Reframing the habitat fragmentation debate around the inferential targets predicted by ecological theory

Juan Andrés Martínez Lanfranco , Erin M. Bayne

The habitat fragmentation debate has persisted for more than three decades because dominant empirical practice has often estimated a narrower quantity than the competing ecological theories jointly require. These frameworks differ not simply in whether fragmentation matters, but in how their predicted effects change along the habitat-amount gradient. The habitat amount hypothesis (HAH) predicts no independent configuration effect once habitat amount is controlled. Threshold dynamics predict increasingly harmful fragmentation effects after structural or demographic breakpoints are crossed at low habitat amount. Colonization–competition coexistence (C–C) tradeoff predicts a sign reversal, with fragmentation neutral or beneficial at high habitat amount where subdivision weakens competitive dominance and harmful where isolation dominates at lower amount. Extinction filtering predicts strongest effects where sensitive taxa remain at high habitat amount, fading as those taxa are lost. In a linear modeling framework, the fragmentation × habitat amount interaction is the minimum linear statistical structure needed to represent this ecological conditionality. The interaction yields β1, the fragmentation effect at mean habitat amount, and β3, how that effect changes along the gradient. HAH predicts β1 ≈ 0 and β3 ≈ 0. Threshold and C–C predict β3 > 0 but differ in the expected sign of β1 and beta-diversity structure. Extinction filtering predicts β3 < 0. The dominant additive model estimates only the fragmentation effect at mean habitat amount and cannot distinguish these alternatives. A null β1 is consistent with true configuration invariance, coexistence crossover, or sampling outside the response-specific recoverability window, the gradient region where effects are both biologically expressed and empirically detectable. Gradient position, beta-diversity decomposition, and metric–mechanism correspondence provide the discriminating signals. Once the target is the conditional response surface, the competing frameworks make distinguishable predictions. The debate can then move from repeated demonstration toward theory investigation.