cwad — Connectivity-Weighted Allocation and Comparison of Field-Plot Designs

cwad is a small, dependency-free R package for plant-breeding trial design.

A design comparison usually changes two things at once: how plots are allocated among genotypes, and whether the analysis borrows information through a kinship matrix. The resulting gain in precision is then reported as a property of the design, although most of it may belong to the analysis model.

cwad keeps the two apart. Every replication allocation is evaluated in one reduced linear mixed model on genotype means, and every allocation can be scored both with and without kinship, so allocation and analysis model enter as crossed factors rather than as a single confounded contrast.

What it does

• A-optimal, connectivity-aware plot allocation via rank-1 Sherman–Morrison updates of the mixed-model equations (greedy_alloc()).
• A crossed allocation × analysis-model comparison table that returns the decomposition of the naive gain (compare_designs()).
• Integer-feasible neutral comparators (balanced_control()), because comparing an integer allocation against a fractional “uniform r = B/N” inflates the apparent gain.
• Two Monte-Carlo stress tests, each running every allocation under both analysis models: rare-outlier shrinkage (sim_transgressive()) and kinship mis-specification (sim_wrongG()).

What it found

On its own 200-genotype, 400-plot example at plot-basis heritability 1/3, the 23.3 % reduction in the mean variance of a genotype difference that a naive comparison attributes to connectivity-weighted allocation decomposes into 22.94 percentage points from using kinship in the analysis and 0.34 percentage points from the allocation itself. Across plot budgets from 1.2N to 3N and heritabilities from 0.10 to 0.67 the allocation contributes at most 1.91 %.

The package is a comparison and calibration tool, not a claim that any one design is universally best.

Install

install.packages("cwad")

# development version
# install.packages("remotes")
remotes::install_github("bkpraveenars-del/cwad")

Quick start

library(cwad)
g      <- make_fs_G(c(rep(24, 5), rep(6, 10), rep(1, 20)))
Ginv   <- solve(g$G); N <- nrow(g$G)
r_unif <- rep(2, N)
r_cwad <- greedy_alloc(Ginv, B = 2 * N, alpha = 2, sig2e = 2)

compare_designs(g$G, alpha = 2, sig2e = 2,
                list(`Uniform r=2` = r_unif, CWAD = r_cwad))

Reading across a row isolates the analysis model; reading down a column isolates the allocation.

Reproducing the published results

Rscript inst/scripts/reproduce.R

regenerates every table and figure. inst/VERIFICATION.md records the checks of the computational engine against independent ground truth.

Note on version 0.1.0

Version 0.1.0 reported the 23 % figure above as a property of the design. It was not: allocation and analysis model had been varied in a single step. Version 0.2.0 reports the same engine, the same example and the same numbers, decomposed, and the API no longer permits the confounded comparison to be expressed. Scripts calling the v0.1.0 compare_designs(), sim_transgressive() or sim_wrongG() will need updating.

License

GPL-3