Breaks a model term down into its basic contrasts (R.W. Payne).
Options
TERM = formula |
Model term to split into basic contrasts |
|---|---|
PSEUDOFACTORS = pointer |
Pseudo-factors representing the basic contrasts |
NEWTERMS = formula structure |
Model formula containing the term followed by the pseudofactors |
No parameters
Description
It is well known that the interaction between factors F1, F2 … Fm (each with prime p numbers of levels) can be partitioned into (p-1)**(m-1) orthogonal sets of contrasts, known as basic contrasts (e.g. Kempthorne 1952, page 321). They are usually written as
F1**k1 F2**k2 … Fn**kn
where conventionally k1=1 and 1≤ki<p. Each set of basic contrasts represents comparisons between p sets of factor combinations, the ith of which contains the factor combinations f1 … fm such that
k1 × f1 + k2 × f2 + … km × fm = i mod(p)
In most straightforward experimental designs, all the contrasts of each treatment interaction are estimated in a single stratum. When this is not feasible, a popular strategy is form the design so that different sets of basic contrasts are estimated in different strata. This underlies the design key method, which is used by the GENERATE directive and the AKEY procedure.
When a design key has been used to generate the design, the FPSEUDOFACTORS directive may be used to form the pseudo-factors required to cope with interactions (or other model terms) whose contrasts are estimated in different strata. If this pseudo-factoring is not included, the ANOVA directive will warn that the design contains partial confounding.
When no design key is available, provided all the treatment terms are main effects or interactions all of whose factors have the same prime number of levels, an alternative strategy is to use the FBASICCONTRASTS procedure to break up each partially-confounded interaction into its sets of basic contrasts. The interaction is specified using the TERM option. The PSEUDOFACTORS option saves a pointer containing the factors generated to represent the basic contrasts. Finally, the NEWTERMS option can save a new model formula containing the interaction followed by the pseudo-factor operator // and then the list of pseudo-factors. For example, for the interaction of two 3-level factors A and B, the NEWTERMS formula would be
A.B // (Pf[1,2])
where Pf[] is the pointer of pseudo-factors.
Options: TERM, PSEUDOFACTORS, NEWTERMS.
Parameters: none.
Method
FBASICCONTRASTS uses the PRIMEPOWER procedure to check that all factors have a prime number of levels. It then constructs a design key with a row for every set of basic contrasts, and forms the pseudo-factors using GENERATE.
See also
Procedures: AFUNITS, FACCOMBINATIONS, FACDIVIDE, FACPRODUCT.
Commands for: Calculations and manipulation, Design of experiments.
Example
CAPTION 'FBASICCONTRASTS example',\
!t('3**3 design with part of the 3-factor interaction',\
'confounded with blocks.'); STYLE=meta,plain
FACTOR [NVALUES=81; LEVELS=3] Blocks,N,P,K
READ Blocks
1 1 1 2 2 2 3 3 3 2 2 2 3 3 3 1 1 1 3 3 3 1 1 1 2 2 2
2 2 2 3 3 3 1 1 1 3 3 3 1 1 1 2 2 2 1 1 1 2 2 2 3 3 3
3 3 3 1 1 1 2 2 2 1 1 1 2 2 2 3 3 3 2 2 2 3 3 3 1 1 1 :
GENERATE N,P,K,3
BLOCKSTRUCTURE Blocks
TREATMENTS N*P*K
ANOVA
" N.P.K is partially confounded with Blocks: split this up into
its basic contrasts, and include these in the model formula."
FBASICCONTRASTS [TERM=N.P.K; PSEUDOFACTORS=Pf; NEWTERMS=Pseudoterms]
PRINT Pseudoterms
PRINT [RLWIDTH=8; WIDTH=62; ORIENTATION=across] N,P,K,Pf[]; FIELD=2
" The basic contrasts account for all of N.P.K,
so N.P.K itself will be aliased."
TREATMENTS N*P*K - N.P.K + #Pseudoterms
ANOVA