Guides
Part of People by country: a sourced reference guide for 2027
People by country by year: an organized year-by-year guide
People by country by year: six events break a country series, two codings answer different questions, and the left edge is usually the source starting.
Plot anything by country across a century and you have plotted a series over a set of containers that kept changing shape. States split, merged, appeared, were absorbed, and swapped territory without ceasing to exist. The line on the chart is smooth because the chart cannot show any of that.
This page is about what to do with that fact. Which year to put on the axis, and the six things that move a count within a stable container, are covered on the state year page and are not repeated here.
What to take away
- A country series is a series over changing containers. Every container event is a discontinuity, and hiding it is the default behavior of every charting tool.
- Coding to the entity that existed at the time and coding to the entity that covers the ground now are both legitimate. They answer different questions and must never be mixed in one column.
- A series that starts at a round year almost always starts where the source starts, not where the subject does.
A world map dated to one year is a list of the containers that existed that year. Every country series is a series of containers, and the containers changed more often than the published charts admit: the code list behind them, ISO 3166-1, keeps a change history for exactly this reason.
Six events that break a country series
| Event | What happens to the series |
|---|---|
| Split | One line must become two, with no defensible way to divide the earlier years |
| Merger | Two lines become one, and the sum is not comparable to either |
| Independence | A line begins, while the same territory's earlier values sit inside another line |
| Absorption | A line ends, and its values continue inside a different container |
| Rename | The label changes with no change in the thing, which tools treat as a new series |
| Territorial transfer | Both containers persist, both change footprint, and neither line breaks visibly |
The last one is the dangerous case. Nothing in the data announces it. Both entities exist before and after, both keep their codes, and a quantity has moved between them for reasons that have nothing to do with the subject being counted.
Two codings, two questions
Contemporaneous coding assigns each record to the entity that existed when the record was made. It answers: what did the authorities of the day record, under the names they used? It is the honest coding for archival work, and it produces a chart with entities appearing and disappearing.
Backcast coding assigns each record to the modern entity covering the same ground. It answers: what happened on this territory, however it was governed at the time? It produces a chart with continuous lines and no visible history.
Both are useful. Neither is the truth. The failure is publishing one while implying the other, which is what happens when a modern country list is applied to historical records without comment, and it is why a register of former entities with start and end dates has to sit beside the data, as set out on the country list page.
Where it matters, code twice and store both. It costs one column and settles most arguments before they start.
Presenting a discontinuity
The rules are simple and almost never followed:
- Break the line. Do not draw across a container event. A gap is information.
- Annotate the break with the date, the event and the treatment applied.
- Never interpolate across it. An interpolated value between two different containers is a number that describes nothing.
- Publish the change log as part of the chart, not as a footnote in a methods appendix nobody opens.
- Recompute rather than splice when a definition changes, and if you cannot recompute the earlier years, say the series begins where the definition begins.
A change log is a small table with a date, the event, the entities before, the entities after, and the treatment you applied. Anyone who has ever inherited someone else's time series will tell you it is the most valuable object in the file.
Where the series really starts
Two artifacts show up at the left edge of nearly every long series.
The first is the source boundary. A series that begins at a round year usually begins there because that is where a digitized run, a survey, or an archive's holdings begin. The apparent growth in the first years after that point is the source ramping up, not the world changing, which is survivorship bias drawn as a trend.
The second is dating itself. Historic records use different calendars, and the year did not always begin in the same month everywhere, so a date lifted from a source can land in the wrong year without anyone doing anything wrong. Where precision matters, record what the source wrote and the conversion separately. The general treatment of dates and of precision that a source never claimed is on the century page.
What such a series can and cannot show
It can show what a defined corpus contains, year by year, for a stated set of entities under a stated coding, with the container events marked. That is a genuinely useful object: it shows a researcher where the material is thick and where it is thin.
It cannot show output, achievement or national character. A count of records rises when archives are cataloged and falls when a war destroys a series, and both movements look exactly like changes in the population. Nor does it support a ranking of countries in any year, which is a page we do not publish for reasons that survive every improvement in the data.
For living people, nothing here changes the rule: we assert no one's nationality, ancestry, religion or birthplace, and a year-by-year table is not a way around that.
Bottom line
The x-axis is the easy part. The hard part is that the series on the y-axis belongs to entities that were created, merged, renamed and redrawn while the data was being collected. Break the line at every one of those events, say which coding you used, and let the chart show the history instead of smoothing it away.
Common questions
Can two countries be compared within a single year of the series?
Only if they were coded the same way and their corpora are comparable, which they rarely are. Two states with different archival histories produce different counts from identical populations. A within-year comparison is a comparison of record keeping, and the country pillar sets out how that bias is structured.
What about counting by decade instead of by year?
Aggregating hides the noise without removing the bias, and it moves container events inside the buckets where they stop being visible. If you aggregate, keep the change log at annual resolution.
How should a year with no data be shown?
As a gap, labeled as no data. A zero on the axis says the count was measured and came out at none, which is a claim about the world rather than about the file.