Stakeholders Involvement, Project Performance, and Service Delivery in Kenya's Water Flagship Projects: A Three-Construct System, Competing Structures, and Cross-Sectional Versus Temporal Coherence

Stakeholder involvement, project performance, and service delivery are usually studied as links in a presumed causal chain, but their joint structure, the alternatives to the chain, and the agreement between project-level and portfolio-level evidence are rarely examined. This study analyzes the three constructs as a system in Kenya's water flagship projects.Survey data from 208 usable questionnaires (65.62% response; stratified sample of 317 of 1,804 projects) were analyzed with partial correlations, maximum-likelihood fit of competing three-variable structures with information criteria, Monte Carlo mediation, dominance and commonality analysis, a Gaussian conditional-independence graph for the four constructs including tendering strategy, and a comparison with recorded annual performance and service delivery indices (2015-2019; n = 5) using correlations, first differences, lags, gap series, and a domain cross-correlation matrix. With tendering strategy added, involvement became a leaf of the conditional-independence graph: its partial correlations with performance (-.031) and service delivery (-.014) were non-significant, and only its link to tendering (.451) survived. Project-level coherence did not simply reappear in time: the recorded composites correlated r = .784 over five years (95% CI -.32 to .98), but their year-to-year changes were negatively related (r = -.67; n = 4), service delivery overtook performance in 2018 (gap +8.4 points), and time and quality performance, not cost, tracked service delivery (r = .69 to .94 against -.01 to .43).The study provides the first system-level account of involvement, performance, and service delivery in Kenya's water flagship projects, shows which structural claims the data can and cannot support, and demonstrates that cross-sectional and temporal coherence are distinct and partly divergent, which cautions against using cross-sectional associations to forecast portfolio dynamics. Received: 13 September 2026 / Accepted: 29 October 2026 / Published: November 2026