# Evidence and parameter provenance

Society Observatory is an exploratory agent-based model. Its literature establishes reasons to test several mechanisms; it does not calibrate a predictive model of human civilization. This document records a focused reading of original research and official sources on 10 October 2026. It is not a systematic review or a meta-analysis.

## What is calibrated

The natural-reproduction male birth probability is approximately **0.512**. UNFPA gives a usual biological range of about 102–106 boys per 100 girls, equivalent to probabilities of approximately 0.505–0.515. The default is within this range. A birth probability must not be substituted for the male fraction among residents, adults or available partners. [UNFPA](https://www.unfpa.org/son-preference)

No psychological distribution, violence hazard, institutional response, technological growth coefficient or relationship preference has been estimated against an observed human dataset. Those values are deliberately exposed as assumptions. Calling the application evidence-informed does not make them empirically calibrated.

## Evidence categories

| Category | Meaning in this project | Examples |
|---|---|---|
| Empirically calibrated | A numerical default has a relevant observational reference, with scope stated. | Approximate natural male birth probability. |
| Theoretically motivated | A mechanism or formula has a published rationale; its use here still requires assumptions. | Scarcity-sensitive partner competition; the idealized sex-ratio effective-size formula. |
| Speculative | A selected functional form, coefficient, distribution or scenario has not been fitted in the represented setting. | Aggression mean and spread; male aggression shift; jealousy response; fertility preferences; resource depletion; institution update; score weights. |

The bibliography additionally labels **empirical context**, **comparative biology** and **methodology**. Those are types of sources, not extra claims that their findings have been used as fitted coefficients.

## Why the model contains competing mechanisms

The human findings do not yield a universal slope between male proportion and violence. The positive China association concerns combined violent/property arrests among young adults; the US county study found an association in the other direction. A later US analysis reported a nonlinear relationship for sexual violence. Their outcomes, populations and observation units differ. None of these results estimates what happens at 0.1% male. [Edlund et al.](https://www.columbia.edu/~le93/EdlundLiYiZhang13.pdf) · [Schacht et al.](https://link.springer.com/article/10.1007/s12110-016-9271-x) · [Pabst et al.](https://www.mdpi.com/2076-0760/11/5/208)

Consequently, the engine never computes violence directly from the global male fraction. Contextual, competition and buffering variants change individual encounter responses. They are alternative constructed explanations, not fitted reproductions of the papers. A shared set of agent traits and the same seed can be used across the variants.

Relationships also mediate the question. A US family-formation study linked relative female scarcity with greater male commitment. This motivates examining commitment, relationship availability and reproductive inequality alongside incident counts. It does not justify assuming that every rare male becomes highly partnered, or that preferences change identically for everyone. [Schacht and Kramer](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0160320)

## Why the default traits are the same across sexes

The default zero male aggression shift is a controlled comparison assumption. It is not an empirical claim that real-world male and female violence rates are equal. UNODC’s 2023 report describes a pronounced male predominance among homicide victims and recorded suspects in 2021. Those outcomes cannot by themselves distinguish underlying disposition, exposure, institutions, age composition, reporting and opportunity. Converting them into an innate aggression difference would require a fitted measurement model. [UNODC](https://www.unodc.org/documents/data-and-analysis/gsh/2023/Global_study_on_homicide_2023_web.pdf)

A user can introduce a trait shift to investigate sensitivity. A result that appears only after a large shift should be attributed to that assumption. The software provides no biologically privileged value for it.

## Source register

| ID | Source | What it contributes | What is not imported |
|---|---|---|---|
| `unfpa-birth` | [UNFPA, *Son preference* (2023)](https://www.unfpa.org/son-preference) | Approximate normal birth-sex range. | A complete demographic model or an adult ratio target. |
| `edlund-2013` | [Edlund, Li, Yi and Zhang, *Sex Ratios and Crime: Evidence from China* (2013)](https://www.columbia.edu/~le93/EdlundLiYiZhang13.pdf) | Positive sex-ratio association in a specific administrative panel. | A universal violence elasticity or a sex-specific trait mean. |
| `schacht-2016-violence` | [Schacht, Tharp and Smith, *Marriage Markets and Male Mating Effort* (2016)](https://link.springer.com/article/10.1007/s12110-016-9271-x) | Contrasting association in US counties. | A conclusion that any male surplus is protective. |
| `schacht-2016-family` | [Schacht and Kramer, *Patterns of Family Formation in Response to Sex Ratio Variation* (2016)](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0160320) | Commitment as an alternative response to partner availability. | Universal preferences or relationship satisfaction coefficients. |
| `pabst-2022` | [Pabst, Walfield and Schacht, *Patterning of Sexual Violence against Women across US Cities and Counties* (2022)](https://www.mdpi.com/2076-0760/11/5/208) | Outcome-specific nonlinear associations. | A fitted sexual-violence submodel; one is not implemented. |
| `pollet-2017` | [Pollet, Stoevenbelt and Kuppens, *The potential pitfalls of studying adult sex ratios at aggregate levels in humans* (2017)](https://pmc.ncbi.nlm.nih.gov/articles/PMC5540859/) | The aggregation level can change an association and its interpretation. | An ecological correlation used as an individual causal rule. |
| `unodc-2023` | [UNODC, *Global Study on Homicide 2023*](https://www.unodc.org/documents/data-and-analysis/gsh/2023/Global_study_on_homicide_2023_web.pdf) | Sex-disaggregated real-world homicide context. | Equal measured homicide rates or a fitted innate aggression shift. |
| `fehr-2002` | [Fehr and Gächter, *Altruistic punishment in humans* (2002)](https://doi.org/10.1038/415137a) | Cooperation responds to incentives in controlled experiments. | The simulator’s institution or reconciliation coefficients. |
| `liker-2013` | [Liker, Freckleton and Székely, *The evolution of sex roles in birds is related to adult sex ratio* (2013)](https://www.nature.com/articles/ncomms2600) | Comparative evidence for interactions among demography, care and mating systems. | Human mating preferences inferred from bird behaviour. |
| `myhre-2017` | [Myhre, Engen and Sæther, *Effective size of density-dependent two-sex populations: the effect of mating systems* (2017)](https://onlinelibrary.wiley.com/doi/10.1111/jeb.13126) | Effective size depends on the reproductive process, not just census size. | A claim that a simple ratio formula measures actual genomic diversity. |
| `grimm-2020` | [Grimm and colleagues, *The ODD Protocol … A Second Update* (2020)](https://jasss.soc.surrey.ac.uk/23/2/7.html) | A reproducible structure for describing entities, scheduling and submodels. | Validation merely through complete documentation. |
| `windrum-2007` | [Windrum, Fagiolo and Moneta, *Empirical Validation of Agent-Based Models: Alternatives and Prospects* (2007)](https://jasss.soc.surrey.ac.uk/10/2/8.html) | Calibration and empirical validation are independent methodological tasks. | Human predictive validity from matching a few aggregate patterns. |

## Interpreting a result

The strongest justified statement currently has the form: “Under these specified rules, parameter values, demographic mode and time horizon, this set of seeds produced this difference.” A narrower interval after more seeds means the simulator’s conditional mean was estimated more precisely. It does not shrink uncertainty about whether its rules adequately describe people.

The demographic phase diagram is a map of sampled model outcomes. Its axes and uncertainty should be read with the model choice, reproduction mode, population size and horizon. A visually smooth region is not evidence that an unobserved human population would behave the same way.

The machine-readable register used in the interface is in `src/evidence.mjs`. The exact implemented dynamics and measurement definitions are in [MODEL.md](MODEL.md); [LIMITATIONS.md](LIMITATIONS.md) records the scope boundaries.