12 ASReml
12 ASReml
12 ASReml
ASReml is a statistical package that fits linear mixed effects models using Residual
Maximum Likelihood (REML). It has been under development since 1993 and is a joint
venture between Biometrics Program of NSW Agriculture and the Biomathematics Unit
previously the Statistics Department of Rothamsted Experimental Station.
Linear mixed effects models provide a rich and flexible tool for the analysis of many data
sets commonly arising in the agricultural, biological, medical and environmental sciences.
ASREML has applications in the analysis of typical applications include the analysis of
• (un)balanced longitudinal data,
• repeated measures data (multivariate analysis of variance and spline type models)
• (un)balanced designed experiments,
• multi-environment trials analysis,
• univariate and multivariate animal breeding and genetics data (involving a relationship
matrix for correlated effects)
• and the analysis of regular or irregular spatial data.
Further
¾ The engine of ASReml forms the basis of the REML procedure in GENSTAT. An
interface for S-PLUS called samm is also available.
¾ Handles large data sets (of 100,000 or more observations/effects).
¾ Supports a wide range of variance models for spatial analysis.
Command File
By convention an ASReml command file has a .as extension. The command file consists
of five sections. These are:
I. Title
II. Definition of data columns
III. Name of the data file
IV. Linear model
V. Variance structure (when necessary)
I. Title Line: The first 40 characters of the first non-blank line in an ASReml command
file is taken as the title for the job. It is used to identify the analysis for future reference.
II. Definition of Data Columns: The Labels for the data field must be given in the order in
which they appear in the data file. The data field definitions must be indented and are
given as
SPACE label [field_type]
Here
SPACE is a required space,
label is an alphanumeric string beginning with a letter and is of maximum of 31 characters
of which only 20 are printed and
field_type indicates how a variable is interpreted if specified in the linear model.
i. For a variate, leave field_type blank or specify 1.
ii. For a model factor, use * or n if the data field has values from 1 to n; !A n if the data
field is alphanumeric; and !I n if the data field is numeric.
III. Name of a Data File: A data file must always be specified after defining the data
columns with complete path if it is not in the same directory as the command file. Its name
must begin in the first position of the line. There are many qualifiers that can be placed on
this line after specifying the name of the data file. The most commonly used are
!skip n indicates to skip over the first n lines (those containing column
headings) in the data file
!maxit m establishes the maximum number of iterations in m. The default is 10
iterations.
IV. Linear Model: The linear model is a list of terms, each separated by a space in the
form
<variable Y> ~ <model>
<variable Y> is the name of the data field which will be analyzed.
<model> lists the terms of the model. <variable Y> is separated from <model> using the
symbol (~).
Some common model terms are:
mu represents a constant term or the intercept
<name> is the name of an explanatory variable or factor
<name>.<name> is the interaction of two terms
!r indicates that the following term is random
!f indicates that the following term is fixed
I-220
ASReml: An Overview
V. Variance structure: In a linear mixed effects models, there are two variance structures,
one for the errors, known as R-Structure and one for the random effects known as G-
structures. The default option is independently and identically distributed errors and
random effects. For specifying a given structure, we use the variance header line as
For example:
Yield~ mu var !r rep
121 # Variance header line indicates that there is one R structure that involves two
variance models and there is one G structure
col col AR1 0.1 # R-Structure indicates that there is auto-regressive structure to each
of the dimensions row and column using 0.1 as initial correlation.
row row AR1 0.1
rep 1 # G-structure header line indicates one variance model.
rep 0 IDV 0.1 # Variance for the replicates is IDV of order 4 σ r2 I 4 . 0.1 is a starting
value for γ r = σ r2 / σ e2 .
Besides, the above, one can get the predicted values/BLUPs for the effects using ASReml.
It gives BLUP directly but it does not do pair wise comparisons. For pairwise comparisons,
there is an approximate test. Take the difference of two BLUPs, if difference is more than
twice the SE of difference, then these two are significantly different.
I-221
ASReml: An Overview
I-222
ASReml: An Overview
# !A variable is alphanumeric
# !GP attempts to keep the parameters in the theoretical parameter space.
-ve values of variances
are replaced by small positive values
# Diag diagonal variance covariance matrix
[End of File]
For a detailed description on the use of ASReml one may refer to User Manual of ASReml.
Some details can also be obtained in the following manual which can be downloaded from
www.cimmyt.org/english/wps/biometrics.
Burgueno, J., A. Cadena, J. Crossa, M. Banziger, A.R. Gilmour and B. Cullis (2000).
Users’s Guide for Spatial Analysis of Field Variety Trials Using ASREML. Mexico,
D.F.: CIMMYT.
I-223