Automatically detects violations in fitted experimental models (non-normality,
heteroscedasticity, boundary singularities, or overdispersion) and executes optimal
guided remedies: Box-Cox power transformations, variance heterogeneity modeling
(weights = varIdent()), family transitions to GLMM (Gamma, Negative Binomial, Beta),
or non-parametric alternatives (Kruskal-Wallis / Friedman).
Usage
remedy(
fit,
method = c("auto", "boxcox", "log", "sqrt", "varIdent", "glmm_gamma", "glmm_nbinom",
"nonparametric"),
lambda = NULL,
...
)Arguments
- fit
An object of class `"agri_fitted_model"` returned by
fit_experiment.- method
Character string specifying remedy method: `"auto"` (default: chooses optimal remedy based on violation audit), `"boxcox"` (optimal Box-Cox power transformation), `"log"` (natural logarithm transformation), `"sqrt"` (square root transformation), `"varIdent"` (heterogeneous variance modeling per treatment group via
nlme), `"glmm_gamma"` (transition to Gamma GLMM with log link), `"glmm_nbinom"` (transition to Negative Binomial GLMM for overdispersed counts), or `"nonparametric"` (Kruskal-Wallis or Friedman rank-sum test).- lambda
Optional numeric value for Box-Cox power parameter. If `NULL` and `method = "boxcox"`, it is estimated via maximum profile log-likelihood.
- ...
Additional arguments passed to refitting functions.
Value
An S3 object of class `"agri_remedy"` containing:
- original_fit
Original fitted model object.
- remedied_fit
New fitted model object with remedy applied.
- remedy_type
Description of the applied remedy.
- comparison_table
Side-by-side comparison of AIC, BIC, LogLik, and assumption p-values.
- lrt_test
Likelihood Ratio Test comparison (if applicable).
- boxcox_details
Details on optimal lambda estimation (if Box-Cox is used).
- verdict
Summary evaluation of remediation efficacy.
Examples
data(wheat_splitplot, package = "agriDesignR")
fit <- fit_experiment(
data = wheat_splitplot,
response = "grain_yield",
main_plot = "temperature",
sub_plot = "genotype",
block = "block"
)
rem <- remedy(fit, method = "boxcox")
print(rem)
#>
#> ======================================================================
#> agriDesignR: Statistical Remedy & Assumption Resolution
#> ======================================================================
#> Remedy Applied : Box-Cox Power Transformation (lambda = 1.9)
#>
#>
#> Before vs. After Model Comparison
#> ----------------------------------------------------------------------
#> Metric Original_Model
#> Model Formula grain_yield ~ temperature * genotype
#> Engine / Family lme4 (gaussian)
#> AIC 188.09
#> BIC 210.04
#> Residual Normality (Shapiro p) 0.6964
#> Variance Ratio (max/min) 21.36
#> Overall Health Status ACCEPTABLE_WITH_CAUTIONS
#> Remedied_Model
#> grain_yield_trans ~ temperature * genotype
#> lme4 (gaussian)
#> 397.53
#> 419.49
#> 0.5648
#> 13.78
#> ACCEPTABLE_WITH_CAUTIONS
#>
#> Remedy Verdict : [OK] REMEDY SUBSTANTIALLY IMPROVED: Residual metrics improved significantly and are now well within acceptable limits for robust inference.
#>
#> ======================================================================
#> Tip: Use 'remedy_fit$remedied_fit' to extract the corrected model for post-hoc analysis.
#>