Title. First built as Ganguli’s Letters; renamed Dear Household at Steffen’s request after the first merge: the paper has two authors and a title naming one was wrong, and a title about the game was wanted. Dear Household is the salutation of the letter the player writes.
Not Steffen’s: a gift for someone else, with the beta line in the footer and no signature or counter. No recipient named.
Ideas given, and they bind the design: “the choice of perspective is important here. perhaps the planner? or we take this outside the survey and look at its real life equivalent. the visuals should reflect the green topic.” Three instructions, taken as: the player is the planner, not a participant; the game is set in the real-life equivalent of the experiment, not in a survey; and the look is green. All three are honoured below, and they do not conflict with the precedent rule: the planner is a controller, which is the position of Jevons’s Noon two games back, but the paper and the brief both want it, so the other three dimensions differ instead (section 6).
Jayant Vivek Ganguli and Friederike Mengel, “Social influence and carbon dioxide mitigation”, Journal of Public Economics 253 (2026), 105558. Received 5 August 2025, accepted 12 December 2025, online 24 December 2025. DOI 10.1016/j.jpubeco.2025.105558. Open access (CC BY). Preregistered as AEARCTR-0010768. Data “available on request”.
What was available: the typeset PDF only, with Tables 1–4, the histogram in Table 2’s right panel, and Figure 1. Not the online appendix (which holds the sample sizes per treatment, the heterogeneity tables, the belief distribution and its type cutoffs, and the wave-2 regressions for several pledges seen), and not the data. The distribution of pledges in the baseline and the main treatment is read off the histogram and adjusted to hit the reported means; it is tagged as such in the config block.
An estimate, primary. An online experiment on 4,701 UK adults in two waves. Everyone gets £10 and chooses how much of it to spend on tree planting (a carbon offset) and how much to keep; the amount is the paper’s “willingness to mitigate”. Wave-1 participants are told, before choosing, what will happen to their anonymised choice:
| Treatment | What the wave-1 participant is told | Mean pledge |
|---|---|---|
| BASE | nothing | £2.83 |
| OBS-(1,1) | one wave-2 person will see it, after they have chosen | £2.82 (−0.014, n.s.) |
| INF-(1,1) | one wave-2 person will see it before they choose | £3.30 (+0.477, p<0.05) |
| INF-(20,1) | twenty wave-2 people will each see it (and only it) before choosing | +0.131 on INF-(1,1), n.s. |
| INF-(1,20) | one wave-2 person will see it among twenty others’ before choosing | −0.496 on INF-(1,1), p<0.05 |
| INF-(20,20) | twenty wave-2 people will each see it among twenty others’ | −0.411 on INF-(1,1), p<0.1 |
Table 2 (column 1), Table 3 (column 1, baseline INF-(1,1), constant 3.307). Wave 2’s response (Table 4, column 1): a wave-2 person who sees one pledge gives £2.54 plus 7.5 pence per pound seen. The social multiplier for a pound of extra mitigation, if influence propagated through many people, is about 1.1 (footnote 14). Beliefs (section 3.3): asked how much a wave-2 person would give on seeing £2, £5 or £8, wave-1 participants expected the step from £2 to £5 to add £1.62 when it added £0.27, and the step from £5 to £8 to add £1.16 when it added £0.68. About one in ten was accurate (within 10%), 15–20% underestimated, most overestimated (Figure 1). Participants also underestimate others’ baseline pledge: £2.42 guessed, £2.87 actual.
An argument, secondary: why influence does not scale. The extensive margin (more people seeing you) is flat, which the paper reads as scope insensitivity; the intensive margin (being one of many seen) kills the effect, which it reads as diffusion of responsibility or dilution. And a boundary the paper draws itself: the choices are anonymous, so social image is switched off; it does not claim image concerns are absent in the world.
A counterintuitive result, in three parts, each the opposite of what a campaign planner would assume. (1) Being watched is not what moves people; what moves them is the chance that their choice shapes someone else’s. The same observer, seeing the same pledge after instead of before, changes nothing. (2) The motive does not scale the way a planner scales things: giving one person twenty followers adds nothing; making one person one of twenty seen takes it all away. (3) The chain reaction people imagine is a ripple. The person who sees your pledge copies seven pence in the pound; people think it is five or six times that.
A town council wants its streets to plant trees. Every household gets ten pounds and can put any of it towards saplings for its front garden, keeping the rest. The council cannot make anyone plant, but it can write the letter: it can tell a household who will get to see what it chose, and when. The planner thinks the way to get more trees is to show everyone’s choice to everyone, or to put the greenest household on a poster. Neither works. What works is the humblest letter: your pledge goes to one household on the next street, before they decide. Told that, people plant about a sixth more. Told that someone will see it only afterwards, they plant what they would have planted anyway. Told that twenty will see it, they plant no more than for one. Told that theirs is one pledge among twenty, they plant as if nobody were looking. And the household that sees a greener pledge plants only a little more, far less than the planner expected.
| Game | Position | Format | Feedback device | Chassis |
|---|---|---|---|---|
| Carbuccia’s Counter | Controller (the fund) | Strategic | The counter (two glowing numbers) and the before/after over a ghost cohort | Techno: a French public counter’s ticket display, institutional cream, LED red; Barlow; serrated ticket stubs; light |
| Leavitt’s Cloud | Experimenter | Discovery | The dial and fold, then the chart | Austere / Instrument: glass plate on a light box, velvet and lavender, violet ink; Spectral + Josefin Sans; pills; dark |
| Jevons’s Noon | Controller (the nation) | Challenge | The fill (squares of coal) and the trajectory | Instrument: Jevons’s engraved plate, cream and soot, furnace orange; Bodoni Moda; engraver’s double-rule buttons; light |
What this game does differently. Position: the planner is a controller, which Carbuccia’s Counter and Jevons’s Noon both were; the paper’s policy question and the brief (“perhaps the planner?”) both want it, so the position is kept and the other three dimensions move. Format: a challenge with a discovery inside it: a total to maximise, where the obvious moves fail and the humble one works, plus a side-bet on the ripple that the paper’s own belief question supplies. Device: the sightlines (who sees whom, drawn as lines between two streets before the player commits) and the gardens filling (each pound a sapling, counted) with an expectation marker the player set beforehand; not a counter, a dial, a fold, a fill of squares, or a trajectory. Chassis: Retro, a 1970s British public-information poster, flat greens on cream with a chunky rounded face; after a Techno counter, an austere glass plate and an engraved chart; after Barlow, Spectral and Bodoni, a rounded poster face; after serrated stubs, pills and double rules, envelopes and solid poster blocks. Light, like Carbuccia’s; the register and its family differ from all three.
The aim, on its own line: Twelve streets, six letters: get the town to plant as many saplings as you can.
The yardstick, in the aim’s units: each pair of streets shows its saplings against the plain pair (what the same two streets plant when the letter says nothing about anyone seeing anything): this street 66, next street 58; plain streets plant 57 each. A player who has just done what was asked sees that this pair beat the plain pair, now.
The verdict leads with: You sent six letters: [the list]. Your twelve streets planted 702 saplings. Plain letters would have planted 679; pass it on everywhere, 741. You expected the next streets to plant about 20 more than a plain street when they saw what you sent; they planted about one more. In the paper’s words, you substantially overestimated your influence; so did about one in five of its participants, and most of the rest overestimated it mildly. Then the paper’s lines: Told that one person will see their pledge before deciding, people gave about a sixth more. Told that the person would see it afterwards, they gave no more. Told that twenty would see it, no more than for one. Told that theirs was one of twenty seen, no more than if nobody saw it. And the person who saw a pledge copied seven pence of every pound. Then the boundary: the pledges in the paper, and in the game, carry no name.
A. The Planner’s Letters — Controller, challenge with a discovery inside. Six pairs of streets, twenty households each. For every pair the planner picks one of six letters, each a real-life arrangement with a plain name: the plain letter (nothing), afterwards (your pledge shown to one household on the next street, after they decide), pass it on (to one, before), the sheet (all twenty pledges to every household on the next street, before: the neighbourhood report), the clipboard (all twenty to one household, before), the spokesperson (one household’s pledge to all twenty, before). Touching a letter draws the sightlines between the two streets. PLANT: this street’s gardens fill. Then the side-bet: how many saplings will the next street plant, having seen what the letter sends? The marker starts at the plain street’s 57. Then the next street fills. Deliberation: which arrangement reaches the most people, and how much will the ripple be. Discovery: the sheet and the spokesperson, the two that sound strong, do nothing or next to nothing; pass it on works; the ripple is one sapling. Fun: anticipation (letter chosen, marker set, gardens fill), agency (the letter decides the street), discovery (the humble letter), mastery (the second half of the game is played knowing it, and the side-bet tightens). Second playthrough: try the letters not tried; aim for the best town; or bet the ripple right from the first pair.
B. The Household — Subject, dilemma. The player is one household on a street, choosing pounds to saplings six times under six letters, and after each choice sees what the paper’s thousand did under that letter, and at the end what their own pattern says. Deliberation: how much would I give if one person will see it. Discovery: the sample’s pattern against one’s own. Fun: discovery only. Second playthrough: none. Fails the author test on contact: the authors know the answer, and a game about one’s own introspection is a quiz about the result. Not picked.
C. The Chain — the reckless one — Controller, challenge, with a play screen that is a map and no menu. Forty houses on a map; the player draws who sees whom by tapping a sender then a receiver; each house may receive one card (so every viewer sees exactly one pledge, which keeps every house in a measured cell), and a second tool, the sheet, gives a house every pledge before it (the “one of many” cell). Cards are limited (eight for twenty houses), so the player chooses where to put them, and the trap is to fan them out from one house. PLANT: the map greens. Deliberation: where do the cards do the most. Discovery: drawn by one’s own hand: the star wastes cards, the chain spends every card on a motive. Fun: agency and discovery; mastery in the second map. Second playthrough: a different drawing. The risk: forty taps on a phone, a drawing that is hard to read once drawn, and the discovery is the same one as A’s, arrived at more slowly; and the best drawing (a chain) is also the obvious one once seen, so the deliberation thins after one map. Its best parts go into A: the sightlines drawn in the picture when a letter is touched, and the star as one of the six letters.
A, the pick. 0:00 A cream poster: a row of flat little houses with round green trees, the title, three lines, the aim in one sentence, BEGIN. 0:15 “A plain street”: twenty plots in five rows of four, most bare soil, some with five saplings in a row, two with ten; “a street that gets a plain letter plants 57 saplings; the paper’s people put £2.83 of £10 into trees.” They read the ten-pound rule in two lines. NEXT. 0:40 Streets 1 and 2. Two blocks of bare plots, a road between, six envelopes under them. The first envelope is the plain one, read already. They touch THE SHEET (“every household on the next street gets the sheet with all twenty pledges, before they decide. No names.”) and twenty lines fan from the top block to a sheet icon and out to every bottom plot. They think: everyone sees everyone, that’s the one. PLANT THIS STREET. 1:00 The top plots fill one by one, a counter climbs to 58, and a thin mark says plain: 57. Frown. The line underneath: “You told every household that twenty people would each see their pledge among twenty. The street planted 58 saplings; a plain street plants 57. The paper: one pledge among twenty is as good as unseen.” 1:20 “The next street sees the sheet. How many saplings will it plant?” The marker sits at 57; they drag it to 75; SEND. The bottom plots fill: 57. “They saw a street much like their own, and planted like a plain street.” 1:50 Pair 2. They touch THE SPOKESPERSON: one top plot gets a megaphone and twenty lines fan down. “That’s the influencer.” PLANT: 58. “One household was told all twenty would see it; the other nineteen were told nothing. One household planted more.” The bet on the next street: they go to 70; it plants 58. “Twenty households each saw one pledge of £3.30 and planted seven pence in the pound more: one sapling.” 2:40 Pair 3. They touch PASS IT ON: twenty short lines, plot to plot. “That’s the small one.” They touch AFTERWARDS: the same lines, dashed. They weigh it; pass it on. PLANT: 66. Grin. “Every household was told one person on the next street would see their pledge before deciding. The street planted 66: about a sixth more.” The bet: 60; it plants 58. “One more.” 3:30 Pair 4: afterwards, to check. 56. “The same observer, after: nothing.” 4:10 Pairs 5 and 6: pass it on, bet 58, right both times. 5:00 The verdict: 702 against 679 and 741; “you overestimated the ripple at first and then had it”; the paper’s five lines. They can say what the paper found: being seen before someone decides makes people give more, being seen by more people doesn’t, being one of many seen undoes it, and the person who sees copies only a little.
C, the runner-up. 0:00 The same poster. 0:30 A map of forty houses, bare. “Tap a house, then a house that decides after it: the second sees the first’s pledge before choosing. Eight cards.” They tap house 1 then 2, 1 then 3, 1 then 4: a star from the greenest-looking house. PLANT: the map greens hardly at all. “House 1 was told seven people would see its pledge; it planted more. The seven saw £3 and planted like anyone.” 1:30 Second map: they chain 1→2→3→…→9. PLANT: nine houses green up. Grin. 2:30 Third map: the sheet tool; they give house 20 the sheet; the nineteen before it plant as if unseen. 3:30 Fourth map: ten chains of two against one chain of nine… the same total. They are done, and the deliberation has thinned to arithmetic. The discovery landed at 1:30 and the rest was confirmation. Good, and slower than A, with forty taps a map.
A, The Planner’s Letters. It is the brief’s seat (the planner), set in the brief’s world (a street, letters, saplings) and the discovery lands on the first or second letter the stranger reaches for, because the letters that sound strong are the paper’s dead cells and the humble one is its live cell. The side-bet is the paper’s own belief question, asked about the next street instead of about a wave-2 stranger, and it gives the paper’s third finding a place in the player’s own hands rather than on the verdict screen. The sentence afterwards has no “guess”: I sent the sheet and nothing happened; I put one household on a poster and one household planted more; I passed each pledge to one neighbour and the street planted a sixth more; and the neighbours who saw it planted one sapling more between them. Originality: no game in the table has had the player design who sees whom, nor drawn the design as sightlines before the commitment.
Author test. The authors would recognise every cell of their own design in the six letters, their own belief elicitation in the marker, the social multiplier in the ripple, and the “no names” boundary they draw themselves. What they will enjoy is watching a player reach for the sheet first, which is the home-energy-report arrangement the literature leans on.
Runner-up: C, The Chain. Switch if the author wants the star-versus-chain lesson drawn by hand for a talk, at the cost of a slower first discovery and a harder phone screen.
What the reckless concept would have given. The structure of influence in the player’s own hand, as a drawing. Kept: the sightlines, drawn in the picture the moment a letter is touched, so the player sees the shape they are choosing; the spokesperson letter, which is the star. Given up: drawing it.
Register: a 1970s British public-information poster (Retro family). From the setting of the data (a UK sample, a tree-planting scheme) and from the one thing the brief insists on: green. The poster register is flat colour, a chunky rounded face, round trees on sticks, houses as two shapes, a cream paper ground; the register that told Britain to plant a tree in 1973. Comic means the register, not jokes.
#F3E9CF (the page); forest #1F5A38 (all type, outlines, the dark of every tree); leaf #7CBB4C (every sapling, and nothing else: the green that grows); soil #9A6A3A (bare plots, the ground); marigold #F2A93B (the chosen envelope, the one filled button, the bet marker).The OBS treatments other than OBS-(1,1) (the paper reports them only in Figure 1); the heterogeneity by attitudes and demographics (Tables D.6–D.8, D.12); the underestimation of others’ baseline pledge (£2.42 against £2.87; read on the About screen); the ceiling-effect checks; the wave-2 regressions on mean, median and mode of twenty pledges (the game uses the mean, which the paper says is not distinguishable from the others); the welfare discussion; the 2,037 tonnes (on the About screen).
Kind: reconstructed. mechanism.js runs a data-generating process built from the paper’s estimates. Each household on the lettered street pledges, in expectation, the plain mean (Table 2’s constant) plus the effect of the arrangement it was told, from Tables 2 and 3; each household on the next street that sees a pledge before deciding adds Table 4’s 7.5 pence per pound by which what it saw differs from a plain pledge. Streets are expected streets, with no draws. The plots are whole saplings taken as quantiles of the histogram beside Table 2 (read off the figure, adjusted to the reported means), nudged so that each street adds up to its rounded total; the histogram decides only how a street looks.
Where each discovery comes from. That pass it on beats every other letter follows from the paper’s cells: twenty households each with the INF-(1,1) motive, against one (the spokesperson), none (plain, afterwards, the clipboard) or the slight INF-(20,20) rise (the sheet). It is not a product of the pairs, the bet or the town. The ripple’s smallness is Table 4’s slope. The calibration check re-runs the pair totals with the extensive margin at its point estimate, with OBS and INF-(1,20) at theirs, with the copy centred on Table 4’s own constant, with twice the slope, with no copy at all, with the sheet copied at its mode (5) instead of its mean, and with INF-(1,1) at the bottom of its 95% interval: pass it on wins in every case, by at least seven saplings a pair except in the last, where it still edges the sheet. The bet’s typology is ours and is presented as the game’s reading in the paper’s words, with the paper’s own shares quoted beside it.
calibration.js checks, against the paper’s typeset numbers: the two histograms add to 100 and reproduce the reported means within 0.01, with 0 and 5 as the two most frequent choices and 10 above its neighbours; Table 3’s constant equals Table 2’s constant plus the INF-(1,1) coefficient (3.307 = 2.830 + 0.477); the 17% of the abstract; the three zeros played are each within one standard error of zero or within 0.03 of the baseline, and INF-(20,20) is at its point; Table 4’s slope gives the text’s 75 pence for £10 against £0, lands within 0.05 of the text’s raw 0.27 for £2 → £5, and gives the multiplier of footnote 14; the belief ratio of about six; the plain street (56.6 → 57), the pass-it-on street (66), the ripple (0.72) and the pair multiplier (1.075); the spokesperson street (one at 3.307, nineteen plain) and its ripple (equal to pass it on’s); the sheet and clipboard streets; every drawn street is twenty whole plots adding to its rounded total, with the plain street showing five bare plots, three or more at five and one at ten, and the spokesperson’s plot at 3; the bet’s cutoffs; a six-pair game, the verdict’s numbers (679, 741), its lines free of decimals; the robustness runs of section 16; and the play-it-in-numbers table. Result: 82 checks, all pass (calibration-result.txt).
The town’s saplings after six pairs; a plain town is 679:
| Strategy | Pairs | Town |
|---|---|---|
| the sheet every time | 115 × 6 | 688 |
| the spokesperson every time | 114 × 6 | 686 |
| afterwards, the clipboard, or plain every time | 113 × 6 | 679 |
| one of each | 113 115 114 123 113 113 | 692 |
| the stranger (sheet, spokesperson, then pass it on, one check of afterwards) | 115 114 123 113 123 123 | 713 |
| pass it on every time | 123 × 6 | 741 |
Reading: one simple rule wins everywhere, and it is the sentence the player should leave with: tell each household that one person will see its pledge before deciding. The spread between the strong-sounding letters and the plain town is two to nine saplings over twelve streets, which is the paper’s “does not scale”; the spread to pass it on is sixty, which is the paper’s main effect. The decisions matter, and the one that matters is the humble one. The bet adds a second axis, the ripple, where every strategy meets the same answer: less than one sapling.
mechanism.js written headless with every number tagged; calibration.js run; three checks failed on the first run: a wrong check (1 and 2 are frequent choices too; the check now asks for 0 and 5 on top and 10 above 8 and 9, and the plain histogram was shifted a point so 5 leads 2 at the same mean), the spokesperson’s plot drawn from the top quantile (a household whose expectation stands apart from the street’s is now drawn at its own rounded mean), and the best town rounding to 741, not 740.loop.js played six pairs on placeholder plots: something visible changed after every action, the dead letters visible as streets on their mark, the ripple invisible in the rounded tally and said in words (“less than one sapling more”), which is the honest size; one wording fixed (“pass it on on every street”).template.html built on the declared chassis: Fredoka embedded; a single canvas for the two streets and the road; plots of soil with whole saplings; eyes, cards, arrows and slots for the preview; envelopes with a drawn flap; poster blocks; the bet as a slider with a marigold marker on the bar. about.html; build.py verifies the embedded mechanism byte for byte.session-test.js drove a full session at 380 px in headless Chromium: every letter previewed, six pairs with the stranger’s plan (sheet, spokesperson, pass it on, afterwards, pass it on, clipboard), bets of 75, 70, 60, 58, 57, help opened mid-play and closed, About, the verdict, then a second town (pass it on everywhere, bet 58): 741, “about right”. No errors, no horizontal scroll.dear-household.html); rebuilt, checks and session re-run.