Chapter 1 — What the Average Leaves Out

What the Average Leaves Out

At 8:30 on a Friday morning, the Bureau of Labor Statistics releases the monthly employment report. The document arrives with tables, definitions, revisions, sampling notes, and distinctions between industries that added jobs and industries that lost them. It separates people who are unemployed from people who have stopped looking. It distinguishes full-time work from part-time work, temporary layoffs from permanent ones, and changes that appear meaningful from those that may still fall within the uncertainty of the survey. The report is not a verdict. It is an attempt to describe a moving system from several angles at once.¹

Within minutes, the report begins to shrink. A news alert reduces it to a number. A headline turns the number into a direction: hiring strengthened, hiring weakened, the economy surprised, the economy disappointed. By midday, television panels have turned the direction into a judgment about whoever currently holds power. By evening, the report has become a mood—the economy is strong, or the economy is failing.

Nothing in the original tables required either sentence. Yet some sentence like them was almost inevitable. Millions of people cannot coordinate around hundreds of pages of statistical detail. Legislators, employers, investors, journalists, and households need something smaller. They need a number they can carry into another conversation.

This is what averages are for: they make a large reality portable. The trouble begins when the portable version is mistaken for the reality itself.

A national employment report can show steady hiring while a factory cancels its second shift. Inflation can slow while a renter opens a renewal notice that demands another two hundred dollars a month. Wages can rise while a family falls further behind because childcare, insurance, and interest payments rose faster. None of these local experiences disproves the national report. The report does not disprove them either. The chart and the receipt are measuring different things, though we often speak as though one of them must win.

The Work of Compression

No one experiences the economy as a whole. We experience fragments of it: a paycheck, a mortgage rate, a delayed shipment, an empty storefront, a job offer, a medical bill, a quieter restaurant, a busier warehouse. Even the people who study the economy professionally encounter it through constructed measures. There is no platform high enough from which the entire system becomes directly visible. Economic knowledge therefore depends on compression.

Gross domestic product gathers countless transactions into a measure of production. The unemployment rate turns millions of individual employment conditions into a proportion. The Consumer Price Index follows the changing cost of a defined basket of goods and services so that price movements can be compared over time. These measures are not crude substitutes for better knowledge. They are among the ways better knowledge becomes possible.

Without compression, public argument would collapse into incompatible anecdotes. One town would report that restaurants could not find enough workers. Another would report that a plant had stopped hiring. A homeowner with a mortgage fixed at a low rate would describe stability. A first-time buyer facing a much higher monthly payment would describe exclusion. Each account might be true, but without some method for combining them, no one could tell whether they represented a broad change, a regional exception, or a temporary disturbance.

The average gives these experiences a shared reference point. It allows a central bank to ask whether price pressure is spreading across the economy. It allows lawmakers to ask whether an emergency response is still needed. It allows employers to compare what they are seeing with conditions beyond their own order books. It helps us notice patterns that no receipt, workplace, or neighborhood could reveal alone. The chart sees what the receipt cannot.

This is why the familiar complaint that economists should stop using averages misses the problem. A country of hundreds of millions of people cannot be understood one household at a time. Coordination requires reduction. Every public description must leave something out. The question is not whether we compress; the question is what happens when we forget that we have done it.

A measure designed to provide orientation can slowly become a substitute for the world it summarizes. A headline that should mean this is the broad direction begins to sound like this is what is happening to you. When people object, the objection is treated as confusion about the number rather than information about what the number does not contain. Orientation becomes substitution.

That shift can occur without anyone intending to deceive. It happens through the ordinary pressure to speak clearly, confidently, and briefly. A cabinet official has several minutes. A television segment has several seconds. A campaign needs a sentence that can survive repetition. A company earnings call must turn a complicated organization into a coherent story before analysts begin asking questions. Under those conditions, the qualification is usually the first thing to disappear.

Employment is strong, though the gains are concentrated in particular sectors. Inflation is easing, though the price level remains far above the one households remember. Wages are rising, though housing and borrowing costs have changed what those wages can buy. The first clause travels. The second clause remains in the report.

Over time, audiences notice this pattern. They learn that the public sentence is not simply a description. It is also an attempt to establish a mood, defend a record, reassure a market, or weaken an opponent. Economic language begins to sound less like a map and more like a signal. The number may still be correct. What changes is the question listeners ask of it—not What does this tell me? but Why are they telling me this now?

The People in the Tail

An average does not conceal variation because it has failed. Concealing variation is part of what an average does.

Suppose rents remain nearly flat in one large metropolitan center while rising sharply in surrounding counties. A regional summary may accurately report moderation. It may even describe the experience of more households than it misses. But the family signing a lease in one of the faster-rising counties does not live in the center of the distribution. Their rent is not made less real by being statistically unusual. They live in what analysts call the tail. Tails sound remote when represented on a graph; they do not feel remote from inside them.

A household with unusual medical needs, a rural contractor dependent on diesel, a family paying for specialized childcare, or a new homeowner buying insurance in a high-risk area may experience an economic pattern very different from the national one. Their expenses are not random objections to the data. They are coordinates the summary was not designed to foreground.

This is especially visible in housing. Shelter measures are built to track broad changes over time, not to predict the exact renewal letter received by a particular tenant. Local inventory, taxes, insurance, ownership patterns, construction, and migration can cause housing costs to move differently across regions and even across neighborhoods.² A national measure can therefore cool while a household’s rent jumps sharply. The statistic remains sound, and the experience remains painful.

Yet public debate often treats this mismatch as though one side must be mistaken. Officials point to moderation in the index. Households point to what they owe next month. Each side suspects the other of refusing reality. The deeper problem is that they are using the word reality for two different scales.

The index is designed to answer a collective question: what is happening to prices across a large and varied economy? The renewal notice answers a personal one: what will it cost me to remain in my home? Neither question can replace the other.

The same pattern appears in wages. A report may show that average hourly earnings have risen. For some workers, those gains represent real improvement: a better position, greater bargaining power, a chance to rebuild savings. For others, the wage increase arrives after rent, food, insurance, and debt payments have already moved higher. They hear that wages are up and look at what remains in their checking account. “Wages rose” and “I am still behind” can be true in the same household.

This is difficult for public language because public language prefers one direction at a time. Conditions improved or deteriorated. Policy worked or failed. The economy is resilient or fragile. But many economic experiences are not moving in one direction. A saver may benefit from higher interest rates while a borrower is squeezed by them. A homeowner with a fixed mortgage may feel protected while a younger family is priced out of buying. A worker may receive a raise while fearing that the employer will cut hours. A small business may see strong sales and shrinking margins at once. The economy does not distribute its meanings evenly.

This is one reason averages can feel dishonest even when they are methodologically defensible. People do not necessarily believe the arithmetic was fabricated. They sense that the arithmetic has been asked to carry a claim it cannot support. The average describes a center; the public message often turns it into a universal.

When the Headline Becomes a Rebuttal

There is a difference between using an aggregate to place an experience in context and using it to rebut the experience. The first says: Your rent rose sharply, even though the national measure is beginning to moderate. Let us understand why those clocks are different. The second says: Inflation is cooling, so your sense that costs remain high is mistaken. Only the first leaves room for inquiry.

The second may never be stated so bluntly. It can appear through sequence. A leader begins with the favorable national indicator, spends several paragraphs explaining its significance, and acknowledges household pain only after being challenged. By then, the structure of the message is clear. The pain is being treated as a complication to the real story rather than part of the story itself. People notice which truth had to be dragged into the room.

This is why adding more nuance does not automatically restore trust. Institutions often respond to criticism by publishing additional charts, methodological explanations, regional breakdowns, and confidence intervals. Those materials can be valuable. But the disagreement is not always caused by a shortage of information. Sometimes the real question is whether the speaker recognizes the listener’s coordinate without being forced to.

A town hall can include twenty accurate slides and still fail in its opening minute. If the first slide celebrates national improvement while the people in the room are worried about insurance, rent, or shortened shifts, the presentation has already told them where their experience ranks. Every later qualification sounds like damage control. The failure is relational before it is statistical.

People are not only asking whether a number was calculated correctly. They are asking whether the people using the number understand what it leaves out. When that understanding is absent, the summary begins to feel like a morale operation. Good news is amplified because confidence is useful. Pain is treated as anecdotal because acknowledging it complicates the message. Revision is delayed because changing the story may look like weakness. This does not require a conspiracy. It requires only ordinary institutional incentives.

Markets often reward confidence. Political coalitions reward clarity. News formats reward conflict. Social platforms reward statements that can be classified immediately as hopeful or alarming, loyal or disloyal, competent or incompetent. A sentence that says conditions are improving unevenly carries less emotional direction than a sentence that says the economy is booming or broken. The thinner the channel, the more pressure there is to remove the second truth.

The original employment release may contain uncertainty, sector detail, and revisions. The headline contains a number. The campaign contains a verdict. By the time the report reaches the public argument, a map has become a flag.³ This is how explanation slides toward signaling.

The audience then performs its own correction. People compare the public claim with the information nearest to them: the warehouse floor, the grocery total, the waiting room, the work schedule, the rent notice. When the two do not match, they trust the thing they can touch.

That response has limits. A household’s experience is not a representative sample. A busy restaurant cannot establish national prosperity. A friend’s layoff cannot prove national decline. Personal experience is vulnerable to recency, selective attention, and the tendency to generalize from vivid events. The receipt does not see what the chart can. But dismissing the receipt because it is incomplete repeats the original error. The chart is incomplete too.

The task is not to decide which partial view deserves exclusive authority. It is to understand what each one can reveal and what each one necessarily misses.

Two Clocks

A useful economic statement often needs two clocks. One clock tracks the aggregate: employment, inflation, growth, credit, or output across a large system. The other tracks transmission: when that change reaches a particular region, industry, firm, or household. The clocks rarely move together.

Inflation may begin to slow before households experience relief because slower inflation means prices are rising less quickly, not that they have returned to their earlier level. Freight indexes may normalize before a small distributor receives a specialized part that has been delayed for months. A labor market may remain strong nationally while a particular employer cuts overtime. Interest-rate policy may reduce future inflation while immediately increasing the cost of borrowing.

Public communication often names only the faster clock. A national measure changes. A headline announces a turn. The slower consequences continue in kitchens, storefronts, and balance sheets. People are asked to accept the new story before their own circumstances have caught up with it. This creates an avoidable form of disbelief.

A leader does not need to deny national improvement to acknowledge that its arrival is uneven. The sentence simply has to hold both scales: inflation is easing, but many households are still paying prices that rose faster than their incomes; employment remains strong nationally, but this region is losing overtime and new orders; supply chains have improved broadly, but specialized components are still delayed for some firms. These sentences are less triumphant. They are also more durable. They give the listener a place inside the description.

The purpose is not to attach a ritual expression of empathy to every statistic. “We know some families are struggling” has become its own kind of compression, vague enough to acknowledge pain without allowing it to alter the story. Structural recognition requires greater specificity. It names the mechanism through which the aggregate reaches people unevenly: the index slowed while the price level remained high; the rate rose while the locked-in homeowner and the new buyer entered different economies; employment grew while the gains arrived in sectors and regions that did not include everyone. This is not a concession to anecdote. It is a more complete use of the data.

Distribution is not an objection to economics. Distribution is where economics becomes a life.

The Report at the End of the Day

By the end of Friday, the employment report released that morning has traveled far from its original form. The technical document still exists. Its tables have not changed. Its qualifications remain available to anyone willing to read them. But most people will encounter the report through the smaller stories built from it.

One worker will hear that the labor market is strong while wondering whether her contract will be renewed. Another will hear the same news after receiving a better offer. A business owner will interpret strong hiring as evidence of demand. Another will interpret it as evidence that labor will remain difficult to find. The report cannot contain every one of these experiences. It was never meant to.

What public speech can do is acknowledge the distance between the measure and the person hearing it. It can present the number as an orientation rather than a verdict. It can name the broad direction without pretending that everyone is traveling in it at the same speed. This is a modest discipline, but it asks leaders to surrender something public life rewards: the clean story.

The clean story says the economy is good or bad. The more honest story says an economy can improve in aggregate while remaining punishing in particular places; that stabilization can be real without feeling like recovery; that a person can misunderstand the national pattern and still understand their own bills perfectly well.

The chart deserves a place in the conversation because no one household can see the whole. The receipt deserves a place because no national average can live anyone’s life.

A summary becomes deceptive not when it is false, but when it asks to be mistaken for the whole.

Footnotes

1. U.S. Bureau of Labor Statistics, The Employment Situation, including documentation concerning household and establishment surveys, sampling, revisions, and sector-level reporting, 2020–2024. ↩

2. U.S. Bureau of Labor Statistics, Consumer Price Index shelter components and regional indexes, 2020–2024; U.S. Census Bureau, housing-cost and rental-market materials, including the American Community Survey. ↩

3. Robert J. Shiller, Narrative Economics: How Stories Go Viral and Drive Major Economic Events (Princeton: Princeton University Press, 2019). ↩