2.4. Statistics

Data analyses were carried out using R 3.5.2 (R Core Team, 2018), SPSS version 23 (IBM Corp. Released, 2015) and Canoco 5 version 5.11 (ter Braak & Šmilauer, 2018). Time series of climate characteristics were plotted using the “ggplot2” and “tseries” packages in R 3.5.2. Boxplots were drawn in SPSS version 23 for all recorded data to detect outliers in the dataset.

Correlation coefficients were calculated to determine the relationships between phase 1 and phase 2 biomass using SPSS version 23. When the data were not normally distributed, the Spearman non‐parametric correlation coefficient was used.

The raw biomass data were sub‐set by soil type and analyzed using a non‐parametric Kruskal–Wallis test to compare biomass per plant across each soil treatment. The root‐to‐shoot and root mass ratios were calculated and used in determining whether there were significant differences between sterilized and field soil within plant groups in the conditioning phase, using the independent t‐test analysis in SPSS version 23.

A general linear model with the main effects phytometer species, conditioning species and soil treatment was analyzed to determine the feedback of growing phytometers in natural field soil, sterilized soil and inoculum‐added soil, using a three‐way analysis of variance (ANOVA) in SPSS version 23. Three‐way ANOVA comparisons were computed across phytometers to compare biomass per plant across phytometers after the normality and homogeneity of variance assumptions were met for the original biomass data sub‐set by phytometer plant species.

The feedback calculations were based on biomass variable and feedback variables to ith individual observations. The biomass variable was calculated as follows: (O_i–F_i) pairwise as described by Brinkman et al. (2010). The feedback variables were calculated as follows; (a) impact of soil sterilization = [(A _i − B _i)/maximum biomass value of either (A _i or B _i)], where A _i is biomass of the phytometer species grown in non‐sterilized soil and B _i is the biomass of the phytometer species grown in sterilized soil and (b) impact of soil inoculum amendment = [(A _i − C _i)/maximum biomass value of either (A _i or C _i)], where A _i is biomass of the phytometer species grown in soil biota without inoculum and C _i is the biomass of the phytometer species grown in soil biota with inoculum as described by Brinkman et al. (2010). The strength of PSFs was categorized into non‐significant, significant, highly significant and very highly significant, where p > .05, p ≤ .05, p < .01 and p < .001 (IBM Corp. Released, 2015; R Core Team, 2018), respectively.

Paired differences in biomass per phytometer species group were tested for normality and equal variance assumptions after ignoring significant outliers in the dataset. The biomass data did not violate these assumptions in the paired t‐test calculations to determine the strength of plant–soil feedback responses in; (a) “own” relative to “foreign” soil biota and nutrients, (b) “own” relative to “foreign” soil nutrients, and (c) “own” relative to “foreign” soil biota.

Box plots were used to visualize the descriptive statistics of soil nitrogen, FDA activity and active carbon legacies data using the “Car” package in R 3.5.2 software. Where soil nitrogen data violated the normality assumption, the Kruskal–Wallis test was performed to test for differences in soil nitrogen legacies among nine categories of conspecific/heterospecific soil origin. Where significant differences were detected, pairwise contrasts were calculated to reveal the treatments responsible for the differences.

The ANOVA test for normality was met for both soil active carbon (mg/kg) and FDA hydrolyzed (μg/mg soil) using the Shapiro–Wilk test. The significant ANOVA test results were analyzed using multiple comparisons based on Tukey's honestly significant difference (Tukey HSD) test across soil conditioning and feedback phase's legacies in SPSS version 23.

Multivariate analysis of conspecific/heterospecific soil legacies data for PLFA biomass was visualized using non‐metric multidimensional scaling (NMDS) ordination diagram of the Bray–Curtis dissimilarities, with two computed axes of soil fungal and bacterial biomass across treatments (Canoco 5 version 5.11; ter Braak & Šmilauer, 2018).