Thirty-Nine, Then Twenty-Five

In 1866 the German Confederation stopped being what it had been since 1815. It had held 39 states under an Austrian presidency. Prussia’s war with Austria ended it, and the North German Confederation took its place.

On 18 January 1871 Bavaria, Württemberg, Baden and Hesse joined a new German Empire of 25 member states.

That is usually told as three steps toward a country. It is also a warning about any series labelled Germany.

The 1815 Confederation, the North German Confederation and the Empire cannot be one reporting unit until somebody decides what the word covers. The decision is often sensible. It is still a decision, and it is rarely written down.

A sum called Germany on either side of 1866 contains a different collection of places, firms, people and revenues. The jump between the two sums is real as arithmetic. Nobody inside either collection experienced it.

The name outlasted the denominator #

Italy runs the same sequence, with the dates close enough together to make the change look instantaneous. The Kingdom of Italy was proclaimed on 17 March 1861 out of Piedmont, the Two Sicilies, Tuscany, the smaller duchies and part of the Papal States. Venice arrived in 1866. Rome was taken in 1870 and made capital the year after.

An annual total labelled Italy across that decade is therefore not a measurement of one territory. It may still be a useful total. It will also manufacture an acceleration, because the later aggregate contains Venice and Rome and the earlier one does not.

The chart can be drawn with complete accuracy and still show the wrong event. Not a change in a place — the arrival of a place in the numerator.

Historical geography does not treat the modern outline as the fix for this. It treats the dated sequence of entities as the record. The North German Confederation, and the kingdom that did not yet have Venice, are political geographies in their own right rather than defective drafts of the later map.

That is the apparatus a panel needs before it is a panel: every observation mapped to the unit that existed when the observation was made.

What the coefficient cannot check #

Regression is good at putting a number on a contrast. A difference-in-differences model makes the contrast explicit, with an interaction between a jurisdiction and a period, and turns a state-by-time code into a fitted effect.

It cannot tell whether the state-by-time code describes a stable state.

Every diagnostic in the standard kit answers a question about the estimator rather than about the coding. Gauss-Markov concerns the errors. R-squared concerns the residual sum of squares, and cannot fall when a predictor is added, so a boundary-date indicator improves the fit whether it marks a mechanism or a filing change. Neither has any way to notice that a region acquired new members halfway through the input. The code can satisfy every arithmetic requirement and still merge observations that were never in the same place.

The old map decided the case #

In October 1983 Burkina Faso and Mali put their land border to the International Court of Justice by special agreement. The Chamber’s judgment of 22 December 1986 was reached from French colonial legislative and administrative texts, from maps, and from colonial effectivités — the evidence of who had actually been administering what. No new survey was commissioned. The old record was the operative one.

The principle is uti possidetis juris: a new state presumptively inherits the administrative boundary it held at independence. The colonial map is not scene-setting there. It decides which ground is in dispute.

A historical revenue file needs the same things kept: the native reporting assignment, its effective date, and the crosswalk used for every later reconstruction. Without them an analyst can produce a clean series only by silently swapping one entity for another.

The result will usually reproduce. The same silent rule reproduces it. Reproducibility is not identity.

The repair happens before the fit #

There are three defensible views of a changed geography and they answer different questions. Native-period units preserve what was reported at the time. A stable crosswalk reallocates everything onto one chosen geography. Geographic intersections keep only the ground that exists on both sides of the boundary.

None of the three is the true series. Each makes its denominator inspectable, which is all that is being asked of it.

The test is not whether one reconstruction gives the sign you wanted. It is whether sign, scale and inference survive all three. A result that vanishes or reverses is a property of the boundary rule. A result that persists among entities never reassigned is evidence of something that happened on the ground.

The map that records a new jurisdiction changes what a regression estimates.

The record that preserves the old one is what says whether the estimate refers to anything that stayed put.