paper-to-game

Design memo — The Wall in the Head

1. Paper

Alberto Alesina and Nicola Fuchs-Schündeln (2005), “Good Bye Lenin (or Not?): The Effect of Communism on People’s Preferences.” NBER Working Paper 11700, October 2005. The paper was later published in the American Economic Review 97(4):1507–1528 (2007), doi:10.1257/aer.97.4.1507, but this game was built from the working paper alone. Only the PDF was available: there was no code and no microdata (the paper uses GSOEP data). Every number in the game is either a reported estimate, a value computed from the reported coefficients, or a value read off Figure 1, and each is marked as one of these.

2. What the paper hands over

3. The discovery, and its kind

There are two kinds. The game is built around the second.

4. The core narrative

For forty-five years a country was cut in two by a wall, for reasons that had nothing to do with the people on either side. One half lived under a government that ran everything. Then the wall came down. You’d think people who had lived under that government would want as little of it as possible. They wanted more. A doubter says: they’re just poorer; or they were always like that; or the ones who liked markets had fled. Your job is to prove the doubter wrong, one piece of evidence at a time, and to find out how long the wall in people’s heads will take to come down.

5. Fidelity level and claim boundaries

Statistical, from reported estimates only. There is no simulated respondent anywhere in the game. Every number shown is one of three things:

Correlation and causation: the player never moves a lever that changes preferences. The player chooses which comparison to look at, which is what an experimenter does.

6. The aim and the verdict

7. Concepts

The playable experience was not obvious: an empirical paper has no model to steer. There were three concepts.

A. The doubter (discovery, Experimenter; “present the evidence”).

B. The survey van (discovery, Experimenter, sampling under a budget).

C. Spot the Easterner (Subject/perception).

8. The first five minutes (pick A)

9. The pick, and why

A. It makes the paper’s real intellectual content, why we believe it was communism, into the player’s decision. Every reveal is a reported finding, nothing has to be invented about individuals, and the “present the evidence” form is a known pleasure. The counterintuitive bet opens it, and the dose–response moment comes in about two minutes.

Runner-up: B. Switch if the author would rather the player gathered data than reasoned about it, accepting a constructed respondent-level process declared as illustrative.

10. The shape

The game takes about five minutes (an estimate; not timed by a person), in three parts:

  1. The bet: one choice.
  2. The doubter: six objections, ordered for the player:
    1. poverty;
    2. the 1990 shock, answered by the dose;
    3. always different;
    4. self-selection;
    5. cheap talk;
    6. why.

    Poverty comes first because it is everyone’s first thought. The dose comes second so the strongest finding arrives early.

  3. The forecast: drag a flag to a year, then reveal.

No presets: the objections are the paper’s own robustness sections. Two tries per objection: a second miss loses the objection to the doubter, and the card the paper plays is shown so the lesson survives the loss. The score is per objection (first try, second try, or lost), and the session verdict counts objections settled and first tries out of 6. There is no running point total.

11. Controls, mechanism, feedback, moment

12. What is cut

13. Divergences

14. Visual world

The paper’s imagery is two survey samples on either side of a line. Rules:

  1. Concrete grey and paper white; East in a warm red, West in a slate blue, used for nothing else.
  2. A wall runs across the top of every screen, cracked.
  3. Cards are large index cards, at most seven, two per row, with big titles and one line each.
  4. Charts are paired bars, three to five per card, each labelled with its number.

Legibility rule. On the play screen the player looks at:

15. Fidelity plan

fidelity.js checks, against the paper’s numbers:

16. Fidelity result and build log

Fidelity result (fidelity-result.txt, all checks pass):

Build log:

  1. mechanism.js written and checked in Node; one bug found by the check (the share of narrowing attributed to minds changing was inverted) and fixed before any page existed.
  2. loop.js played a plausible session headlessly: wrong cards produce a specific rebuttal and still show the card’s finding; the yardstick reads “settled at the first try” or “on try n”.
  3. template.html skinned (concrete wall, East red / West slate, large cards two per row, bar charts with the number on every bar); build.py inlines the mechanism verbatim and verifies it byte for byte.
  4. Screenshots at 380 px: start, bet, card grid, a settled card, the cohort card, a wrong card, forecast, end. One fix from the screenshots: the end-screen table squeezed the objection column.
  5. Builder’s self-review: the aim can be stated from the start screen alone (“prove to a doubter that communism changed what people want”); the discovery can be stated from the top of the end screen alone (East Germans want about 16 points more state; the gap grows with years under communism; it is not money, history or migration). This is the builder’s review, not stranger testing.
  6. session-test.js drove the final page through a full session at 380 px with two wrong cards, the help page opened and returned from mid-play, a forecast, the verdict, and a retry: no console errors, no horizontal scroll (scrollWidth 380). Session length is an estimate of about five minutes; a person has not timed it.
  7. Revision after the author’s first play: the start-screen wall made much bigger (a 260 px cracked wall with WEST and EAST on it), and a two-try rule added in mechanism.js (CONFIG.maxTries): a second miss loses the objection to the doubter, marked red, and the paper’s card is shown beneath the wrong one. Fidelity check (j4b) covers the rule; the session test now loses objection 4 on purpose. All 25 checks pass; the browser session runs clean at 380 px.