Dataset DOI: 10.5061/dryad.0rxwdbshg

Dataset overview

This deposit contains the plot-level analysis dataset and R scripts supporting the associated Journal of Ecology article. The dataset has 64 rows and 26 columns. Each row represents a different 5 m x 5 m experimental plot. All observations were collected in August 2025, during the fourth growing season of an experiment established in May 2022 at Yanchi, Ningxia, China (37 degrees 49 minutes N, 107 degrees 22 minutes E).

Plant species richness treatments are 1, 4, 6, 8 and 10 species, represented by 30, 9, 9, 9 and 7 plots, respectively. Soil samples combine five cores from the 0-10 cm layer per plot. Biomass was harvested within one 1 m x 1 m quadrat per plot.

The data include plant richness, realised legume biomass proportion, six community-weighted leaf traits, two herbivore feeding-guild abundances, nematode faunal indices and composition-derived footprint metrics, biomass numerator/denominator values, and soil physicochemical measurements. The main statistical scripts do not use the supplementary soil columns or the exported c-p 1 percentage.

Authors and contact

Chengru Ma; Xu Luo; Cui Han; Jiali Lian; Xueqin Yang; Siyi Liu; Jianping Li.

Data and analysis contact: Chengru Ma, leomcr@163.com, Ningxia University, China.
Corresponding author of the article: Jianping Li, lijianpingsas@nxu.edu.cn.

Funding: Ningxia Key Research and Development Program (2026BEG02047) and Natural Science Foundation of Ningxia (2023AAC02029).

Files and their relationship

Unzip analysis_code.zip in the same directory as analysis_dataset.csv. The result
should have run_all.R, install_packages.R, CODE_README.md, R/ and environment/
beside analysis_dataset.csv. Do not leave the CSV inside a separate download folder.

Variables, units and missing values

The complete field-level dictionary is data_dictionary.csv. Column names are case-sensitive and include legacy spaces, brackets and punctuation. Read them with check.names = FALSE in R. The scripts use Plot_ID as an identifier, not a numeric value or a treatment code. Use Species_Richness for the richness treatment.

• RATIO is legume dry biomass divided by total plant dry biomass, a fraction between 0 and 1.
• The two biomass columns Legume_Proportion and total Proportion are in grams per sampled quadrat; despite their names, neither is a proportion.
• CA is leaf contact angle (degrees).
• LDMC is dry mass divided by saturated fresh mass (g g^-1).
• LMA is leaf dry mass per area (g m^-2).
• LTM is density-normalised leaf punch strength (J kg^-1).
• TD is trichome density (trichomes mm^-2).
• leafN is mapped to Nmass and is expressed in g nitrogen per 100 g leaf dry mass. The six leaf-trait columns are biomass-weighted community means for each plot.
• Chewing and Sucking are numbers of chewing and piercing-sucking insects per
  plot, pooled across the sampling methods. Total herbivore abundance is calculated in the scripts as Chewing + Sucking.
• SI, EI and CI are the structure, enrichment and channel indices, respectively.
• MF, BF, FF and HF are the composite, bacterivore, fungivore and herbivore footprint metrics. HF must not be substituted for MF.
• The nematode indicators are supplied NINJA outputs derived from genus-level relative sequence composition and database parameters. The footprint values are retained on their output scale. They are not microscopic individual counts, absolute carbon flux measurements or energy-use efficiencies. The legacy [MEAN] labels are export labels, not means across richness treatments.
• FF is zero in all 64 rows. Zero is a supplied value, not a missing-value code. Constant FF is excluded from min-max scaling and the 40 estimable correlations. Undefined correlations involving FF are saved as NA in output tables.
• SOM g/kg contains soil organic matter, TN and TP contain soil total nitrogen and total phosphorus (all g kg^-1). pH is dimensionless. EC is in microsiemens per centimetre (uS/cm). Soil:water ratios for pH and EC are 1:2.5 and 1:5.
• The input CSV has no missing cells. Missing or non-estimable values in generated output tables are written as NA, not replaced by zero.

Data preparation and provenance

The analysis starts from the supplied plot-level community-weighted traits and NINJA indicator outputs. Leaf-level observations, species-by-plot biomass/trait input tables, and sequence reads are not contained in this deposit. The workflow therefore reproduces the reported statistical analyses from the analysis table; it does not reconstruct community-weighted means or NINJA outputs from upstream measurements. Figure 1 maps, photographs and sampling artwork are not part of the statistical workflow.

The soil field originally labelled SOC g/kg has been relabelled SOM g/kg to reflect the measurement identified by the authors. No organic-carbon/organic-matter conversion factor was applied. All 64 data rows and all numerical values are unchanged from the supplied analysis table. Nitrogen, biomass and conductivity units are documented as confirmed by the authors; no numerical rescaling was applied.

Code and software

The analyses were run with R 4.5.1. Required packages and the recorded versions are listed in environment/package_versions.csv inside analysis_code.zip. The environment/sessionInfo.txt file records the prior complete verification run.

Open a fresh R session in the directory containing run_all.R and the input CSV:

# Only if required packages are missing:
source("install_packages.R")

# Execute the complete workflow:
source("run_all.R")

Alternatively, run R script run_all.R from that directory in a terminal. The optional installer installs missing packages. Use the recorded versions for closest replication.

Each run creates its own output/run_YYYYMMDD_HHMMSS directory, copies the input to its data subdirectory, and writes tables, models, figures and logs into its output subdirectory. RUN_STATUS.csv records module success or failure. An error stops the workflow.

The workflow includes additive trait models; factorial OLS and HC3 inference; paired feeding-guild mixed models and five Holm-adjusted conditional contrasts; PCA and PERMANOVA; random forests; nematode sensitivity analyses; descriptive LOESS summaries with 2,000 stratified whole-plot bootstrap replicates; 40-test BH-adjusted correlations; and the retained piecewise SEM. Random seeds are explicitly set. The primary analyses retain all 64 plots.

Figures 2-7 are statistical reconstructions from the archived data. Layout, fonts, and decorative elements need not be pixel-identical to the published artwork. The Fig. 5 reconstruction includes all 64 points. The Fig. 7 reconstruction shows fitted connections and the residual association; hypothetical connections not fitted in the final model are not statistical model outputs.

Interpretation of generated results

• Standardised interaction models use centred/scaled responses and predictors.
• Main-effect coefficients are conditional slopes at the other predictor's sample mean.
• The five richness-specific guild contrasts are one Holm adjustment family. Their confidence intervals are pointwise 95% intervals, not Holm-adjusted intervals.
• Random-forest importance uses type = 1, scale = TRUE. Output tables retain negative raw importance scores and a separate nonnegative display column. These values are not percentage increases in MSE or significance tests.
• Fig. 6 uses equal-weight plot-level module means, fixed original min-max ranges, and LOESS degree = 1, span = 1. Resampling is by whole plot within richness
  treatment. Bands are pointwise 95% percentile intervals. The main heatmap uses
  BH-adjusted P values; panel b uses raw OLS P values. Extra unadjusted figures
  are explicitly named exploratory and are not the main heatmap.
• SEM component models use original response scales; standardised coefficients are calculated after fitting. The log-likelihood AIC is approximately 2989.84. AIC values from differently transformed response scales are not comparable. The fitted TD-MF path is included in tables even though it is not drawn in the reconstructed diagram.

Access, reuse and citation

The data and author-provided analysis scripts in this Dryad deposit are made available under CC0 1.0 Universal: https://creativecommons.org/publicdomain/zero/1.0/.
Third-party R packages remain subject to their own licences and are installed separately. No third-party images or journal PDFs are included.

Please cite this dataset using the DOI provided on this Dryad record, together with the associated article citation when appropriate.