Multi-Criteria Prioritization of Rural Road Rehabilitation Using Surface Condition, Population, and Market-Access Indicators

Multi-criteria prioritization ranks candidate projects by combining indicators into a score, and the standard robustness check perturbs the weights. Using 7,068 rural road links across 7,707 km in Osun and Kwara, scored on surface condition, served population, market access, health access and flood exposure, we compare that check against a perturbation of the indicator layers themselves. The comparison that carries the paper uses the one layer whose error has been independently measured. Redrawing the surface condition grade alone, through a confusion kernel reproducing an agreement coefficient of 0.041 between two readings of the same map data, gives a Kendall rank correlation of 0.546 against the base ranking. Drawing all five weights uniformly across the entire admissible simplex gives 0.669. A single indicator, redrawn at its own measured reliability, therefore disturbs the ranking more than abandoning all knowledge of the weights. Perturbing all five layers together gives 0.435 and retains 1.32 of the base top ten against 6.21 under weight perturbation. Only 77 links, 10.9 percent of the top decile and 294 km, hold a top-decile place under both perturbations; the weight check alone certifies 112 and the layer check alone 161, so each test on its own certifies links the other rejects. We conclude that where an indicator layer’s reliability has been measured and is poor, weight sensitivity is the smaller of the two questions, and that where it has never been measured the honest robustness statement is that the ranking’s reliability is unknown.