WHAT DEMAND-MANAGEMENT CONFIGURATIONS IMPROVE PROCUREMENT SATISFACTION IN UNIVERSITIES? A NEURAL-NETWORK-BASED NONLINEAR RECOGNITION AND CSQCA ANALYSIS
Abstract
After the implementation of procurement demand-management rules, whether university procurement satisfaction improves cannot be explained adequately by any single institutional arrangement or isolated managerial measure. Based on 50 valid questionnaire responses collected in September 2024, this study adopts a sequential mixed analytical design. First, neural-network-based nonlinear recognition is used to assess the independent and joint explanatory capacity of graded review rules, procurement-demand coordination, professional support, financial/audit supervision, flexible review arrangements, and several extended governance-context variables. Second, after the neural-network results show that single-condition and net-effect explanations are insufficient, crisp-set qualitative comparative analysis (csQCA) is used to identify sufficient configurations associated with improved procurement satisfaction. The results show that 82% of the cases report improvement, yet the mean balanced accuracy of single-condition neural-network models is only 0.544, and the balanced accuracy of the five-condition model is 0.470. When whole-process cognition, goal understanding, review-team scale, and other contextual variables are added, the balanced accuracy rises to 0.564 and AUC rises to 0.620, suggesting a limited but meaningful nonlinear configurational space. The csQCA results identify no single necessary condition with adequate discriminating power. Under a frequency threshold of 2, a consistency threshold of 0.85, and conservative minimization without logical remainders, three sufficient paths emerge: professional-flexible compensation, graded coordination-supervision, and professional-led transition. The overall solution consistency is 1.000 and the overall solution coverage is 0.415. The findings suggest that university procurement demand-management reform should move beyond the isolated optimization of rules, experts, or review modes and instead construct matched bundles of classification rules, cross-departmental coordination, expertise, supervision, and procedural adaptation.