2.5 Statistical analyses
All analyses were performed using R version 4.0.5 (https://www.r-project.org/). We analyzed litter mass loss using linear mixed effects models (LMMs). Data of the deciduous and coniferous forests were analyzed separately. In each LMM, type of deposited compounds (control, N, Na, PAHs), earthworms (with and without), mesh size (fine and coarse) and time (five sampling dates) were treated as fixed factors. Mesh size was nested in mesocosms and included as random factor to account for non-independence of litterbags within mesocosms and repeated sampling. To evaluate the effects of type of deposited compounds, earthworms and mesh size, we used planned contrasts between the control and the respective treatment. Data were log(x + 1) transformed to meet normality if needed. Due to the response variables being log transformed, contrasts are analogous to log response ratios (Piovia-Scott et al., 2019). We used ‘nlme’ to fit mixed-effect models and ‘emmeans’ for planned contrast. The changes in litter mass loss, C and N loss were shown as percentages and were calculated as\(\left|m_{t}-m_{c}\right|\times 100\%\), with \(m_{t}\) and\(m_{c}\) are the mass loss percentages of the treatment and control, respectively. In another word, the percentages shown in the results are true mass percentages to total litter mass (not the ratio of treatment to control, or the ratio of the difference in treatment and control to control). To evaluate effects of the addition of E. fetida , we modelled abundance and biomass of earthworms at the end of the experiment using generalized linear models with quasi-Poisson distribution to account for model over- or under-dispersion.
Structural equation models (SEMs) were used to inspect pathways linking earthworms and deposited compounds to litter mass loss (see Fig. 4 and Fig. S4; Tian et al., 2018, Yin et al., 2022). To compare the effects of different types of deposited compounds and earthworms on litter mass loss, we merged six models for each forest as one including the three types of deposited compounds and two mesh sizes. In each model, deposited compounds and earthworms were included as categorical variables (with and without) and the other variables as numeric; values of all variables were scaled to 0 - 1 before modeling. Direct effects of Na on litter mass loss were built-in models of the coniferous forest according to the modification indices in R; direct effects of earthworms on SIR were removed to improve fitting (Shipley, 2009; Yin et al., 2022).
3 Results
3.1 Changes in the abundance of earthworms
The addition of E. fetida resulted in significantly increased abundance and biomass of total earthworms including original existingE. fetida in both deciduous and coniferous forests (P < 0.001, Table S1). Specifically, the addition of E. fetidaincreased the abundance of total earthworms by factors of 4.66 and 2.93 in mesocosms of deciduous and coniferous forests, respectively (Table S2).