This study investigates the complex configurational mechanisms driving breakthrough green innovation (BGI) among construction firms, which remain inadequately explained by conventional variable-centered approaches.
Grounded in push-pull theory, a hybrid analytical framework is developed that integrates machine learning for antecedent screening with dynamic qualitative comparative analysis (QCA) for configurational identification. Panel data from 65 A-share listed Chinese construction enterprises spanning 2017 to 2024 are analyzed. Ensemble machine learning algorithms with SHAP attribution identify six core antecedent conditions across push, pull and resistance dimensions, which are then subjected to dynamic QCA using the Garcia-Castro and Ariño (2016) panel set-theoretic framework.
No single antecedent condition constitutes a necessary condition for BGI. Four sufficient configurational pathways are identified (overall consistency = 0.764, coverage = 0.217), classified into push-dominated and pull-dominated archetypes. All pathways demonstrate cross-period structural stability, though their individual coverage varies across configurations (raw coverage ranging from 0.060 to 0.127), indicating differences in their empirical prevalence.
Firms facing stronger external pressure should strategically leverage peer demonstration effects, which can substitute for internal innovation resources in driving green innovation. Policymakers should prioritize targeted financing support mechanisms and equity governance improvements to ease the structural frictions surrounding breakthrough innovation.
This study is among the first to integrate machine learning dimensionality reduction with dynamic QCA for studying configurational drivers of BGI in construction. The hybrid paradigm provides a replicable methodological template while advancing push-pull theory's application to organizational green innovation research.
