Statistical Test of Network Analysis: Can it Detect Differences in Food Web Properties

• Published in: Ecosystems (New York) 2007-09, Vol.10 (6), p.906-923
• Main Authors: Dame, James K, Christian, Robert R
• Format: Article
• Language: English
• Subjects: Animal and plant ecology; Animal, plant and microbial ecology; Applied ecology; Biological and medical sciences; Biomass; Conservation, protection and management of environment and wildlife; Disturbance; ecological network analysis; ecopath; Ecosystem management; Ecosystem models; ecosystem-based management; Ecosystems; Environmental conditions; Food chains; food web; Food webs; Fundamental and applied biological sciences. Psychology; General aspects; Insect larvae; Marine ecosystems; Network analysis; Ponds; salt marsh pond; Salt marshes; stress; Studies; Synecology; Taxa; Trophic levels; trophic network; Variance analysis
• ISSN: 1432-9840; 1435-0629
• DOI: 10.1007/s10021-007-9068-1

Ecological network analysis (ENA) is a modeling approach increasingly being used to examine food webs. However, most studies do not replicate networks, and a statistical evaluation of ENA is lacking. The major objectives of this study, therefore, were to evaluate statistically the effectiveness of ENA in detecting differences in food web properties and to compare ENA output with established community level indices. Quantitative trophic networks (n = 12) representing four high salt marsh ponds during three times (corresponding to low stress, high stress, and post-disturbance) were constructed from an extensive field sampling program augmented by literature values. Food webs of salt marsh ponds were used because these systems contain relatively simple food webs, have well defined boundaries, and allow for adequate replication. A null hypothesis was tested to determine how values of 12 indices from ENA output differed among the three stress/disturbance conditions (H o: low stress = high stress = post-disturbance). Results of both ANOVA and Friedman's tests indicated most ENA indices were significantly different among the three stress/disturbance conditions. The amount of covariance among the indices was relatively low (7 of 66 were significant). Results were compared to differences in community indices (richness, evenness, and diversity) among the three stress/disturbance conditions. ENA output identified differences beyond those recognized by the community indices. Overall, networks were unique enough under different environmental conditions to provide statistically significant differences in ENA results. Our findings are supportive of the use of carefully constructed networks in food web analysis and for decision making in ecosystem-based management.