NOTE-1267: Population ethical intuitions

Contribution

It is the first systematic set of experiments on lay population ethics. It asks three questions philosophers have argued over and finds that lay judgement fits no single theory. Do people weigh happiness and suffering alike? Do they count the average or the total? Do they count only people who exist, or also people who could?

Key insight

Ordinary judgement counts merely possible people, but not in large numbers. People think a world is made better by adding a happy person who would not otherwise exist, and worse by adding an unhappy one, by equal amounts. That is against the view that morality is about “making people happy” and not “making happy people” (Narveson 1973: 80, as quoted). But the value of more happy people fades as their number grows, while the disvalue of more unhappy people grows. When average and total conflict, the average tends to win.

Assumptions

• Hedonic welfare only. Lives are described by a happiness scale from −100 to +100 with a neutral 0, “a life worth living” above it (Studies 1b–3d). Non-hedonic goods, such as Van Gogh’s achievements, are set aside (Limitations).
• Axiological, not deontic, judgements. Participants say which world is better, or which they would prefer to come into existence, not what anyone ought to do. The authors grant that the question forms differ between studies (Limitations).
• Fixed-level homogeneous populations. Everyone in a civilization has the same level, and resource, crowding and environmental concerns are stipulated away (Studies 3a–d).
• Attention and comprehension checks drop many participants: 161 of 622 in Study 3c, 131 of 613 in Study 3d. The authors report that the pattern holds without exclusions (Supplementary Materials, unread).

Key results

• Studies 1a–c: trading happiness against suffering. “What percentage of happy and unhappy people would there have to be for you to think that this world is overall positive … better for the world to exist rather than not exist?”

  • 1a (N = 474, pre-registered). Extreme happiness (“falling in love”) against extreme unhappiness (“being tortured”). Required happiness: 75.62% for a population of 1,000, 74.98% for the share of time within each life; no difference, t(468) = 0.41. The less willing participants were to relive their worst day to relive their best, the more happiness they required (r = −.11). Women required more than men (78.4% vs 72.35%).
  • 1b (N = 356, pre-registered; 2 × 2 of mild or extreme happiness and suffering). Above 50% in every cell. Extreme suffering outweighed by mild happiness needed 81.97%, by extreme happiness 72.03%. Mild suffering needed 64.56% and 61.9%.
  • 1c (N = 278, pre-registered). The happiness and suffering were mild and stated to be “exactly equally intense”, and each person in the world was said to weigh an hour of each as exactly neutral. Required: 61.84% (population) and 60.37% (time within a life); the modal answer was 51. 37.0% said 50% or 51% was enough.

• Studies 2a–b: adding one person. 1–7 scale, 4 = equally good.

  • 2a (N = 157, within-subjects, not pre-registered). A world of one million people at level 0. Adding one at +100: M = 5.06. Adding one at −100: M = 3.08. Reverse-scored, the two did not differ (t(156) = 1.23, p = .22), also with a weak preference for adding a neutral person (M = 4.23) controlled.
  • 2b (N = 369, between-subjects, Prolific, not pre-registered). An empty world or one of 10 billion neutral people. Empty: happy M = 5.64, unhappy M = 2.38. Full: happy M = 4.74, unhappy M = 3.14. No effect of valence (F(1, 365) = 0.24); adding mattered more to an empty world (F(1, 365) = 54.16). Only seven participants chose “This question doesn’t make sense to me / I don’t know”.

• Study 3a: totals vary, averages fixed (N = 789, pre-registered). Pairs of civilizations lasting millions of years, all happy or all unhappy, at ten-fold ratios (1,000 vs 10,000; 1M vs 10M; 1B vs 10B). “Which would you prefer?“

  condition	1,000 vs 10,000	1M vs 10M	1B vs 10B
  happy	5.12	4.33	3.71
  unhappy	3.04	2.56	2.32
  neutral	4.65	3.98	3.48

  (7 = strongly prefer the larger.) The preference for fewer unhappy people (M = 5.36 reversed) was stronger than for more happy people (M = 4.39), t(785) = 8.67. At a billion, participants preferred one billion happy people to ten billion (t(401) = −2.90). Median ideal population: 1.5 million happy; 100 unhappy.

• Study 3b: averages vary, totals fixed (N = 825, pre-registered). 4,000 at ±60 against 6,000 at ±40, and the same at millions and billions. People strongly preferred the better average: happy M = 2.28, unhappy M = 2.34 (1 = the population with the better average). Only 47% correctly said the totals were equal, but the pattern held among those who did.

• Study 3c: average against total (N = 461, pre-registered; 2 valence × 2 thinking style). Civilization A, 1,000 at ±100, against 100,000 at ±90, ±70 or ±50, or a million or a billion at ±50.

  • Responses favoured the total for 100,000@90, were indifferent at 100,000@70, and favoured the average from 100,000@50 on. In the happy condition they still favoured 1,000 at +100 over a billion at +50 (t(212) = −5.56).
  • Instructed intuition was more averagist than instructed reflection, F(1, 457) = 7.42, ηp² = .02.

• Study 3d: mere addition (N = 482, pre-registered). 1,000 at +100 (or −100) against the same plus 1,000 at ±10, ±50 or ±90.

  • Averagist answers predominated. They were stronger for unhappy populations (F(1, 478) = 21.89) and under intuition (F(1, 478) = 7.99). In the unhappy condition, adding 1,000 people at −90 to 1,000 at −100 was preferred (intuition M = 3.3, reflection 3.65, reverse-scored; 1 = averagist).
  • Strict averagists (choosing 1) were 40%, 30% and 17% across the ±10, ±50 and ±90 tasks; strict totalists 7%, 10% and 16%.
  • Asked in the abstract, participants said adding happy people below the average is good (M = 4.85) and adding unhappy people above it is bad (M = 3.31): totalist. 27% (happy) and 35% (unhappy) answered totalist in the abstract and averagist in the concrete.

Claims

id	claim	strength	support
C1	Laypeople require more happiness than suffering for a population to be net positive, even at stated equal intensity	moderate: three pre-registered studies; the size of the asymmetry shrinks as intensity is equalized (75% to 61%)	Studies 1a–c
C2	Laypeople reject the intuition of neutrality: adding a happy person makes the world better	moderate: two studies, within and between subjects, neither pre-registered; one person added, at +100	Studies 2a–b
C3	Laypeople’s judgements of adding a happy or unhappy person are symmetrical, against the procreation asymmetry	moderate: same two studies; axiological only	Studies 2a–b
C4	Laypeople are sensitive to totals and to averages, and in conflict tend to favour averages	moderate to strong: four pre-registered studies with varied designs	Studies 3a–d
C5	The value of more happy people declines with population size, and reverses at a billion; the disvalue of more unhappy people does not	moderate: one study, comparative preferences	Study 3a
C6	Reflection moves judgements toward totalism	weak to moderate: an instruction manipulation, ηp² = .02 in both studies; in Study 3d the post hoc difference is significant only for happy populations	Studies 3c–d
C7	Averagist intuitions may come from substituting one life’s distribution of experiences for a population’s	weak: suggested by equal trade ratios across populations and lives; not tested	Studies 1a, 1c; General Discussion

Concepts

• intuition of neutrality (Broome): adding a life worth living is morally neutral, all else equal.
• procreation asymmetry (McMahan): adding an unhappy life is bad, adding a happy one neutral.
• averagism / totalism: a population is better as its average, or its total, welfare is greater.
• asymmetric scope sensitivity: preferences over the size of unhappy populations stay steep as numbers grow while those over happy populations flatten.
• axiological versus deontic judgement: which outcome is better, versus what one may or must do.
• attribute substitution (Kahneman and Frederick): answering how good a population is by answering how good its distribution would be within one life, “such as themselves”.

Connections

• Narveson (LIT-1769), unread. The paper quotes his slogan, “We are in favor of making people happy, but neutral about making happy people” (1973: 80), as the source of the intuition of neutrality. Lay judgement rejects it in Studies 2a–b. Two qualifications, both the authors’. First, the test is axiological: it asks which world is better. Narveson’s view, as the record states it, is about what utilitarian concern gives reason to do. Second, people “may consider it morally forbidden to create a new unhappy person but only morally supererogatory … to create a new happy person” (General Discussion), an asymmetry the studies did not test. So the result counts against neutrality about value, not yet against a person-affecting view of obligation.
• Schubert et al. (LIT-1748). Cited here as a reason the question matters: if creating happy people is good, reducing extinction risk could be a priority. That paper’s Study 1 found people at the midpoint on whether a happy person coming into existence makes the world better (M = 4.45); this paper finds yes when the person is at +100 and described concretely.
• Parfit (LIT-469). Study 3c’s dilemmas resemble the Repugnant Conclusion. The authors infer that people would prefer a small population at +100 to any population at +1. They did not test it.
• Longtermism (LIT-1678 and the works the question cites). The authors end on it: reflective totalism “implies that future generations should be weighed much more heavily” (General Discussion). Study 3a’s fading at large numbers is the obstacle they do not stress.

Bearing on the record

• What the living owe the far future (QUESTION-299). The question cites this paper as finding that ordinary judgement “counts adding happy people as good, against the person-affecting view”. That is right about one person at +100. It needs two qualifications. The finding is axiological, so a person-affecting view of obligation survives it. And concern for more happy people fades with numbers and reverses at a billion, so lay axiology does not deliver the astronomical value longtermism needs.
• The owner’s thesis (CLAIM-491). These are inferences; the paper measures no self and no identification.
  • Impersonal, but thin. Ordinary judgement values possible people the way an impersonal view does. A world is better for containing a happy stranger who would not otherwise exist. That is the ground of obligation the thesis has least to say about. But its reach into large numbers is weak, and it is reached for less when people respond quickly.
  • A single self as the model of a population. The trade ratio of happiness to suffering was the same for a population as for one life (Studies 1a, 1c). The less willing people were to relive their own worst day for their best, the more happiness they required of a population (r = −.11, small). The authors’ explanation of averagism is that people judge a population as if it were the distribution of experiences in one life, “such as themselves”. That is not an extended self in the thesis’s sense. It is a population pictured as one subject, using the self as template. If it holds, it gives the thesis a route into impersonal judgement it has not considered: by projection, not identification.
  • Neither result shows concern tracking a self extended in time. No participant’s descendants, legacy or continuity with the future figure in any vignette.
• The triviality objection (CLAIM-492). Not addressed.
• A dated note should be added to CLAIM-491, jointly with the Schubert et al. reading. Proposed text: “Caviola et al. (NOTE-1267) find that ordinary judgement counts merely possible people impersonally: adding a happy person makes the world better, as much as adding an unhappy one makes it worse. That valuation is thin at large numbers and gives way to averages in conflict. Their account of averagism is that people judge a population as if it were one life, their own as template: projection of a single self, not extension of one. Neither this nor Schubert et al. measures a self, and neither gives evidence that concern for future people tracks one.”
• No THEORY filed.
• No ML instruction.

Limitations

• The preprint, not the published paper. Results may have changed in review.
• Online US samples, mostly MTurk through Positly; Study 2b on Prolific.
• Studies 2a–b were not pre-registered, and carry the paper’s main finding against neutrality.
• The scale is stipulated. Whether participants read −100 and +100 as equally intense is uncertain, as the authors say; Study 1c’s 37% answering 50–51% suggests many did.
• Discrepancies.
  • The abstract’s N = 5,776 does not match the studies: final samples sum to 4,191, and recruitment to 4,776.
  • Study 2b says Study 2a used a world of “10 million neutral people”; Study 2a says one million.
  • Study 3a’s median ideal happy population is 1.5 million in Results and 1 million in Discussion.
  • Study 3b’s unhappy-condition effect is printed d = 2.34, the same as its mean. From M = 2.34 and SD = 1.10 it is about 1.51.
  • Study 2a’s covariate test prints ηp² < .001 with F(1, 156) = 1.52, which gives about .01.
  • Study 3d’s gender test prints t(571) with N = 482.
  • Starmans and Bloom is cited as dissertation chapter 3 in the introduction and chapter 5 in Study 1a.
  • Cross-references: Study 3c cites “Figure 7” for Figure 8 and “Studies 3a and 3c” for 3a and 3b; Study 3d cites “Figure 8” for Figure 9. Study 3d’s Discussion gives the existing average as “−90” where it was −100.
• Typos: “falling love” and “exampless” (Study 1a); “Net,” for “Next” (Study 2a); “more bad than than happiness” (General Discussion heading); “observerved”; “Rozyman”; “MechanicalTurk”.

Open questions

• Is the rejection of neutrality also a rejection of person-affecting obligation? Ask whether one must, or merely may, create the happy person, as the authors propose.
• Does the fading of concern for more happy people come from the numbers (scope insensitivity) or from residual worry about crowding? A design that varied numbers while holding vivid, uncrowded descriptions fixed would separate them.
• Does judging a population “as one life” vary with a measure of how far the judge identifies with humanity or with future people? That would bring this paper to CLAIM-492’s test.