NEWS

SpATS 1.0-19 (2024-10-16)

Bug corrected. In the PSANOVA function, the number of segments, nseg, must be a multiple of the divisor of the number of segments, nest.div. This is now checked, and an error is returned if the condition is not met.
Bug corrected. In the PSANOVA and SAP functions, the number of segments (nseg), order of the polynomial (degree), and penalty order (pord) should be integers. This is now checked, and if any of these values are not integers, they will be rounded to the nearest integer, with a warning issued to inform the user.

It is now possible to include the interaction between genotypes and a treatment factor when both are modelled as fixed effects.

In the new version, we allow the user to specify that the fitted spatial trend (2D surface) should be centered at zero for the observed data (i.e., the average of the fitted spatial trend will be zero at the observed data). To that end, the "spatial" functions SAP() and PSANOVA() have a new argument, center. By default center = FALSE for compatibility with previous versions of SpATS.

The controlSpATS function has now a new argument which allows the dispersion parameter for the Gaussian family to not be estimated (it is fixed at 1).

The plot function now allows the fitted spatial trend to be depicted either in the original scale (raw) or as a percentage of the (average) observed response variable of interest across the field.
Predictions for fixed factor effects can now be obtained conditional on the reference value (predFixed = "conditional"), or averaging over all levels of the fixed factor (predFixed = "marginal"). For compatibility with previous versions of SpATS, by default predictions are obtained conditional on the reference value.

The way of calculating the nominal dimension associated to each random term in the model has been corrected. The nominal dimension corresponds to the upper bound for the effective dimension (i.e., the maximum effective dimension a random term can achive). This nominal dimension is now calculated as \eqn{rank[X, Z_k] - rank[X]}, where \eqn{Z_k} is the design matrix of the k-th random term and \eqn{X} is the design matrix of the fixed part of the model. In most cases (but not always), the nominal dimension corresponds to the model dimension minus one, “lost” due to the implicit constraint that ensures the mean of the random effects to be zero. For the genotype (when random), the ratio between the effective dimension and the nominal dimension corresponds to the generalized heritability proposed by Oakey (2006). A deeper discussion can be found in Rodriguez - Alvarez et al. (2018).