Simulated Soil Data with Inherent Factors and a Nested Design

180 simulated soil samples carrying the columns an inherent-property adjustment needs and neither other dataset has: soil type, land-use history, a current management factor, and a plot identifier. Five samples per plot, 36 plots.

Usage

soil_inherent

Format

A data frame with 180 rows and 16 columns:

SampleID

Character. Unique identifier (INH001-INH180)

PlotID

Character. Plot the sample came from (P01-P36), five samples each

soil_type

Factor. Acrisol, Cambisol or Fluvisol – inherent

land_use_history

Factor. primary_forest, secondary_forest or long_pasture – inherent

management

Factor. conventional or improved – not inherent, and the thing an index is meant to detect

Sand, Silt, Clay

Numeric. Particle-size fractions (%), summing to 100

BD

Numeric. Bulk density (g/cm3)

pH

Numeric. Soil pH in water

OM

Numeric. Organic matter (%)

SOC

Numeric. Soil organic carbon (%)

N

Numeric. Total nitrogen (%)

P

Numeric. Available phosphorus (mg/kg)

K

Numeric. Exchangeable potassium (mg/kg)

CEC

Numeric. Cation exchange capacity (cmol/kg)

Source

Simulated. The generating script is data-raw/create_soil_inherent.R, which states every effect size used, so the injected truth can be checked against what an analysis recovers.

Details

Why the three factors are separated. A soil should not be scored down for being what its parent material made it. Maaz et al. (2023) built scoring functions that account for inherent properties precisely to stop them biasing the overall score, and no additive-index paper in the corpus does the same. Demonstrating that requires data where the inherent and the manageable are distinguishable, so here they are generated separately:

Factor	Inherent?	What it drives
soil_type	yes	Clay, CEC, pH
land_use_history	yes	legacy OM and N
management	no	OM, P, BD

The two-sided property that makes it useful. Regressing an indicator on the two inherent factors and keeping the residuals should remove the soil-type difference and leave the management difference standing. An adjustment that erased both would be useless. On this data the management signal does not merely survive – it sharpens, because the inherent variation was masking it:

Indicator	soil_type p	management p
OM	0.83 to 1.00	2e-04 to 6e-23
CEC	5e-56 to 1.00	0.09 to 8e-07
pH	3e-91 to 1.00	0.48 to 5e-03

How much of each indicator is inheritance. The R-squared of the inherent model is informative in its own right: Clay 0.95, pH 0.94, CEC 0.88 and OM 0.82 are mostly inherited, while P 0.22 and BD 0.19 are mostly managed.

It is nested, on purpose. Five samples share each plot, and the plot-level variance deliberately exceeds the within-plot variance, giving intraclass correlations of 0.81 to 0.99. Maaz found ICC above 0.75 for every indicator – samples within a plot are not independent, which invalidates ordinary standard errors, and most soil campaigns never check.

See also

soil_structured for covariance structure without design factors; soil_data for the simple case

Examples

data(soil_inherent)
table(soil_inherent$soil_type, soil_inherent$management)
#>           
#>            conventional improved
#>   Acrisol            30       30
#>   Cambisol           30       30
#>   Fluvisol           30       30

# Parent material dominates the exchange capacity
tapply(soil_inherent$CEC, soil_inherent$soil_type, mean)
#>   Acrisol  Cambisol  Fluvisol 
#>  9.128333 12.695000 15.446667 

# Management moves phosphorus, and parent material does not
tapply(soil_inherent$P, soil_inherent$management, mean)
#> conventional     improved 
#>     6.736667     9.605556 

# Adjusting for inheritance leaves the management signal standing
adjusted <- residuals(
  lm(CEC ~ soil_type * land_use_history, data = soil_inherent)
)
anova(lm(adjusted ~ soil_inherent$management))[["Pr(>F)"]][1]
#> [1] 8.136476e-07