Abstract
Building retrofit research relies on deterministic models treating occupant behaviour as fixed, an assumption failing in practice. In the Global South, rapid urbanisation creates millions of climate-insensitive housing units, yielding persistent gaps between predicted and actual performance. No validated framework yet combines Socio-Technical Systems theory, stochastic occupant behaviour data, and calibrated simulation into a testable causal model for tropical-highland climates. This study mathematically operationalises STS theory as a predictive engineering instrument beyond descriptive sociology via the novel Integration and Adoption Procedure (IAP) index, validated through two-level Multilevel Modelling with bootstrapped mediation in Global South public housing. A sequential mixed-methods design targeted two Addis Ababa condominium typologies (G+4, G+10). Components were: (1) ASHRAE Guideline 14-calibrated EnergyPlus simulation with behaviour-informed dynamic occupancy; (2) Longitudinal Stochastic Occupant Behavior Tracking protocol (n=36 households, two seasons, twice-daily ASHRAE seven-point reports) converting empirical frequencies to probabilistic inputs; (3) two-level MLM with bootstrapped mediation (5,000 replications, 95% CI), interaction moderation, and IAP as mediator (power>0.85; Cohen’s f²=0.15; ICC=0.21; α=0.05)., Behaviour-informed passive retrofits cut overheating by 22–34% and peak temperatures by 3.8°C versus baseline (487–521 hours/year). IAP correlated with Thermal Resilience (β=0.62, p<0.001) and Perceived Comfort (β=0.58, p<0.001), proving socio-technical integration drives outcomes beyond technical performance. Security-constrained night window closure (p=0.12) eliminated 40–60% of natural ventilation potential, quantifying the behavioural contribution to the tropical-highland performance gap. The Predictive Socio-Technical Modelling Framework scales to rapidly urbanising Global South cities (Nairobi, Dhaka, Lagos, Kampala, Bogotá) where stochastic behaviour, institutional constraints, and climate sensitivity converge on performance gaps.

This publication has 26 references indexed in Scilit: