The potential effects of precipitation regime and landscape position on response variables (i.e., soil hydraulic properties and other characteristics measured at the eight sampling locations) were investigated through mixed-effects linear modeling. In the cases of soil porosity, carbon, nitrogen, and EDP, models were fit with all data records included. However, subsets of data were used in models of infiltration rate (subsetted by pressure potential), θv (divided into intervals of 0.5 pF units (log10[−hPa]), particle size [divided into intervals of integer log10(μm) units], aggregate stability (subsetted by sieve size), and CWM root diameter (divided into 5-year intervals). This allowed us to estimate the effect sizes of precipitation and landscape position on response variables in the absence of nonlinear effects of pressure potential, pF, particle/aggregate diameter, or community transitions. In the analyses of infiltration rate and particle size, data were log-transformed to achieve residual normality. Precipitation, landscape position, and their interaction were entered into models as fixed-effects terms. Random-effects terms were also included to account for variation due to transect identity and interactions with precipitation regime and landscape position, as well as for repeated measurements within plots and samples, if applicable. Continuous variables were standardized by two SDs, while discrete variables with two levels were coded as ±0.5, making all coefficients comparable in scale (42). Rather than using binary indicators of significance (i.e., P values) to gauge the importance of model terms, we evaluated terms’ effect sizes and the strength of evidence that their coefficients were nonzero (43). Effect sizes were determined from the absolute values of model-averaged coefficients, with larger deviations from zero being indicative of greater effects. Model averaging involved fitting the full model, all hierarchically complete reduced models, and the intercept-only model. For model variants that had corrected Akaike information criteria (AICc) scores sufficiently close to the variant with the smallest AICc (ΔAICc < 6), standardized coefficients for all retained fixed-effect terms were averaged. Evidence of an effect being nonzero was considered strong when the 95% confidence interval of the corresponding standardized, model-averaged coefficient excluded zero.