survival for each individual must be calculated. This process determines who dies and is removed from the simulation.
The model for mortality relies on age- and sex-specific survival probabilities derived from UN data. As with the initial population structure and fertility patterns, we use data for the United States in 2023 to ensure a realistic demographic baseline.
The core of this process is the life table parameter \\(p_x\\), which is the probability that an individual of age \\(x\\) will survive to their next birthday at age \\(x+1\\). These probabilities are provided separately for males and females.
The simulation of mortality proceeds as follows:
Aging: Each person's age is incremented by one year.
Death: Using these survival probabilities \\(p_x\\), simulated individuals are removed from the population. This represents their death.
Crucially, this survival mechanism also applies to the cohort of newborns. A newborn enters the simulation at Age = 0. Before the simulation proceeds to the next year, they face the infant survival probability, \\(p_0\\). Only the newborns who survive this initial check are added to the population for the following year.
The mortality for the population, male and female by age, including newborns, is modeled in R as follows. This final code snippet represents the end of our loop—one full year of simulation.
Method: The Eugenics Policy
Our eugenics policy is simple:
The IQ of everyone in the population is known through mass testing. When a couple produces a child, the government takes the average of both parents' IQ scores.
That couple's average IQ score is then compared against the general population for that year. "For that year" is crucial, since a population's average IQ, especially under a eugenics policy, can be changing year over year.
For every standard deviation (15 IQ points) the couple's IQ is above the population mean for that year, a child benefit of $20,000 USD is administered to the couple annually—inflation adjusted—with the last payment being made on the offspring's 20th birthday.
This policy is a *positive eugenics* policy; that is to say, couples with an average IQ below the population mean simply receive no benefit, instead of having to pay.
The payment is an addition to all present child benefits that may or may not exist for this hypothetical country or scenario. Additionally, it occurs regardless of marital status, couple living arrangements, or offspring death.
This last clause is to keep our scenario simple and applicable to all couples. Realistically, one would provide incentives to maintain and promote marriages and avoid rampant single-motherhood. However, modeling such effects is completely beyond the scope of our simulation.
Statistically, it's important to note that variance in the general population is less than the variance between couple averages.
This is because part of the variation cancels itself out when averaging two random people. In our sample, for example, the standard deviation in couple mean phenotypic IQ was 12.574 (this adjusts for assortative mating), not 15 as it is for the general population. As another example, for our couples, 11.7% had an average IQ greater than 115, but for any one person it is simply 15.9%. This is somewhat offset by assortative mating, which mitigates this phenomenon.
Policy Effectiveness
To determine the effectiveness of this policy on fertility for the targeted individuals in our simulation, we draw on Lyman Stone's article on pro-natal policies.
Appendix: Academic Research on Pro-Natal Policies
I have identified 34 academic studies since 2000, with the vast majority published since 2005, assessing the effectiveness of specific pro-natal policies. Of those studies, 22 contain sufficiently detailed estimates of effects and have sufficiently quantifiable costs, so that they can be used to estimate plausible elasticities of fertility with respect to various pro-natal incentives. The studies are listed in an appendix linked here. However, the summary finding is simple: an increase in the present value of child benefits equal to 10% of a household’s income can be expected to produce between 0.5% and 4.1% higher birth rates. The figure below presents 26 different elasticity estimates from the 22 studies with enough data to produce such an estimate.
The real median household income for the USA in 2023 was $80,610. Using this figure and Lyman Stone's findings, we can calculate the effectiveness of our positive eugenics policy.
We need to convert these figures into an effect per standard deviation. The mathematics for this is as follows:
Let \\(B_{\text{SD}}\\) be the final policy effect multiplier per standard deviation. This value determines the multiplicative fertility boost applied for each standard deviation a couple's average phenotype is above the population mean.
It is derived from several economic and effect-size parameters. Let \\(C_{\text{policy}}\\) be the total financial cost of the policy incentive per child, \\(I_{\text{household}}\\) be the median household income, and \\(E_{\text{base}}\\) be the baseline policy effectiveness, representing the fertility increase in percentage points for a financial incentive equivalent to 10% of household income.
The total percentage effect, \\(E_{\\%}\\), calculates the total fertility increase our policy is expected to generate. It first determines the policy's cost relative to a 10% slice of household income, and then multiplies this ratio by the base effectiveness.
The total percentage effect is then converted into a multiplicative factor, \\(B_{\text{SD}}\\), which will be used in the exponential model for the policy's dose-response.
Combining these steps into a single formula provides the final calculation for the policy effect per standard deviation.
With this calculation, we can determine the quantitative impact on fertility for our couples.
Crucially, we exponentiate the effect of the policy on fertility. For example, if a couple is 3 standard deviations from the mean, their fertility is multiplied by \\({B_{\text{SD}}}^3\\) instead of \\( (B_{\text{SD}} - 1) \times 3\\). This is a theoretical choice; TFR, when correlated to most outcomes—GDP per capita, NIQ, HDI, etc.—often has a higher correlation when it is log-linear. In layman's terms, fertility is relative—that is to say, a country going from a TFR of 5 to 3 is not analogous statistically to another country going from a TFR of 2 to 0.
For our simulation, this policy effect multiplier manipulates couples' fertility by raising the probability they will produce offspring. In R, this is coded as follows:
IQ x TFR Feedback Loop
There's one more modification we make to the simulation. Realistically, as our simulated country's standard of living, GDP per capita, and years of education increase causally commensurate with IQ, we should expect fertility to fall.
To incorporate this into our simulation, we set the TFR of our simulation equal to the predicted TFR in 2023 for a country with an NIQ of 100. Then, every year, we proportionally adjust the TFR of our country according to its change in NIQ. This is coded into our simulation as follows:
Policy Cost
For calculating the total cost of this policy, once again we decided to keep it relative to the USA since we used its household income to calculate policy effectiveness.
To estimate the policy cost, we created a shortened simulation up until the birth of the newborns. Then, we calculate the sum total of standard deviations above the average that the couples' phenotype was for those newborns.
The total sum of the standard deviations above the mean is then multiplied by the total expected cost the government would have to pay over the course of the entire policy for each standard deviation—that is, $400,000 USD (20 years × $20,000).
Since our population is scaled to 1 million, we multiply the cost by the US population in millions to convert it back to the US (343.4773 million as of 2023).
Results
Finally, the moment you've been waiting for: the results of our simulation. Given the large variation in the theoretical effectiveness of such policies, we ran the simulation four times. Once as a control with no policy effectiveness, and three times varying Lyman Stone's \\(E_{\text{base}}\\)—that is, the expected percent increase in births given a 10% increase in the present value of child benefits.
Remember that \\(E_{\text{base}}\\) was between 0.5% and 4.1%. The confidence intervals became the "Low" and "High" effectiveness scenarios respectively, with the average being the "Medium". Our simulation ran for 100 years for each scenario.
Policy Cost
The total estimated policy cost, if implemented in the US, totals approximately $886 billion for the first year. Taking the US GDP of 27.72 trillion in 2023, this is a cost of ~3.2% of GDP.
That is roughly what the US spends on the military and nearly half of what the US spends on education.
NIQ
As expected from the control scenario, IQ was a stable 100 beginning to end. By the end of the medium scenario, IQ had increased by an average of nearly 10 points.
Comparing this increase to other countries globally, this is equivalent to moving from a national IQ near the Yookay, the Netherlands, Finland, or China, to 2 IQ points higher than Singapore, the highest IQ country.
So how effective was the policy at boosting fertility by an individual's IQ? Well, this varies by the assortative mating in our population, given that our policy functions on couples, not individuals.
We quickly built a mini-simulation on our policy to calculate the effect by IQ and assortative mating values.
In our scenario, we assumed a correlation of 0.4, which is the second line from the bottom in this plot. So the average female in our simulation, with an IQ between 110-115, should have a TFR near replacement. Those couples with an IQ between 140 and 145 should expect to have at least 4 children.
It should also be noted that females with an IQ below average can still benefit from this policy! That is, if they happen to couple with a male with sufficiently high IQ to bring their average above the mean.
Demographics
Next, we have the effect of a pro-natalist policy on the demographics of our hypothetical million-person country. It should be noted that the TFR adjustment took into account the policy and the change in TFR as a result of the effect of changing NIQ over time.
By the end of the century, the TFR of our country had dropped below the baseline!
However, the positive effects on the overall population remained. The no-policy scenario resulted in a catastrophic loss of population. With its TFR near 1.2, nearly half the replacement rate, this is hardly surprising.
Economic and Innovative Effects
The resulting economic effects of our policy are miraculous for GDP per capita. At the start of our simulation, the GDP per capita was ~$37,600 (constant 2015 USD), equivalent to France, Japan, or Italy.
However, by the end of the medium scenario, it had reached ~$108,000, equivalent to Luxembourg, Bermuda, or ~160% of the USA. This growth is approximately an additional 1.06% compounding annual growth rate.
As for the total GDP, only in the radical "High" scenario did it result in total economic growth sufficient to offset its own population decline.
Finally, we applied our innovation index (this is a per capita metric) from our previous post on dysgenics. The initial 100 IQ population starts equivalent in innovation index to approximately Poland, Greece, or Ireland at 2.3 and ends with an innovation index similar to Switzerland, USA, or Sweden at 3.7. For reference, the most innovative country is Germany at an innovation index of 4.04.
Conclusion
Now, naturally, the question arises: would eugenics work? Given the large variation in theoretical effectiveness, our simulation provides a definitive answer—though perhaps not the one eugenic proponents might expect.
Demographics Summary
First, the cognitive effects are undeniable. Even the modest "Low" scenario produces a 2-point IQ gain over a century. The "Medium" scenario generates a substantial 10-point increase, while the "High" effectiveness scenario yields an 18-point improvement that would place a nation 2 IQ points above even Singapore, currently the world's highest-IQ country.
However, and this is important, the demographic story tells a more complex tale. All scenarios, including the control, experience catastrophic population decline. The no-policy scenario sees a 70% population collapse, while even our successful interventions witness 42-66% declines. Notably, the Base TFR—that is, the fertility rate that would prevail if the policy were suddenly abolished—reveals an alarming trend. By year 100, the Medium scenario's base TFR drops to 0.86, and the High scenario plummets to just 0.66. This means that the policy becomes not merely beneficial but essential for demographic survival; without it, fertility would crash to levels incompatible with civilization itself.
Economic and Innovation Summary
Next, the economic transformation is nothing short of miraculous. GDP per capita increases by 22% in the Low scenario, nearly triples in the Medium scenario, and increases by over 6.5-fold in the High scenario—transforming a population from middle-income status (comparable to France or Japan) to among the world's wealthiest nations. The innovation index similarly improves dramatically, suggesting accelerated technological progress that could fundamentally alter the trajectory of human development.
Crucially though, total economic output tells a different story. Despite the massive per-capita gains, total GDP actually falls in all scenarios except the most optimistic "High" scenario. While individuals become dramatically wealthier and more innovative, the shrinking population undermines total economic power until only the most effective policy implementation maintains both cognitive enhancement and demographic viability.
So, to return to our original question—would eugenics work? The answer is yes. The policy succeeds in enhancing intelligence and individual prosperity, yet it simultaneously creates dependency. At an estimated cost of 3.2% of GDP annually, the intervention represents a substantial but potentially justified investment, particularly given that the alternative—as shown by the collapsing base TFR—may be demographic extinction.
The most optimistic scenario suggests truly transformative benefits: a nation could become a technological and economic powerhouse within a century. However, this requires unprecedented policy effectiveness sustained over multiple generations, and the fundamental tension between cognitive enhancement and fertility presents challenges that extend far beyond selective breeding policies alone. Eugenics would work, but it would also become essential—a policy that, once implemented successfully, could never safely be abandoned. Of course a world with eugenics would yield a completely alternative political landscape. One such that may permanently find a solution to the demographics crisis we face.
I see a lot of pragmatic barriers like 140 IQ married couples deciding to mate for life and forgo working to create value, but they could probably be overcome with the right adjustments.