We therefore also analyzed the effect of SOC on the three response variables in a model resulting from backward model reductionofthefullmodelona5%significancelevelwiththemodificationthatthemaineffectofSOCwaskeptinthemodelevenifit wasnon-significant.Thereducedmodelissubsequentlyreferredto as the “final model”. Following model reduction, the effect of SOC andtheotherremainingvariableswastestedbycomparisonwitha modelsimilartothefinalmodelbutwithoutthevariabletestedfor significance. After model reduction, variables were removed that were not significant when SOC was included in the model. If the test showed a significant effect of SOC, there are two possibilities. Either there is a causal effect of SOC, or a confounding variable with a causal effect on the response variable has been removed. Hence, we can not be sure that there is an effect of SOC. However, in the final model, if there is no effect of SOC we can with stronger confidence say there is no evidence of an effect of SOC. Model selection was done by F-tests when no random effects remainedandotherwisebychi-squaretests.Coefficientsinthefinal modelsarereportedwith95%confidenceintervalsandp-valuesfor beingnon-zero.Incaseofmultipletestswithineffects,thep-values forbeingnon-zeroarealsoreportedwithadjustmentusingthefalse discovery rate (Storey, 2002). For models with random effects the parameter p-values are computed by simulations, and may differ slightly from the p-values used in model selection. All statistical analyzes were performed using the Linear Mixed Effectspackage(lme4),Goodnessoffit(GOF)andstandardRpackages (www.r-project.org).