Abhishek S.
Shipping in public. Listening in private.

Abhishek

I lead women’s Indo-Western & Premium at Max Fashion. I also wrote the AI that runs the buying floor.

Rare profile. Category operator who ships production code.

Senior Buying Leader · Max Fashion Women’s Indo-Western & Premium · 530+ India stores NIFT ’12 · On the women's wear floor

abhishek@bengaluru ~ %
>role: senior buying lead
>dept: women’s indo-western + premium
>floor: 530+ stores india

Alabama Paradox

The 1880 census produced a table in which a 300-seat House gave Alabama 7 seats, while a smaller 299-seat House gave it 8. Alabama’s population did not fall. One extra seat changed the order in which fractional claims were rounded.

The Alabama paradox is not ordinary rounding error. It is a reversal created when several rounding errors compete for the same indivisible object.

How one extra seat removes a seat

Hamilton’s method gives state (i) an exact quota:

q_i(h) = h × p_i / P

Here, (h) is the House size, (p_i) is the state’s population, and (P) is the total population. Each state first receives floor(q_i). The remaining seats go to the states with the largest fractional remainders:

r_i(h) = q_i(h) − floor(q_i(h))

Every exact quota rises when (h) rises. The remainders do not. Each remainder climbs toward 1, resets near 0 when its quota crosses an integer, and then climbs again. Changing (h) therefore reshuffles a ranking.

A population split of 53:33:14 exposes the mechanism:

State Quota at 10 Seats at 10 Quota at 11 Seats at 11
A 5.30 5 5.83 6
B 3.30 3 3.63 4
C 1.40 2 1.54 1

With 10 seats, the whole-number allocations consume 9 seats. State C has the largest remainder, 0.40, so it receives the tenth.

With 11 seats, the whole-number allocations consume 9 seats again. The two spare seats go to A and B because their remainders, 0.83 and 0.63, now outrank C’s 0.54. C’s exact claim rises from 1.40 to 1.54, yet its allocation falls from 2 to 1.

This failure has a precise name: Hamilton’s method is not house monotone. A house-monotone rule never takes a seat from any state when the total number of seats increases.

What the 1880 calculation changed

The February 7, 1882, Congressional Record preserved Census Office official C. W. Seaton’s account of calculations for House sizes from 275 to 350. His table assigned Alabama 8 seats at 299 and 7 at 300. These were alternative House sizes, not two consecutive legislatures, but the comparison exposed a defect in the rule used to construct the table.

The United States did not move directly from Hamilton’s method to today’s formula. The Census Bureau records the sequence as Vinton or Hamilton from 1850 through 1900, major fractions for the 1910 and 1930 apportionments, and equal proportions from the 1940 apportionment onward.

Equal proportions belongs to the family of divisor methods. It prevents an Alabama reversal, but it can assign a state fewer than its lower quota or more than its upper quota. The paradox disappeared because the rule changed which promise it was willing to break.

What’s contested

Nobody disputes the arithmetic. The dispute is over which guarantee deserves priority: quota, house monotonicity, or population monotonicity.

Method Respects quota House monotone Population monotone
Hamilton Yes No No
Balinski–Young quota method Yes Yes No
Equal proportions Not always Yes Yes

Balinski and Young’s 1975 quota method matters because it blocks a tempting overstatement. The Alabama paradox does not prove that quota and house monotonicity are incompatible. Their rule can preserve both. Under the standard axioms developed in their later work, the deeper conflict lies between quota and population monotonicity: a rule cannot guarantee both for every population change.

Frequency adds another uncertainty. Janson and Linusson proved that, under their asymptotic model, the expected number of states harmed when one seat is added is at most 1/e, about 0.368. When population shares are themselves sampled uniformly and the number of states grows, their expectation approaches about 0.123. Those figures describe mathematical models, not the historical frequency among legislatures.

Randomization changes the bargain again. Correa, Gölz, Schmidt-Kraepelin, Tucker-Foltz, and Verdugo showed in 2024 that randomized rounding can respect quota in every realized allocation while matching exact proportionality in expectation. Their work also finds harder failures at the coalition level: improving several parties’ shares need not improve every probability that the coalition crosses a seat threshold.

Why this has to do with other realms

Apportionment is a form of quantization. concept information theory asks what disappears when a continuous signal becomes a finite code; Hamilton’s method shows that quantization can discard monotonicity as well as precision. The lost information is not merely a decimal. It is the order in which claimants receive physical units.

Biology performs a related conversion in concept genetic drift. An allele frequency is continuous, but a generation contains an integer number of offspring. Wright–Fisher sampling makes the rounding random rather than deterministic, yet the structural question survives: when expected shares cannot all become exact counts, who absorbs the error?

concept braess paradox supplies a third version. Adding one road can increase journey times because the new road changes route rankings. Alabama loses a seat and drivers lose time for the same broad reason: increasing capacity changes the allocation rule’s internal order.

An open question

A lottery can give a state its exact quota on average across 100 apportionments, but a legislature lives through only one. When does fairness in expectation justify randomness in the single allocation that governs the next decade?

Key Sources

Further Reading

See Also

Abhishek's take

I care less about the 8-to-7 oddity than about the hidden ranking that causes it. On a buying floor, demand arrives as decimals and inventory arrives as units; choosing the rounding rule means choosing which errors become policy. If a lottery matched the decimals across 100 cycles, would I trust it for the one cycle that sets the assortment?

Tags: #apportionment #voting-theory #discrete-mathematics #fairness #paradoxes