This Graph Changes The Way You View The World

Key Concepts

Normal Distribution: A statistical distribution where most data points cluster around an average value, with extreme outliers being rare.
Power Law: A relationship where one quantity varies as a power of another. Characterized by a few very large events and many small ones, with no inherent physical scale.
Log-Normal Distribution: A distribution that arises when random effects are multiplicative rather than additive. It has a long tail, indicating a higher likelihood of extreme events than a normal distribution.
St. Petersburg Paradox: A theoretical game with an infinite expected value, illustrating the extreme nature of power law distributions.
Self-Organized Criticality (SOC): A property of dynamical systems that naturally evolve to a critical state, exhibiting power law behavior without external tuning.
Universality: The phenomenon where systems in a critical state, regardless of their specific physical details, exhibit the same universal behavior.
Fractals: Geometric shapes that exhibit self-similarity at different scales, often associated with power law distributions.

Summary

Introduction to Normal vs. Power Law Distributions

The video begins by contrasting two fundamental types of data distributions: the normal distribution and power laws. Normal distributions, exemplified by human height or IQ, show data clustering around an average, with extreme values being highly improbable. In contrast, power laws are prevalent in nature and human systems, characterized by a few extremely large events and many small ones. This leads to a much higher likelihood of significant events than predicted by a normal distribution, skewing averages and making predictions difficult due to the absence of an inherent physical scale.

Vilfredo Pareto and Income Distribution

The concept of power laws was first observed by Italian engineer Vilfredo Pareto in the late 1800s. While studying income tax records, he discovered that income distribution did not follow a normal curve. Instead, a small percentage of the population earned a disproportionately large amount of income. Pareto plotted this data on a log-log plot, transforming the distribution into a straight line with a gradient of approximately -1.5. This led to the formulation of a power law: the number of people earning above a certain income $x$ is proportional to $1/x^{1.5}$. This pattern, with varying exponents, holds true for income distributions in most countries to this day.

Casino Games: Illustrating Distribution Types

To illustrate the differences, three casino games are presented:

Table 1 (Normal Distribution): A simple coin toss game where winning $1 per head over 100 tosses. The expected payout is $50, and playing multiple times leads to results converging around this average, characteristic of a normal distribution. This is linked to Abraham de Moivre and the concept of random additive effects leading to normal distributions.
Table 2 (Log-Normal Distribution): Winnings are multiplied by factors (1.1 for heads, 0.9 for tails) over 100 tosses. While the expected payout is $1, the distribution of outcomes is log-normal. This arises from multiplicative random effects. A few outcomes can be extremely large (e.g., $14,000), while the median payout is much lower (61 cents). This demonstrates how multiplicative processes lead to significant inequalities and a long tail of extreme events.
Table 3 (Power Law - St. Petersburg Paradox): The payout doubles with each consecutive tail until a head appears. The expected value of this game is theoretically infinite, a phenomenon known as the St. Petersburg Paradox. This occurs because the probability of extremely large payouts, though small, is sufficient to drive the expected value to infinity. When plotted on a log-log scale, the payout distribution forms a straight line with a gradient of -1, representing a power law ($P(x) = 1/x$). This highlights the absence of a measurable width (infinite standard deviation) in power law distributions.

The Nature of Power Laws and Their Implications

Power laws signify systems lacking an inherent physical scale, making future outcomes highly unpredictable. The average in such systems can increase indefinitely with more measurements due to the significant impact of rare, extreme events. This is likened to the average wealth in a room with billionaires being skewed by their presence.

The Underlying Mechanism: Exponential Conspiracies and Self-Organized Criticality

The St. Petersburg paradox's power law arises from the interplay of two exponential functions: an exponential increase in payout with the number of tosses and an exponential decrease in the probability of reaching that number of tosses. This "conspiracy of exponentials" is a common mechanism for generating power laws.

This concept extends to natural phenomena like earthquakes. The frequency of earthquakes decreases exponentially with magnitude, while the energy released increases exponentially. Combining these leads to a power law distribution of earthquake magnitudes.

Power laws are intrinsically linked to fractals, exhibiting self-similar patterns across different scales. This fractal nature is observed in systems at a critical point, a state of delicate balance between order and disorder.

Self-Organized Criticality (SOC) describes systems that naturally evolve to this critical state without external tuning. Examples include:

Magnets near the Curie Temperature: At the Curie temperature, a magnet transitions from magnetic to non-magnetic. Simulations show fractal-like patterns of magnetic domains of all sizes, leading to a power law distribution of domain sizes. This indicates a scale-free system at criticality.
Forest Fires: A forest fire simulator demonstrates SOC. Trees grow, and lightning strikes initiate fires. The system naturally reaches a state where small fires are common, but occasional massive fires occur, following a power law distribution. The 1988 Yellowstone fires, burning 1.4 million acres, exemplify this. Unlike magnets, forests self-organize to criticality.
Earthquakes: The Earth's crust, under stress from tectonic plate movement, exhibits SOC. Small slips dissipate stress, but sometimes trigger chain reactions, leading to earthquakes of all sizes, following a power law. The Kobe earthquake of 1995 is a case study of such a cascading event.
Sand Piles: Per Bak's thought experiment of dropping sand grains onto a pile demonstrates SOC. Avalanches of sand occur when the pile becomes too steep, and the size distribution of these avalanches follows a power law. This model, though simplified, captures the essence of SOC and its link to power laws.

Universality and its Implications

A key aspect of systems at criticality is universality. This means that systems in the same universality class, regardless of their specific physical composition, behave identically at their critical point. This allows for powerful theories to be developed using simple models, as the underlying physics becomes less important than the critical state itself. This principle applies to diverse phenomena, from phase transitions in fluids to the spread of diseases.

Real-World Applications and Behavioral Adjustments

Understanding power laws is crucial for navigating different environments:

Insurance: Power law distributions mean extreme events are more common than expected. Insurance is designed to protect against these rare, high-impact events. Insurance companies, however, face the challenge of pricing policies to cover potential massive payouts. The bankruptcy of Merced Property and Casualty after the 2018 California wildfires illustrates this risk.
Venture Capital and Publishing: Industries like venture capital and book publishing thrive on power laws. A small percentage of investments or books generate the vast majority of profits (e.g., Horsley Bridge, Bloomsbury's Harry Potter). Success depends on taking calculated risks and benefiting from rare, high-performing outliers.
Streaming Platforms: On platforms like Netflix and YouTube, a small fraction of content accounts for a disproportionately large share of viewership, demonstrating power law dynamics.
Restaurants and Airlines: In contrast, businesses like restaurants and airlines operate closer to normal distributions. They rely on consistent, average performance rather than rare, massive successes.

Adapting Behavior to Power Law Systems

The core lesson is that behavior must adapt to the governing distribution:

Normal Distribution: Emphasizes consistency and average performance.
Power Law: Emphasizes persistence and taking calculated risks, as rare, high-impact events can dwarf average outcomes. The goal is to make repeated intelligent bets, understanding that most will fail, but a few outliers can lead to extraordinary success.

The Internet and the "Preferential Attachment" Mechanism

Albert-László Barabási's research on the internet revealed a power law distribution of webpage links. He proposed the preferential attachment mechanism: new websites are more likely to link to existing popular pages. Simulations confirmed this, showing that as new nodes are added, they preferentially connect to those with more existing connections, leading to a power law distribution of connectivity. This "rich get richer" effect is a common driver of power laws in human systems.

Conclusion: The Critical State and Unpredictability

The world is largely shaped by power laws, suggesting we often exist in a critical state where small actions can have vastly different outcomes. While this makes systems unpredictable, it also means that pursuing areas governed by power laws requires a strategy of making repeated, intelligent bets, rather than seeking guaranteed average results. The unpredictability means that even the most carefully chosen bet could yield nothing or transform one's life. The video concludes with a personal anecdote from Casper, illustrating how a single email, a small action, led to a significant opportunity due to the unpredictable nature of such systems. The availability of simulations and information about the Veritasium game are also provided.