The average income in a city tells you almost nothing about the person standing next to you. Averages hide the distribution.
The 'Problem with Average' is a cognitive bias where reliance on the arithmetic mean obscures the true distribution of data, leading to misleading conclusions when outliers or skewed distributions are present.
A Scene Worth Recognising
A small tech startup has five employees earning $45,000, $48,000, $50,000, $52,000, and the founder earns $500,000. The average salary is about $139,000, which suggests everyone is well paid, but four of the five earn far less. Relying on this average could mislead a job candidate about typical earnings. The median salary of $50,000 better reflects what most workers actually receive. This shows how a single high outlier can distort the mean and hide the real spread of incomes.
What it means and how it works
The mean is calculated by summing all values and dividing by the count. A single outlier contributes disproportionately to the sum, shifting the mean toward its value. If the underlying distribution is skewed or multimodal, the mean fails to capture the central tendency experienced by most observations.
Averages summarize a dataset with a single number, but they can be heavily influenced by extreme values. When data are not symmetrically distributed, the mean may not represent any typical case and can hide important variability, making decisions based solely on averages risky.
Why it matters
Depending on averages can lead to poor judgments in finance, health, safety, and policy. For example, designing infrastructure based on average depth of a river ignores dangerous deep sections, and investment strategies based on average returns may overlook risk of catastrophic loss.
The verified research on this pattern supports the following:
- A single extreme outlier can increase the arithmetic mean by orders of magnitude, rendering it unrepresentative of the majority of data points.
- Reliance on averages alone can mask dangerous variability, leading to underestimation of risk in real-world contexts.
Common misunderstandings
Misunderstanding 1: That the average always represents a 'typical' value, or that a small change in the average reflects a small change in the underlying data. In reality, the average can shift dramatically due to a few extreme values while the majority remain unchanged.
See Problem with Average in everyday decisions
Pick a life context to see how this bias can show up outside the textbook.
Average Ticket Resolution Time Misleads Staffing After Update
A team lead uses mean resolution time to set staff levels after a software release, missing the surge in complex tickets.
Scenario
After releasing a new feature, the support lead checks the weekly report and sees the average ticket resolution time is 6 minutes. Assuming the team can handle the load with current staff, she approves a shift reduction for the following week. However, most tickets are simple password resets resolved in under 2 minutes, while a growing subset of tickets tied to the new feature require 15 to 20 minutes for troubleshooting. The reduced staffing causes queues for those complex tickets to build, leading to longer wait times and frustrated users. The lead later realizes that the average hid the bimodal distribution and that relying on it left the team unprepared for the tail of difficult cases.
Where The Bias Enters
The arithmetic mean is pulled upward by the long, feature-related tickets, making the typical ticket appear to need more time than most actually do. When staffing is set on this mean, the majority of short tickets are overstaffed while the few long tickets are understaffed, creating a bottleneck that the average does not reveal.
Decision Check
Before adjusting staff levels, the lead should examine the full distribution-look at the median, the 80th percentile, and a histogram of ticket durations. If the median is much lower than the mean or a clear second peak exists, she should maintain a separate buffer for feature-related tickets or create a triage tier that routes complex issues to specialists.
This pilot example is illustrative and review-gated. It is designed to explain the pattern, not to claim a documented public case.
Sources
- Niroula, Rishab. REV 2.0 Topic Catalog. Hello to Halo.
- Nisbett, Richard E., and Lee Ross. Human Inference: Strategies and Shortcomings of Social Judgment. Prentice Hall, 1980.
- Kahneman, Daniel. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011.
The next time this pattern surfaces, the move is not to fight it — it is to notice it. Naming Problem with Average creates a moment of pause before the decision. That moment is usually enough.