Thought Toys · Strategy & computation · Exhibit 97

Same votes. Different winner.

Twenty-five voters. Ten of them back the city party. Nobody changes their mind, ever. Move only the lines between districts and that party wins three seats of five — or none.

The electorate — 10 city (filled), 15 country (hollow)

your turn — press any square to move it to the next district, and watch the winner change while the votes never do

What you're seeing

Twenty-five voters live on this grid. Ten of them — the filled circles — back the city party, and they live together in the two left-hand columns, the way city voters usually do. The other fifteen back the country party. That is the entire electorate, and it never changes. No voter here ever switches sides, stays home, or moves house.

The grid has to be cut into five districts. The rules are the ordinary ones: each district holds exactly five voters, and each district has to be one connected piece. Whoever has three or more voters inside a district wins that district's seat.

Start with five columns, which is what the geography suggests. The city block fills two whole columns, so the city party wins two seats out of five. Ten votes in twenty-five, two seats in five. That looks fair.

Now press five rows. Nothing about the voters has moved. But every row now cuts across the city block and picks up two city voters and three country ones, so the country party wins all five districts by the same narrow margin. Forty per cent of the electorate has just been reduced to nothing. That is cracking — spread a group thin enough and it is a minority everywhere.

Then press packed & cracked. Here the map hands one district entirely to the country party, which wastes five of their votes on a seat they were going to win anyway, and then takes three of the remaining four by the smallest margin that works. The city party now holds a majority of the chamber with forty per cent of the vote. That is packing — concentrate your opponents where their extra votes are thrown away. Every district is still exactly five voters, still one connected piece. Nothing has been broken.

Press squares yourself and you will feel the real lesson quickly: the outcome is easy to move and the votes are impossible to move. Whoever draws the lines is choosing the answer before anyone votes.

The rule, exactly. A map is legal when the grid is cut into five contiguous: one connected piece, reachable square to square without lifting your finger districts of exactly five voters, and a district goes to whoever holds three of its five. Fixing the electorate at ten city voters to fifteen country ones, the question is what the legal maps can do: seats(city) {0, 1, 2, 3}  over all 4,006 legal maps This is settled by exhaustive enumeration, not by example (improve/verify/97-gerrymander.js): every one of the 4,006 ways to cut a 5×5 grid into five connected pentominoes is generated, checked against the page's own legality test, and scored. The count 4,006 is a known combinatorial quantity, so reproducing it independently checks the enumerator itself rather than merely reporting a number. Exactly one legal map leaves the city party with nothing — the five rows — and 276 of them, close to one map in fourteen, hand it a majority of the seats on 40% of the vote. Notice also what the rules do constrain: no legal map gives the city party four or five seats. The rigging has a ceiling; it just sits far above what its votes deserve. Four negative controls. A map with a six-voter district must be rejected; a map whose district falls into two separate pieces must be rejected; a legal but ugly S-shaped district must still be accepted, so the test rejects disconnection rather than mere untidiness; and the comfortable intuition that 40% of the votes buys 2 of 5 seats is refuted by counting — only 3,033 of the 4,006 legal maps, 76%, are proportional at all.

Also in Strategy & computation: The evolution of trust →

All 27 in Strategy & computation
  1. 10The evolution of trust
  2. 23Sorting algorithms
  3. 24PageRank & the random surfer
  4. 25Huffman coding
  5. 26Dijkstra's shortest path
  6. 27Nash equilibria
  7. 33The learning-rate cliff
  8. 36A* pathfinding
  9. 37Braess's paradox
  10. 44Diffie–Hellman key exchange
  11. 45Preferential attachment
  12. 46Aliasing & the Nyquist limit
  13. 47The secretary problem
  14. 49Freeze too fast, stay stuck
  15. 51Cross one line, and its territory closes
  16. 56Catch one error, miss the next
  17. 57Why more processors stop helping
  18. 58Why a busy line explodes
  19. 59The set that's only sure when it says no
  20. 60The fit that memorizes instead of learns
  21. 61When the wire breaks, pick one
  22. 67Better at both, and still better off trading
  23. 69Everyone was consistent. The vote wasn't.
  24. 78Every world map is lying. You get to pick the lie.
  25. 82Your computer can't hold one tenth
  26. 85The shape that has only one side
  27. 97Same votes. Different winner. — you are here

← the cabinet · Thought Toys — a cabinet of explorable explanations. Exhibit 97.