ASCREEN procedure

Performs screening tests for designs with orthogonal block structure (R.W. Payne).


Options

PRINT = strings
Which tests to print (conditional, marginal); default cond, marg

FACTORIAL = scalar
Limit on the number of factors in each treatment term; default 3

EXCLUDEHIGHER = string
Whether to exclude higher-order interactions in the initial model for the conditional test of each term (yes, no); default no

FORCED = formula
Terms that must be included (together with any covariates) in the initial models for every term; default * i.e. none


Parameter

Y = variates
Variates to be analysed


Description

ASCREEN can be used to assess the treatment terms in an analysis of variance when the design is unbalanced but its error terms that are all orthogonal to one another. This includes any design with a hierarchical block structure, for example

Blocks / Plots

or

Replicates / Wholeplots / Subplots

ASCREEN thus provides a way of testing treatment terms in designs that cannot be analysed by ANOVA. Once ASCREEN has been used to decided which terms need to be included in the treatment model, the treatment effects and means can be estimated using REML.

   Before using ASCREEN, the block and treatment models for the design must be defined by the BLOCKSTRUCTURE and TREATMENTSTRUCTURE directives, in exactly the same way as for an analysis by ANOVA. As in ANOVA, the FACTORIAL option sets a limit on number of factors in each treatment term (default 3). You can also define covariates using the COVARIATE directive. The y-variate is specified by the Y parameter of ASCREEN.

   ASCREEN forms marginal and conditional tests for the treatment terms similar to those produced by the RSCREEN procedure. These are produced for the analysis of each stratum of the design (i.e. for the variation associated with each error term). The PRINT option has settings conditional and marginal to control which tests are produced if there is more than one error term; by default both are printed. However, if there is only one error term, ASCREEN uses procedure RSCREEN, which always prints both.

   In a marginal test, each term is assessed by adding it to the simplest possible model. So, with a treatment model of

A + B + C + D + A.B + A.C + A.D + B.C + C.D + A.B.C + A.B.D + A.C.D + B.C.D + A.B.C.D

the main effect of A is added it to the null model, while the interaction term A.B is added to a model containing only the main effects of A and B.

   In a conditional test, each term is added to the most complex possible model. So the main effect A is added to an initial model excluding any term that has A as one of its margins. A is a margin of any term that contains A as one of its factors. So the terms to exclude for A are A.B, A.C, A.D, A.B.C, A.B.D, A.C.D and A.B.C.D. Similarly the interaction A.B is added to a model excluding any term that has A.B as a margin; i.e. any term that contains A and B amongst its factors. So A.B.C, A.B.D and A.B.C.D are excluded with A.B. The other terms to be included in the initial model depend on the setting of the EXCLUDEHIGHER option. With the default setting of no, all other terms are included in the initial model. So, the initial model for A would be

B + C + D + B.C + C.D + B.C.D

Alternatively, if EXCLUDEHIGHER=yes, the initial model contains only terms with no more factors than the term being tested. So, the initial model for A would be

B + C + D

   The FORCED option allows you to specify a model formula with terms that must be included in the initial model for the conditional and marginal tests of every treatment term. The forced model automatically includes any covariates.

 

Options: PRINT, FACTORIAL, EXCLUDEHIGHER, FORCED.

Parameter: Y.


Method

ASCREEN uses RSCREEN if there is only one error term. Otherwise, it first uses ANOVA to check that the design has orthogonal block structure. Then, if so, it calculates the relevant sums of squares by regression with matrices of weights calculated using FPROJECTIONMATRIX. The weight matrix for each stratum is its projection matrix; for further details see Payne & Tobias (1992).


Action with RESTRICT

ASCREEN takes account of any restrictions on the y-variate.


Reference

Payne, R.W. & Tobias, R.D. (1992). General balance, combination of information and the analysis of covariance. Scandinavian Journal of Statistics, 19, 3-23.