How People by US City Rankings Are Built and Why the Order Changes

A city ranking of famous people is a construction, and it can be shown as one. Below is a worked ranking of eight cities, the rules it was built on, and the four points where a different rule would have printed a different order.

The counts are illustrative, drawn from named public sources, and labelled as such. Substitute your own figures and the method still holds.

What to take away

The worked ranking, and the rules behind it

The table below ranks eight cities on one stated relation: the person's birthplace, as recorded in a public biography index. The count is the number of entries in that index whose birthplace field names the city.

The index is Wikidata, queried on its place of birth property (P19). The figures in the table are illustrative and are not a live query result, so they carry no access date. A real run prints its access date with the total, because a live database changes by the day.

Ties are broken by the founding year of the anchor institution; where two institutions share a year, the city name decides, alphabetically.

Read the fourth column before the first. Each city's total tracks the age and size of its collecting institutions and the digitization of its newspapers. Boston and Philadelphia lead because their records were kept, catalogued and scanned early, not because they produced more notable people per resident.

The fifth column is the trap in any dated table. A corpus weighted to 1750 to 1900 will rank colonial ports high and twentieth-century migration hubs low, and that is a property of the index, not of the cities.

Where the ranking breaks, point by point

The relation is a choice. Run the same eight cities on residence at mid-career and New York and Chicago climb while Charleston drops to last place. The person did not move. The rule did. This is the same problem that makes country-level groupings unstable, one level down.

Re-ranked on residence at mid-career, on the same index, the same eight cities give the order below. The figures are illustrative, read from the residence property (P551).

Only the relation changed. The index, the corpus and the cities are all the same.

The birth certificate is not the city line. A modern birth is recorded where the delivery happened, which is a hospital.

A hospital serves a catchment that crosses municipal lines, so a birth recorded in a city hospital may belong to a family living in a suburb the city never annexed.

A city that annexed its suburbs carries people its neighbour counts as suburban.

The denominator is a choice too. Municipal, urban, metro and county populations differ by large factors for the same place. Divide the table by any two of them and you get two orders. A count spanning two centuries divided by a present-day population compares a numerator from one era to a denominator from another.

Small totals are noise. Charleston's 98 and St. Louis's 121 sit within a few editorial decisions of each other. One index adding a batch of entries reorders the bottom half of the table. A measure that moves that much on one cataloguing choice is measuring the cataloguing.

Why the count tracks archives, not people

Every source you could count from is a record of attention. Five effects push the same way.

• Digitization is uneven. Cities whose newspapers were scanned are searchable; cities whose newspapers sit on microfilm are not, a limit set out in library guidance on digitized newspaper collections.

A table built on these effects reproduces the shape of the archive with high precision. That is survivorship bias printed as a league table.

What the year in the title cannot deliver

A ranking labelled with a year implies three things: that something was measured, that it was measured on a stated date, and that a later run would use the same method.

None of that is available here. With no fixed instrument, there is no measurement to date.

A dated ranking would need a frozen query, a fixed corpus and a rule for late corrections to the index. Without those three, the year in the title is version control, not a measurement. Two runs are comparable only when the query, the corpus and the correction rule are the same.

Reading somebody else's city ranking

Ask five questions before you look at the order: which relation, which geography, which corpus, which date, which tie rule. A list that answers all five is a legitimate statement about that corpus.

This is the discipline of reading notability maps and data without fooling yourself, and it is the same whether the table covers eight cities or eight hundred.

The same reasoning applies to state league tables, and the state pillar sets it out in the same terms, with extra force for living people, whose birthplace and residence we do not assert.

Common questions

Could the ranking be built better?

Better coverage would reduce one bias and leave the grouping problem untouched, because the grouping problem is definitional. A person genuinely attached to four cities cannot be assigned to one of them by any improvement in the archive. You can improve the count; you cannot make the assignment true.

Is there an honest ordering in this subject?

Yes, of documents. Archives can be ordered by holdings, digitization coverage or date range, and that is useful because it tells a researcher where to look. What you cannot do is relabel it as an ordering of places. The record layers this depends on are set out on the city pillar.

What can one document prove about a person's city?

A birth certificate proves where a birth was registered, which is a hospital or a registrar's district, not a family home. A census entry proves where someone was enumerated on one night. A probate record proves where an estate was administered.

Each supports one relation, and none of them settles belonging.

For a contested attribution, ask the holding archive or a credentialed genealogist.

How should the table above be cited?

As an illustration of method, not as a finding. The counts are illustrative, the sources are named so you can check their scope, and the ranking changes with the relation, the geography and the corpus. Cite the method and the source list, never the order.