Fit linear mixed models using advanced restricted maximum likelihood (reml) techniques.



Find resources, tutorials and more in the Knowledge Base.

Download ASRemlDownload ASReml

Harness the power of REML

ASReml is powerful statistical software specially designed for mixed models using Residual Maximum Likelihood (REML) to estimate the parameters. Linear mixed effects models provide a rich and flexible tool for the analysis of many data sets commonly arising in animal, plant and aqua breeding, agriculture, environmental sciences and medical sciences.

    Using the Average Information (AI) algorithm and sparse matrix methods ASReml handles large or extremely large and complex data analysis (of 500,000 or more observations / effects). It provides flexible methods to model a wide range of variance models for random effects or error structures.

    Change the text below to read – The typical applications of ASReml include the analysis of:

    • (un)balanced longitudinal data
    • (un)balanced designed experiments
    • multi-environment trials
    • both univariate and multivariate animal breeding
    • genetics data
    • regular or irregular spatial data
    • repeated measures analysis

    New in v.4

    Introduction of an alternative functional method of associating variance structures with random model terms and the residual, akin to that used in ASReml-R, as an alternative to the former structural method, where the variance models were specified separately from the model terms. Using the functional specification, the variance model for random model terms and the residual error term is specified in the linear mixed model by wrapping terms with the required variance model function. The functional approach leads to a simpler, more concise and less error-prone specification of the linear mixed model, that is more automatic for specifying multi-section residual variances.

    There are also many changes to improve efficiency and convenience. These include:

    • computationally efficient fitting of random regression models when there are more variables than observations motivated by the use of SNP marker data to explain genotypes
    • fitting linear relationships among variance structure parameters
    • automatic generation of initial values for variance parameters
    • generating a template to allow an alternative way of presenting
    • parametric information associated with variance structures
    • new qualifiers !ASSIGN, !FOR and !IF to simplify job flow
    • stabilized updates to improve convergence of factor analytic models
    • enhanced syntax for VPREDICT, allowing specification of functions in terms of names rather numbers
    • calculating information criteria
    • writing out design matrices to external files

    ASReml in industry

    Be in good company. Our software is cited in thousands of papers. Read a few examples.