How easy is it to rig an election?

Live: https://yaqzans.github.io/who-should-count-more/

The page has a good candidate, a bad candidate and 1,001 voters. Someone backing the bad one tries to rig the election: by paying people for their votes, by threatening them, or by buying the national news. Every election is counted twice, on the same ballots:

• Equal votes: one person, one vote
• Educated x5: an educated person's vote counts as five

The question is which count is harder to steal.

What is on the page

1. Pick a country.

   • Bangladesh, India and Pakistan are tuned to real survey data.
   • Two made-up worlds show the idea in its pure form: everyone shares 3 channels, or everyone reads widely.

2. Pick how to rig it. Pay for votes, threaten voters, buy the TV news, or any mix of them.

3. Hold elections. There are three ways to run one:

   • Hold an election: watch one get counted.
   • Find a stolen election: skip ahead to one the rigging flipped.
   • Run 1,000 elections: a live chart of how often each count picks the good candidate.

   A scoreboard under the buttons keeps a running total for elections held with "Hold an election": how many, how many were stolen under each count, and what the rigger spent.

4. Who won, and why.

   • Headline: each election gets one, written from its own numbers, in red when the result was stolen.
   • Two panels, one per count. Each has its own bar, which ticks up live while the votes are counted and marks where the result stood before any rigging. Each also has its own write-up:
     • in short: what decided it
     • is this normal: the good candidate's win rate over 300 replayed elections with the same settings
     • who has the say
     • the margin
     • what the rigger did, or what flipping it would have cost
     • a verdict
   • What everyone heard: shared by both counts. It covers each group's news, the worst and most helpful local source, the parties, and the swing voters.
   • Comparing the two: the exact arithmetic of why the counts differ, and the lesson.

5. Over 1,000 elections: how many each count won, and how many the rigging stole.

The picture. Every dot is a voter, linked to the sources they follow:

• Layout: educated voters sit above the dashed line, everyone else below. Squares are voters loyal to a party.
• Colour: dots turn white for a good vote and red for a bad one.
• The news: before anyone votes, small dots travel from each source to its listeners, coloured by which way that source leaned.
• The rigger: a red RIGGER node is wired to the channels it bought and sends red dots to the voters it paid or threatened. Those voters get an x.
• Interaction: hover anything for its numbers, and click a source to highlight its listeners.

Every one of the 21 variables is listed under "Every variable in the model", each with a one-line meaning and its source.

How one election is simulated

1. The electorate is drawn once per country and setting.

   • Each voter is educated or not.
   • Each voter is loyal to one of the two parties, or is a swing voter.
   • Each voter has an asking price.
   • Each voter is wired to news sources:
     • educated voters follow 3 national outlets
     • everyone else follows one of 30 local sources, plus, with the country's TV share, one national channel

2. Every election, every source slants a random amount. This is drawn from a normal distribution, even if nobody paid it. Everyone who follows a source gets the same push. That shared push is what makes whole groups wrong together.

3. Each voter weighs the candidates:

   perceived gap = the good candidate's real edge
                 + the average slant of the sources they follow
                 + their party's pull (if loyal)
                 + the educated standing lean (if educated)
                 + their own random misjudgement
   

   They vote good if the gap is positive, bad otherwise. An educated voter's own misjudgement is smaller (about +-0.8 against +-2).

4. The votes are counted twice: once with weight 1 for everyone, once with weight 5 for the educated. The good candidate needs more than half the weight, and an exact tie goes to the bad candidate.

5. The rigger acts on each count separately.

   • Bought news pushes every national channel toward the bad candidate.
   • Cash and threats turn voters who voted good. It picks the most vote weight per taka first and uses whichever is cheaper for each voter, cash or a threat. It stops once the bad candidate reaches half, and only pays if the whole job fits the budget.

6. The same election is replayed with no rigging, using the same random numbers (common random numbers). The difference between the two runs is exactly what the rigging changed. That is how the write-up can say "the good candidate was ahead until the rigger paid 22 voters".

The random numbers are seeded (mulberry32, with Box-Muller for normal draws), so any result can be reproduced.

Where the numbers come from

Some settings are measured, some are reported figures, and some are assumptions. The page shows each one's source next to its slider.

This is the beginning of the README. Read the rest at the original. Read the full README ↗