Skip to content
Why $1.5M Earners Feel Poor: The Scale-Free Pareto Trap
Industry Insights 10 min read

Why $1.5M Earners Feel Poor: The Scale-Free Pareto Trap

This article explains why even high earners like a $1.5M/year AI researcher feel inadequate, using Pareto distribution's scale-free property showing those above you always average 3x your income, compounded by sampling bias where extreme outliers dominate perception, and how AI's superstar economics steepens this curve.

Model Perspective
Model Perspective
Model Perspective

01: High Income Is a Scale-Free Ruler

Ordinary incomes resemble height: most cluster in the middle, and the tallest person is not an order of magnitude taller. Top incomes, however, follow a Pareto distribution: the number of people earning above x is roughly proportional to x<sup>-α</sup>. This distribution has a key property: it is scale-free. Regardless of your income level, the average income of those above you divided by your income equals α/(α-1), a constant.

average income of those above you
your income

Using U.S. top-income data (Atkinson, Piketty, Saez 2011), the inverted Pareto coefficient rose from ~1.7 in the 1970s to ~3 recently, implying α=1.5 and a 3x ratio. The median of those above you is ~1.6x your income, and 35% of them earn at least 2x. The accompanying chart contrasts this with a normal "height-type" distribution where the multiple shrinks near the top.

From any level looking up, the multiple is the same
From any level looking up, the multiple is the same

Thus, the roommate Xiaoshu's feeling that "everyone around me earns several times more" is not an exception but the standard experience at every Pareto layer. A person earning ¥20k/month looking at ¥60k, a director looking at a VP, and a $1.5M earner looking at ¥1.5B — the proportional gap feels similar. Climbing one rung reveals a new rung that looks just as distant.

02: You Only Hear About the Highest Earners

A second amplification comes from sampling bias. Xiaoshu says "I asked people around me." But whose salary gets asked about, retold, and remembered? The highest few. "5 years ¥1.5B" spreads across dinner tables; "about the same as me" goes unmentioned. The author simulates this: if the probability of hearing about someone is proportional to their income, the median income among those you hear about becomes 4x yours, while the true median of those above you is only 1.6x. This mirrors the "friendship paradox" (Feld 1991): your friends' friends have more friends than yours because highly connected people appear in more samples. Luttmer (2005) found that higher neighbor income reduces well-being as much as an equivalent drop in own income. Combining the scale-free distribution (3x gap at every level) with upward-biased sampling means the higher you go, the more likely you are to feel "shortchanged."

The people you hear about vs the real people
The people you hear about vs the real people

03: AI Makes the Ruler Steeper

Why is this especially visible in AI? Rosen's 1981 "Economics of Superstars" explains: when marginal cost of replication is near zero (e.g., a record or a model serving hundreds of millions), tiny quality differences produce enormous income gaps. Large language models fit this market: users pick the single best model, so "making the model slightly better" commands huge value. In 2025, Zuckerberg reportedly offered multi-year packages worth over $100M to top researchers for a "superintelligence lab," and Altman stated Meta offered $100M signing bonuses. The article notes that big-company interns access 80 GPUs while university labs have 8 — resources concentrate where the best model is most likely to emerge. For the income distribution, this means α decreases, the tail lengthens, and the upward multiple grows. If α falls from 1.5 to 1.2, the average income of those above you jumps from 3x to 6x. In such an industry, anxiety is not a mindset issue but a distributional inevitability.

04: Where the Simplifications Lie

The author acknowledges three simplifications: (1) Pareto only describes the top slice; middle and lower incomes follow a log-normal distribution where the upward multiple varies with income. (2) α=1.5 is based on U.S. top-income aggregates; the AI industry's true α is unknown, and the drop to 1.2 is illustrative. (3) The assumption that hearing probability is proportional to income is crude — some hide income, some exaggerate — but the simulation only demonstrates directional bias: your heard sample skews high.

05: A Mental Tool to Carry Forward

Next time you hear "everyone around me earns several times more," ask two questions: Does this distribution have a scale? How did these people enter my view? If the answers are "no" and "because they are prominent," the statement holds at every level. In a scale-free distribution, climbing higher never brings a feeling of "not losing out"; only changing the ruler can change that feeling.

References

[1] Frontal AI. Model Generation. 2026-11.

[2] Atkinson A B, Piketty T, Saez E. Top Incomes in the Long Run of History. Journal of Economic Literature, 2011, 49(1): 3–71.

[3] Rosen S. The Economics of Superstars. American Economic Review, 1981, 71(5): 845–858.

[4] Luttmer E F P. Neighbors as Negatives: Relative Earnings and Well-Being. Quarterly Journal of Economics, 2005, 120(3): 963–1002.

[5] Feld S L. Why Your Friends Have More Friends Than You Do. American Journal of Sociology, 1991, 96(6): 1464–1477.

[6] Axios. Behind the Curtain: Zuck's AI moonshot. 2025-07-03; TechCrunch. Meta is offering multi-million pay for AI researchers. 2025-06-27.

[7] Article's own calculations: Pareto α=1.5, hearing probability proportional to income simulation.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

sampling biascognitive biasincome inequalityAI compensationPareto distributionrelative incomescale-free distributionsuperstar economics
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.