Structural Luck: An Agent-Based Model of Workplace Career Dynamics (1.0.0)

Why do career outcomes in organizations follow highly skewed distributions when the individual traits presumed to drive success — competence, effort, and social skill — are approximately normally distributed in the population? This agent-based model investigates the relative contributions of individual attributes versus random events (“luck”) to career outcomes in hierarchically structured organizations.

The model places 500 agents in a five-level pyramid-shaped organization over 80 six-month evaluation periods (40 simulated years). Each agent is characterized by four attributes drawn from normal distributions at initialization: competence, effort, social skill, and adaptability. In each period, agents may encounter stochastic career events (good or bad projects, supportive or poor managers, market booms or busts), which modify their performance and visibility scores. Promotion decisions are made competitively when vacancies arise at higher levels, using a weighted combination of recent performance, accumulated visibility, network position, and tenure. Social connections form and decay through a proximity-based model influenced by agents’ social skills.

The model extends the talent-versus-luck framework of Pluchino et al. (2018) in four substantive directions: (1) it embeds agents in a hierarchical organization with finite positions at each level, making promotion a zero-sum competition rather than an abstract encounter with random events; (2) it represents individual differences along four independent attribute dimensions rather than a single talent score; (3) it incorporates dynamic social network formation that mediates access to career-relevant opportunities; and (4) it implements the Peter Principle mechanism through which promoted agents must acquire competence appropriate to their new role.

Using permutation-based feature importance decomposition across 1,900 simulation runs spanning seven experiments, the model finds that random events account for 43–60% of the variance in final career level, exceeding the combined contribution of competence (17–18%) and effort (19–20%). Key findings include: organizational hierarchy depth amplifies the role of luck by creating bottlenecks where timing and circumstance decide which of similarly qualified candidates advances; preferential-attachment network dynamics reduce the meritocracy index by approximately 50% relative to a no-network baseline; the Peter Principle emerges when skill transferability across hierarchical levels falls below approximately 0.75; and “blind evaluation” policies that remove visibility and network effects from promotion decisions can raise the meritocracy index from 0.45 to 0.95. A systematic one-at-a-time sensitivity analysis across 750 simulation runs confirms that the qualitative conclusion — luck is a major and often dominant structural factor in career outcomes — is robust across a wide range of parameter assumptions.

The model is implemented in Python using the Mesa agent-based modeling framework and is fully configurable through experiment configuration dictionaries. All seven experiments from the accompanying paper can be reproduced using the provided run scripts.

Release Notes

This is the first public release of the model. This release contains the complete model code and analysis pipeline used to produce all results, tables, and figures reported in the paper.

Associated Publications