Skip to content

validation_study.md

Latest commit

 

History

History
127 lines (95 loc) · 5.29 KB

validation_study.md

File metadata and controls

127 lines (95 loc) · 5.29 KB
Raw

PortusSIM Validation Study: Greiner, Ockenfels & Werner (2012)

Summary

This document reports a pre-registered validation experiment comparing PortusSIM's behaviour against a published laboratory experiment on the dynamic interplay of inequality and trust. One of three pre-registered predictions was confirmed. The pattern of mismatch is itself informative about PortusSIM's mechanism design.

Paper tested

Greiner, B., Ockenfels, A., & Werner, P. (2012). The dynamic interplay of inequality and trust — An experimental study. Journal of Economic Behavior & Organization, 81(2), 355–365.

The paper's central findings

Greiner et al. ran a laboratory experiment with human subjects playing a dynamic growth game across multiple periods. Subjects could invest wealth in a trust-based exchange that, if reciprocated, generated growth benefiting both parties. The experiment had two treatments:

  • EQUAL: All players started with equal endowments
  • UNEQUAL: Players started with different endowments

The authors reported three principal findings:

  1. Trust is initially high in EQUAL but decreases over time
  2. Trust is initially lower in UNEQUAL but remains more stable
  3. Wealth disparity narrows over time across both treatments — the EQUAL economy develops inequality and the UNEQUAL economy reduces it, and they partially converge

Translation to PortusSIM

PortusSIM does not directly model "inequality aversion" or "conditional trust" as Greiner et al. do — it models trust accumulation via repeated successful trades. The validation question is whether PortusSIM's mechanism produces qualitatively similar dynamics despite lacking explicit inequality aversion.

Pre-registered experimental design

  • Common parameters held constant: 7 elite + 7 middle + 7 poor merchants, trust_max = 0.15, trust_increment = 0.02, 500 simulation days, identical daily income/expenses across classes
  • EQUAL treatment: all classes start with wealth = 100
  • UNEQUAL treatment: starting wealth 200/100/40 (elite/middle/poor)
  • 20 random seeds per treatment (40 simulations total)
  • Checkpoints: Gini measured at days 50, 100, 200, 300, 500

Pre-registered predictions (BEFORE running)

  • P1: UNEQUAL Gini DECREASES over time (the headline Greiner finding)
  • P2: EQUAL Gini INCREASES over time (inequality emerges)
  • P3: The treatments CONVERGE — gap narrows from day 50 to day 500

Results

Step EQUAL Gini (mean ± 95% CI) UNEQUAL Gini (mean ± 95% CI)
50 0.183 ± 0.012 0.226 ± 0.010
100 0.228 ± 0.017 0.221 ± 0.009
200 0.305 ± 0.017 0.273 ± 0.011
300 0.359 ± 0.014 0.313 ± 0.011
500 0.424 ± 0.015 0.380 ± 0.010

Predictions evaluated

  • P1 (UNEQUAL Gini decreases): NOT CONFIRMED. Gini rose from 0.226 to 0.380.
  • P2 (EQUAL Gini increases): CONFIRMED. Gini rose from 0.183 to 0.424.
  • P3 (Treatments converge): NOT CONFIRMED. The gap inverted: EQUAL ended with HIGHER Gini than UNEQUAL (0.424 vs 0.380).

Overall: 1 of 3 pre-registered predictions confirmed.

Interpretation

The result is partial and informative. PortusSIM reproduces the pattern that dynamic trade interactions generate inequality from equal starting points (P2), matching the spirit of Greiner et al.'s observation that the EQUAL treatment developed inequality over time.

PortusSIM does NOT reproduce the convergence-to-lower-inequality finding that is the paper's headline. Two structural reasons:

  1. No inequality aversion. Greiner's human subjects displayed conditional trust — they adjusted their investments based on observed wealth differences. PortusSIM merchants accumulate trust purely from repeated successful trades; they have no awareness of relative inequality.

  2. No reciprocal growth coupling. In Greiner et al., trust → investment → shared growth, with the rich benefitting the poor when they invest. PortusSIM's trust only affects the probability of trade execution, not the redistribution of trade gains.

The crossover at day ~100 (EQUAL Gini overtaking UNEQUAL) is consistent with these structural absences: in PortusSIM, an equal start gives all merchants similar opportunity to accumulate, but trade interactions inherently advantage some over others, so emergent inequality compounds. In the UNEQUAL start, the initial gap is fixed and only modestly amplified.

What this means for PortusSIM as a research tool

This experiment confirms that PortusSIM is a model of mechanistic trust dynamics, not a model of human social cognition. Its predictions should be interpreted accordingly:

  • ✅ Suitable: questions about how repeated trade interactions affect wealth distribution under different geographic, demographic, or trust-mechanism configurations
  • ❌ Not suitable: questions about how human inequality aversion, conditional cooperation, or psychological reciprocity affect wealth outcomes

The mismatch with Greiner et al. is therefore not a defect — it is a clear specification of the model's scope.

Reproducing this experiment

The exact configuration and run script is available in validation/greiner_2012_test.py. Running it on the same code version will produce bit-identical results (the seeds are 1000–1019 for each treatment).