Statistical analysis
All statistical tests and visualizations were conducted in R (R
Development Core Team 2008) using the lme4 (Bates et al. 2015)
and MuMin (Barton 2018) packages. We used mixed-effects models with site
as a random effect to examine the relationship between depth, SOC, MB,
clay, and fungi:bacteria and EE activity (expressed on soil mass, SOC,
and MB bases). We similarly used mixed-effects models with site as a
random effect to examine the effect of soil stoichiometry (using ratios
of SOC, total N, and available P) on enzyme stoichiometry. These models
were conducted on the complete dataset, the surface soil dataset (depth
< 20 cm), and the subsoil dataset (depth > 20 cm)
to determine differences in the controls of EE activities between the
surface and subsoils. Because we did not characterize the horizonation
of the sampling pits, we a priori chose 20 cm to represent the
subsoil because most EE studies do not sample below this depth. However,
we also conducted our analysis using a 30 cm threshold, and statistical
significance and overall interpretation remained unchanged. Therefore,
for clarity, we report results using only the 20 cm threshold for the
subsoil. To report the variance explained by the model, we report the
marginal R2 value, which expresses the increase in
explained variance by including the fixed effect(s) (Nakagawa &
Schielzeth 2013). We also used analysis of variance (ANOVA) to determine
if the fraction of EE activity below 20 cm differed by soil order. We
assessed significance at the α = 0.05 level and marginal significance at
α = 0.10. If significant differences were detected, we used Tukey’s Test
of Honest Significant Differences to determine which soil orders were
significantly different.
We used QQ-plots and scale‐location plots to inspect normality and
homoscedasticity, respectively. Because many of the mixed-effects models
failed to meet parametric assumptions, all dependent and independent
continuous variables were natural log-transformed and re-analyzed. The
resulting models, along with the ANOVAs, met the assumptions of
parametric tests. For visualization purposes, data are left
untransformed unless otherwise stated.
RESULTS