Model selection results for trophic cascade analysis.

We used LMMs with terms for average temperature for ecosystem j in week w (Twj), weeks 2–9 (Wk), and their interaction. We treated the power level (e.g., 100 W, 200 W, etc.), our temperature treatment, as a random effect to account for repeated measures on ecosystems over time. We compared models using likelihood ratios (LogLik), AICC, Akaike weights (w), and δAICC weights. The model was fit to 79 observations in 10 groups. The full model (TCFull) includes all terms, and models representing alternate hypotheses excluded terms indicated by “NA.” Coefficients were pooled (Methods: Statistical analysis) to estimate slopes and intercepts for Fig 2.