Routine to fit several spatial models and suggest the best model

spatial.selector Generates a table with the goodness-of-fit statitics to select the best spatial model for an analysis of different analyses (unreplicated and replicated). Models that can not fit are reported as NA. The models require some input conditions from user: 1) Genotypes: fixed/random; 2) Blocks: TRUE/FALSE (if TRUE then fixed/random); 3) Covariates (always fixed, max 2); and 4) Incomplete Blocks: TRUE/FALSE (always random). The model selector will run over a series of models with the conditions: 1) Rows within replicate: TRUE/FALSE (always random); 2) Columns within replicate: TRUE/FALSE (always random); 3) Spline on rows: TRUE/FALSE (fixed linear covariate + random spline); 4) Spline on columns: TRUE/FALSE (fixed linear covariate + random spline); and 5) Residual structure: 'indep' or 'ar1'. In addition the best spatial model (with ar1) and selected according to A-optimality is fitted with an additional nugget effect. List of 32 models is found on data MODLIST.Rda.

spatial.selector(
  data = NULL,
  gen = NULL,
  block = NULL,
  ibk = NULL,
  row = NULL,
  col = NULL,
  cov1 = NULL,
  cov2 = NULL,
  resp = NULL,
  type.gen = "random",
  type.block = "fixed",
  nugget = FALSE,
  model = NULL
)

Arguments

data: dataframe with all relevant columns for spatial model and response variables.
gen: factor name for genotypes (or treatments)
block: factor name for full block (or replicates) (optional)
ibk: factor name for incomplete block (optional) (optional)
row: column name for row coordinates of each experimental unit
col: column name for column coordinates of each experimental unit
cov1: column name with additional covariate 1 (optional)
cov2: column name with additional covariate 2 (optional)
resp: column name for the response variable to analyze
type.gen: model assumption for genotypes: 'random' or 'fixed' (default: 'random')
type.block: model assumption for full blocks: 'random' or 'fixed' (default: 'fixed')
nugget: logical to add nugget to any spatial model fitted (dafualt: FALSE)
model: model number to be fitted (optional)

Value

A table with goodness-of-fit statistics for models evaluated. This includes columns: number of variance components in the model (n.VC), log-likelihood (logL), Akaike information criteria (AIC), Bayesian information criteria (BIC), and heritability based on predictor error variance (heritPEV, only for genotypes 'random').

If parameter 'model' is specified the object with fitted model number is provided, with several objects with details of the fitted model.

aov: Wald-test (mixed model ANOVA-like table) call: String with the ASReml-R call used to fit the requested model gt: Goodness-of-fit statistics for the model evaluated are reported as summary. This includes columns: number of variance components in the model (n.VC), log-likelihood (logL), Akaike information criteria (AIC), Bayesian information criteria (BIC), A-optimality value, logarithm of the D-optimality value, and heritability based on predictor error variance (heritPEV, only for genotypes 'random') mod: ASReml-R object with all information from the fitted model predictions: Predictions for all genotypes, with an additional column of 'weight' is to be used on a on a second-stage analysis. These weights are the diagonal of the inverse of the variance- covariance matrix of predictions

Author

Salvador A. Gezan. VSN International

Examples

# Example 1: Selecting best model from Replicated Trial
# library(agridat)
# testREP <- durban.rowcol
# testREP$bed <- as.factor(testREP$bed)
# testREP$row <- as.factor(testREP$row)
# testREP$gen <- as.factor(testREP$gen)
# head(testREP)
# test.sel <- spatial.selector(data=testREP, gen='gen', row='row', col='bed',
#                              resp='yield', type.gen='random')
# View(test.sel$parms)
#
# # Example 2: Fitting only selected model 27
# mod.sel <- spatial.selector(data=testREP, gen='gen', row='row', col='bed',
#                              resp='yield', type.gen='random', model=27)
# ls(mod.sel)
# mod.sel$call                  # Call for model fitted
# plot(mod.sel$mod)             # Residual plots
# plot(varioGram(mod.sel$mod))  # Variogram fitted model
# summary(mod.sel$mod)$varcomp  # Variance Components
# mod.sel$aov                   # Wald-test Table
# mod.sel$gt$herit.PEV          # Heritability (other statistics on gt)
# head(mod.sel$predictions)     # Predictions for fitted model with additional weights