2.4 Statistical analysis.
For each chemical and physical parameter, comparisons between treatments were carried out using a two-way ANOVA following the general linear model of the ”CAR” R package (Fox et al., 2012), followed by posthoc multiple comparisons (LSD test). All the data was checked to meet the assumptions of normality of residuals and homoscedasticity and transformed when needed. For P Olsen, we used a Negative Binomial Hurdle model approximation because of the amounts of zeros (concentrations below the detection limit). Soil data were subjected to a principal component analysis (PCA) for data ordination and reduction and distinguished if conditions were different, and determined what groups of soil indicators were more effective on differentiating grazing intensity. The data was auto-scaled previous to the PCA because of the different natural scales between variables and to attribute an equal weight to each variable (Sena et al.,2002). All statistical analysis was conducted in R (R Core Team, 2013).
3. RESULTS