Attractiveness, Aptitude, and Alcoholism
The Curiosity of the Pareto Distribution in Returns to Human Capital
Tinder and Hinge, popular dating apps, have conducted studies over the past few years to investigate the distribution of attractiveness on their respective platforms. As each app’s respective measure of attractiveness, each company was using whether a female “likes” a male or whether a male “likes” a female using some in-app functionality to convey romantic selection. The results from each of these investigations had concluded a very similar result. For Tinder, it was determined that “the bottom 80% of men are competing for the bottom 22% of women, and the top 78% of women are competing for the top 20% of men.”[1] Essentially, 78% of “likes” from women were directed towards the same 20% of men on the app. Similarly on Hinge, the top 10% of men received over 58% of “likes” from women and the top 10% of women received over 45.7% of “likes from men, whereas the bottom 50% of men received 4.3% of all likes and the bottom 50% of women received 7.9% of all likes.[2] Altogether, this distribution of attractiveness for males and females in a heterosexual, cisgender marketplace resembles the distributions of wealth across some of the most inequal countries in the world.
But this distribution, the underlying mathematical structure that explains the relative frequency of values of attractiveness, is curious, as it describes the distribution of a valuable human quality (albeit, a subjective one) across a population – these studies suggest that the returns to attractiveness (which are proxied by likes on the respective application) are not uniformly distributed (i.e., everyone receives roughly the same number of likes) or normally distributed (i.e., the median receives the highest number of likes). Rather, a very small percentage of the population receives the majority of the “likes” on these apps, whereas the majority of the population receives a minor fraction of “likes.” This distribution of attractiveness within dating app populations seems to reflect a larger tendency I’ve observed in human capital.
For the sake of consistency, I would like to define “human capital” in the context of my usage throughout this analysis. I define “human capital” as any quality of a human that generates monetary or economic returns – in this case, it’s not necessarily one’s intellect or skills. One’s attractiveness or personal qualities could also fall under the aegis of human capital, if these qualities are, broadly speaking, monetizable. Further, these monetary or economic returns need not be positive ones necessarily, and one’s human capital may generate negative returns as well. While human capital is an imprecise and problematic way to measure true value, it is helpful for modeling equity in personal traits across society.
It is curious that attractiveness is not the only quality of human capital that follows this unequal distribution. Across the field of management, and at all levels of seniority and organizational maturity, there are a wide variety of principles that fit the following semantic structure: “N% of _______ can be attributed to X% of ________.” Some of the most common examples include:
· “80% of outcomes are due to 20% of causes”
· “90% of problems are due to 10% of your staff”
· “80% of returns are due to 20% of the portfolio”
· “90% of a company’s performance is due to 10% of the employees”
· And so on…
These statements tend to be the result of ad-hoc empirical observations – either rigorously conducted by a researcher in the field of management science/business or intuitively concluded by a professional or long-time manager who is drawing from their career experience. The conciseness of these statements or principles implies a universality about the observation, which is curious because that would in turn suggest that the values (N, X) in the statement formula relate to some intrinsic or special quality of the world.
The first example principle is generally referred to as the Pareto principle or “80-20 rule,” and can be attributed to polymath Vilfredo Pareto’s observation that roughly 80% of the land in 19th century Italy was owned by roughly 20% of the population (i.e., 80% of the returns to land ownership were realized by 20% of the population). The Pareto principle (or set of principles that follow the same structure), can further be generalized by the Pareto distribution. This probability distribution, in its general form, is part of the natural exponential family (NEF) of probability distributions, and broadly speaking, it is a mathematical formulation of these curious observations. I say “broadly,” because, at the risk of either being too technical or too casual, the Pareto distribution intends to describe phenomena such as the aforementioned observations; however, there is more nuance to the relationship than simply filling parameters into a function. For the sake of our discussion, however, we will just discuss its general probability distribution shape.
With respect to the name’s history, the Pareto principle was named by a Romanian-American management consultant, Joseph Juran, who had formulated a similar set of observations to the ones listed above in the context of observing organizational challenges in the corporations he was advising. In Juran’s world, he observed that a small segment of employees at a company, regardless of the company size or industry, were directly responsible for a majority of the company’s success. In other words, the returns to aptitude in a professional setting are also distributed according to this type of distribution. This principle (or principle structure, as the values themselves tend to vary), sometimes referred to as the “law of the vital few,” has been observed in many disciplines and fields outside of management science and economics, including the natural sciences, as well as other social sciences (including the field of datingappology).
A rather concerning manifestation of the Pareto principle can be observed in alcohol consumption, as according to researchers William C Kerr and Thomas K Greenfield, “the top 10% of drinkers drank 55.3% of the total alcohol consumed.”[3] The distribution of alcohol consumption seems to follow the same distribution patterns of attractiveness and aptitude, with the exception being that the returns to alcohol consumption are wholly negative. In other words, the negative returns associated with alcohol consumption are disproportionately borne by 10% of the population. While addiction plays a covariant role at the individual level in the lopsided distribution of the consumption of substances, this excessively inequal distribution raises the question of why the returns to human capital follow this structure?
Between alcoholism, aptitude, and attractiveness, the Pareto principle seems to govern a wide slice of human qualities. Further, while the question of why this distribution recurs across studies on human capital is puzzling, this distribution is also strange because it bumps against our preconceived models of human capital. Traditionally, we use the Gaussian (normal) distribution to model the distribution of human capital, a model that states the majority of the population are average and small percentages of the population tend to be exceptional. For example, we use the Gaussian distribution for metrics such as IQ (which is, by definition, standardized to the normal distribution) – however, using a Gaussian standardized metric may actually be inherently flawed, as it would “force” the numerical approximation of a person’s intellectual quotient to a comparative metric that doesn’t really reflect differences in ability. As such, even assuming the processes for measuring IQ were sound (which is a massive and specious assumption to make), the IQ metric itself would be an irresponsible comparator of ability between individuals.
Essentially, we assume that most people have “mean” qualities – assumptions of this structure would look like
· 50% of results are attributed to 50% of the labor force
· 50% of income is earned by 50% of the labor force
· 50% of alcohol consumed is consumed by 50% of alcohol drinkers
· 50% of Tinder users receive 50% of the likes on the app
And so on.
However, the empirical studies suggest that none of these hypotheses are quite accurate. The consistency of the Pareto distribution with respect to returns on human capital (and we aren’t even getting into the discussion of physical capital, which have returns that also seem to follow these distributions), both positive and negative, suggests that results tend to be driven by extremes. Further, this consistency suggests that models which rely on a mean value to represent the median (50th percentile) of a population are wholly inaccurate. Let me clarify this vague point with an example.
Let’s say a government agency calculates consumer price index (CPI) for a given year using a mean basket of goods (the average bundle of consumer goods purchased across the income distribution). Assuming normal distribution of income, that mean basket of goods can also be assumed to reflect what the 50th percentile of the population purchases. However, this is an assumption – if the underlying income distribution is actually Pareto, that would mean the median basket of goods is a lot different from the mean basket of goods; in fact, it’s a lot smaller than the mean basket of goods. Consequently, if the government agency calculates CPI accordingly, and then firms use this metric to price cost-of-living adjustments for their employees, employees’ cost-of-living will be improperly assessed. This could be potentially devastating, especially if the cost-of-living changes reflected by the mean basket are lower than the real cost-of-living changes reflected in the median basket and workers need higher cost-of-living adjustments than what they are given. In this theoretical (or, perhaps not so theoretical) world, workers’ real incomes would decrease over time relative to their living expenses.
The mismatch between our conception of how returns to human capital are actually distributed (Pareto, exponential, etc.) and the way many of our mental, social, and policy models are engineered (Gaussian or uniform) has many implications which are not fully explored or understood. It’s clear that returns to human capital, both negative and positive, are generally not distributed in a uniform or Gaussian way, but it’s far from clear as to what we exactly should do about it. While there have been empirical studies to suggest that returns to human capital tend to not follow our more “traditional” or “assumed” models, but the literature explaining why this happens tends to be extremely varied and fragmented, especially depending on the lens with which you are trying to unpack the problem. If you were to examine at the issue from the perspective of tax policy, corporate law, sociology, or historical economics, you will find four different theories that could plausibly explain the nature of this distribution. And while the truth may lie at the confluence of narratives, the problem remains the same – we need to better understand why human qualities such as alcoholism, aptitude, and attractiveness are so unevenly distributed. All things considered, these qualities are subjective and socially constructed; there is nothing naturally occurring or scientifically intrinsic about these qualities.
So why does human society collectively distribute these qualities so unevenly?
[1] https://medium.com/@bellehookwrite/attraction-inequality-and-the-dating-economy-a70cbee8e4db
[2] https://qz.com/1051462/these-statistics-show-why-its-so-hard-to-be-an-average-man-on-dating-apps/
[3] https://pubmed.ncbi.nlm.nih.gov/17651465/