It's common to think of local knowledge as a smaller version of national knowledge.

We imagine information beginning in individual places and gradually moving upward. Enough observations become a survey, enough surveys become statistics, and eventually somebody looking across the whole field can see what no individual community could see alone.

There is truth in that picture.

There is also something important missing from it.

When information is aggregated, it doesn't merely become larger. It changes resolution.

Suppose we want to know whether a region is becoming more or less prosperous. At one scale we can look at employment, income, business formation, housing costs and other measures that allow one place to be compared with another. Those abstractions are useful precisely because they discard many local particulars. If every town described prosperity in its own terms, comparison would become extraordinarily difficult.

Now move closer.

A business owner may know that customers are postponing repairs they would normally have made immediately. A school may notice that the families leaving the district aren't the same kinds of families who left ten years ago. A volunteer fire department may discover that recruiting isn't difficult because fewer people care, but because fewer residents have jobs that allow them to leave work when a call comes in.

Those observations don't necessarily contradict the regional numbers.

They contain information the regional numbers weren't designed to preserve.

That's the distinction I find useful.

Local knowledge isn't merely national knowledge waiting to be collected, because aggregation can't retain every property of the observations from which it was built. Some information survives compression remarkably well. Other information depends on relationships, sequence, unusual cases or knowledge of what normally happens in one particular place.

The same trade occurs in the other direction. A person living in one town can accumulate extraordinary knowledge about that town while remaining unable to see a pattern occurring simultaneously across fifty other places.

Different resolutions reveal different things.

This is easy to understand with something as ordinary as a map. A highway map can show me how to travel from Albany to Buffalo far better than a detailed street map of one neighborhood. The neighborhood map can show me an alley, driveway or footpath that would be useless clutter on the statewide map.

Neither map is more real.

Each preserves information appropriate to a different question.

Human knowledge becomes harder because we don't always remember that we're using maps.

Someone who has lived in a community for forty years may know which roads usually flood first, which organizations actually cooperate during an emergency and which official procedures are routinely supplemented by informal arrangements. That knowledge can matter enormously when something goes wrong.

It can also be wrong.

The road that always flooded first may have been rebuilt. The family everyone assumes will help may no longer have the capacity. A reputation formed twenty years ago may continue circulating after the person or institution has changed.

Experience provides information.

Time can deepen it, and time can make parts of it stale.

That is why I don't want to give local knowledge a special exemption from correction. Being close to something doesn't guarantee that we understand its causes, and lived experience doesn't make conflicting interpretations simultaneously accurate. People standing in the same town can experience the same change differently and tell incompatible stories about why it happened.

Those disagreements are information too.

The value of local knowledge is not that it arrives already correct. Its value is that it can preserve distinctions that disappear when we zoom out.

Large-scale knowledge has the complementary advantage. It can reveal comparisons that no local observer could produce alone. A town may believe its young people are leaving because of something uniquely wrong there until regional data shows similar demographic movement across dozens of communities. A business owner may interpret declining sales as evidence of local economic weakness while businesses elsewhere are experiencing the same industry-wide change.

The wider view can rescue us from local explanations that became too convincing.

The local view can rescue us from averages that became too complete.

Problems begin when information only moves one way.

A central organization creates categories so that thousands of cases can be compared. Those categories work well enough to produce a useful model, and eventually the model becomes the language through which new observations must enter. Anything that doesn't fit becomes difficult to report because there is no field for it.

At first, people improvise.

They put an explanation in a notes box. They send an email. They mention the unusual case during a meeting. Someone says, “That's not quite what's happening here,” and tries to explain the distinction.

Whether the larger system can learn now depends on what happens to that exception.

Sometimes the category is still good and the unusual case really is unusual. We shouldn't redesign every system around every exception it encounters. Other times several exceptions accumulate around the same missing distinction, and what looked like noise begins telling us something about the model.

The problem isn't abstraction.

The problem is abstraction without a return path.

Any large system needs compression. Governments, businesses, researchers, news organizations and software systems all need categories that allow them to work with more information than any individual person could hold in detail. The question is whether observations that don't fit those categories can eventually modify them.

That is feedback.

Local knowledge can be particularly valuable in that process because consequences often become visible in particulars before they become obvious in aggregates. A policy may work well for most places while interacting badly with some feature of one community. A statewide program can improve an average while creating a specific problem that the average doesn't reveal.

Neither observation automatically cancels the other.

We need to know both.

This is also why telling people that the data contradicts their experience often produces more heat than understanding. Sometimes the data really does reveal that an intuitive local story is mistaken. That happens, and we shouldn't pretend otherwise.

Other times the disagreement exists because the measurement and the experience refer to different things.

“The economy is improving” may describe a regional change in employment and income.

“I can't afford to live here anymore” may describe one household's relationship between wages, housing and other expenses.

Those statements can coexist without either becoming dishonest.

The next question is what each one actually measures.

Local journalism has an interesting position inside this exchange because it can move between resolutions. A reporter hears the individual account, attends the local meeting, notices the business closure or receives the phone call from someone who says the official description doesn't match what they're seeing.

That is one observation.

The reporter can then look outward. Does the record support it? Are other people seeing the same thing? What do the numbers show? Is this community unusual? Is the apparent local change part of something larger?

Then the reporting can return.

Perhaps the broader evidence supports the original observation. Perhaps it contradicts it. More often, it changes the question enough that the original story becomes more specific.

That movement matters more to me than deciding whether local or centralized knowledge deserves priority.

Local knowledge without comparison can become folklore.

Aggregate knowledge without return can become abstraction that no longer notices what it discarded.

The useful relationship is recursive.

Particulars move outward far enough to be compared. Comparison produces patterns that can be brought back to particular places. The people living there can then see whether the broader explanation actually describes what is happening around them.

If it doesn't, we have another observation.

We don't need to decide in advance which scale gets the final word.

This is especially important because some local knowledge really is difficult to transfer. Knowing that a road floods is easy to record. Knowing that it floods differently when the ground is already saturated after several warm winter days may require more context. Knowing which informal relationships become important during the resulting emergency may be harder still.

We can document more of that knowledge than we sometimes assume.

We simply shouldn't confuse documentation with perfect transfer.

The person receiving the information elsewhere doesn't inherit the years of experience through which the original observer learned what mattered. They receive a representation of it.

That representation may be excellent.

It may also leave something out.

The answer isn't to declare the missing part inaccessible to outsiders. It's to preserve enough contact with the source that questions can travel back.

That may be the boundary worth protecting.

A healthy information system doesn't require every decision to remain local, and it doesn't require every local observation to be accepted as truth. It needs a way for information to move across scales without pretending that movement is lossless.

Some things become clearer when we zoom out.

Other things disappear.

Then we zoom back in and see what the larger picture taught us about the place where we started.

Knowledge doesn't become complete merely because we've gathered more of it.

Sometimes understanding comes from changing the distance and noticing what appears—and what vanishes—each time we move.