Unless a multivariate model is being fitted or missing values
are being estimated, ASReml discards records with a missing response variable
as the data is being read. The data file is not read until the
linear model line has been parsed and the response variable identified.
Warning:
Consequently, the data summary and any
tabulation
is based on those records where the response is not missing.
Estimating the missing values
It is sometimes
computationally convenient to estimate missing values, for example, in spatial
analysis of regular arrays, see example below.
This is done by including mv in the fixed model.
Formally, mv creates a factor with a covariate for each
missing value. The covariates are coded 0 except in the record where the
particular missing value occurs, where it is coded -1.
Missing values in multivariate analysis
In
multivariate
analyses, the R matrix is
modified to reflect the missing value pattern.