People by US state by year: an organized year-by-year guide

Plot a biographical dataset by year and state and you get a chart with structure in it: rises, dips, a spike here, a long flat stretch there. Almost none of that structure is about people. A yearly series drawn from a reference dataset is, first and foremost, a series about the dataset.

Reading one properly means knowing which year is on the axis and what makes the line move.

What to take away

Historical dates are frequently reconstructed rather than registered, and reconstruction has a signature: values pile up on years ending in zero and five.
Be explicit about which quantity the chart shows.
State which year field is on the axis, in the chart itself.
Locating discontinuities in a collection, which is the fastest way to understand how a dataset was built.

Five different years, five different charts

One person carries several dates, and each produces a different series from the same records. The state pillar sets out which custodian holds each of those dates.

Year plotted	The series describes	Where it misleads
Birth year	A cohort	Nothing that person did appears near their position on the axis
Event year	When something happened	Requires the event to be dated, which thins the set
Publication year	When a work appeared	Lags the work, sometimes by decades
Record year	When a document was created	Registration lag, and delayed filings
Entry year	When the database row was added	Purely a fact about the project

Mixing these inside one chart is common and invisible. If a dataset has a single "year" column assembled from several sources, find out what it was populated from before you plot it.

Six things that move a yearly count

Only one of them is the world.

An ingest. A collection is imported, a partner institution's records are added, another language edition is merged. The line jumps on the import date, spread backwards across whatever years the collection covers. This is the largest single cause of structure in these charts and the easiest to check, because projects keep release notes.

A rule change. An inclusion threshold is revised and a whole class of entries appears or disappears at once: across all years simultaneously, which is the tell.

A digitization program. A funded effort to scan one archive lights up one state and one period. The resulting bump is a fact about a grant.

Documentation lag. Coverage of any year keeps accumulating for decades afterwards. Recent years are always sparse, and the sparseness is not a decline.

A jurisdiction change. Territorial status, statehood, county reorganization, or a change in which authority registered births all shift where records land. The state timeline page covers the mechanics.

A registration law. Statewide registration of births began at different times in different states, and early compliance was partial. A state's series will show an artificial onset at whatever year its records begin to survive in quantity, which is why federal guidance on vital records points at state offices rather than a national register.

Before interpreting any feature of the chart, check these six. If the project publishes a change log, plot it as annotations on the same axis. Most apparent history dissolves.

The two edges are both unreliable

The right-hand edge always slopes down. Recent years have had less time to accumulate documentation, coverage, and reference-work entries. Nothing has declined. Truncate the last several years, or shade them and label them incomplete, and never fit a trend through them.

The left-hand edge is thin for the opposite reason: record survival falls off going back, unevenly by state and by county. What looks like a smooth ramp is usually the aggregate of many step functions, each one a registry starting or a courthouse burning. The holding agency's guide to census records is explicit about which rounds survive and which do not.

Round numbers are over-populated

Historical dates are frequently reconstructed rather than registered, and reconstruction has a signature: values pile up on years ending in zero and five. Ages reported in censuses and depositions show the same digit preference, and any birth year derived by subtracting a reported age inherits it.

So a yearly series built from reconstructed dates will have regular spikes at round years. That is a measurement artifact, not a birth pattern. Two defenses: keep the approximation marker in the data rather than storing a bare integer, and check whether your spikes fall on multiples of five before looking for a historical explanation.

Stock, flow, and the thing being counted

Be explicit about which quantity the chart shows.

A flow is how many events fall in a year: births, deaths, publications.
A stock is how many people were alive, resident, or active in that year, which requires spans rather than points and is a different calculation entirely.
A cumulative count rises forever by construction and says nothing at all about any individual year.

Charts that silently switch between these are common. A "people by state by year" series is nearly always a flow of records, and readers nearly always interpret it as a stock of people. Period bucketing fails the same way, as the century pillar describes.

Presenting a yearly series honestly

State which year field is on the axis, in the chart itself.
Annotate every known change to the collection: ingests, rule changes, digitization programs, source additions.
Break the line at method changes rather than drawing through them. A continuous line asserts continuity you do not have.
Mark incomplete recent years distinctly and exclude them from any trend.
Render missing years differently from zero years.
Give counts, not just rates, so a reader can see when a dramatic percentage rests on a handful of records.
Say in one sentence what the chart is evidence about, the dataset, and what it is not evidence about.

What a year-by-year view is genuinely good for

Locating discontinuities in a collection, which is the fastest way to understand how a dataset was built.
Assessing coverage before using the data for anything else: where is it thick, where is it empty, and from when.
Comparing a state against itself over a period where the collection method did not change.
Finding candidate years for archival research, which is what these catalogs were built to support.

What it cannot show

How many notable people a state had in a given year. That quantity is not in any dataset.
A comparison between states with different registration histories, which is most pairs.
Any per-capita rate for early periods, where the population denominator is itself a reconstruction.
Anything that would support a ranking of places by the people counted in them. The reasoning is in the maps and data page, and the short version is that such a ranking orders archives, not people.

Bottom line

A year-by-year state chart is a diagnostic instrument pointed at the dataset. Used that way it is genuinely informative, and it will tell you more about how a reference work was assembled than any documentation will. Used as history, it converts import schedules and registration laws into apparent facts about places. Label the axis, annotate the changes, cut the last years off, and say what the line is made of.

Common questions

How many recent years should be cut off the right-hand edge?

Enough that the slope stops being an artifact, and the number depends on how fast your sources accumulate. Say what you cut and why, rather than picking a round number silently.

Is the digit preference at years ending in zero fixable?

Not in the data. It can be shown: plot the last digit of the year and the spike is obvious. That is more honest than smoothing it away.

Should an ingest be plotted as an annotation or excluded?

Annotated. Excluding the affected years hides the fact that the series records collection activity as much as history.

Can a per-year state series be compared between two states?

Only where both registration histories and both coverage histories are known and similar. Otherwise the comparison measures two registration regimes.