Location-scale models and cross validation to advance quantitative evidence synthesis.

Area of Science:

• Ecology
• Ecological Synthesis
• Quantitative Ecology

Background:

• Generality in ecology is often assessed via central tendencies, with less focus on heterogeneity.
• Ecological evidence syntheses typically assume constant variance (homoscedasticity) in effect sizes across studies.
• Assessing the transferability of ecological effects across diverse contexts is crucial but challenging.

Purpose of the Study:

• To demonstrate how location-scale models and cross-validation enhance ecological evidence synthesis.
• To explore scale-dependent heterogeneity in plant species richness relationships.
• To investigate factors influencing species richness in habitat fragmentation studies.

Main Methods:

• Utilized location-scale models to relax the assumption of homoscedasticity in meta-analyses.
• Applied cross-validation techniques to assess model transferability and generalization.
• Conducted two case studies: one on native-exotic plant species richness and another on habitat fragmentation.

Main Results:

• Heteroscedastic models quantified the certainty of effect transferability across contexts.
• Scale-dependent heterogeneity was identified in plant species richness relationships.
• Habitat fragment size and study covariates influenced unexplained variation in species richness.

Conclusions:

• Location-scale models provide a more nuanced understanding of ecological effect transferability.
• Cross-validation is a valuable tool for assessing the generalization of ecological findings.
• Assuming homoscedasticity can limit the assessment of transferability in ecological evidence syntheses.