Help System (web edition)

The equality of several means may be tested by the F-test in one-way
analysis of variance with or without assuming equal group variances.
Both tests are automatically produced. An example of the definitions
in the ANOVA operation:

        ANOVA IEADATA,30
        DEPENDENT=KNOWLDGE   GROUPING=ATTITUDE
        ATTITUDE=1(Best),2(Same),3(Worst)
        METHOD=ONEWAY

Multiple comparisons of means may be performed either by using
comparisonwise (t tests) or experimentwise significance levels.
The names of selected methods are given by the parameter METHOD:

TTESTS  pairwise t-tests both for equal and unequal group
        variances.
TUKEY   Tukey's (equal sample sizes) or Tukey-Kramer's (unequal sample
        sizes) method. Equal population variances assumed. The latter
        method can be used only in pairwise mean comparisons.
T3      T3 method for both equal and unequal group variances. Pairwise
        mean comparisons only.
BF      Brown and Forsythe's method for both equal and unequal group
        variances. Pairwise mean comparisons as well general contrasts
        of means.
SCHEFFE Scheffe's method for equal group variances. Pairwise mean
        comparisons and general contrasts of means.
GABRIEL Gabriel's approximative method for equal group variances.
        Usage as Scheffe's method.
NK      Newman-Keul's procedure, the desired significance level
        must be given as the option, equal sample sizes only,
        only significance tests.
TWELSCH Newman-Keul's or Dunacn's procedure with Tukey-Welsch
        specification, the significance level must be given, only
        significance tests.

If the letter S is given in parenthesis after the name of the method
e.g. TUKEY(S) then the p-values corresp. to the test statistics are
given too. Instead of significance tests you may ask for simultaneous
confidence intervals by giving the wanted significance level in
parenthesis, e.g. TUKEY(0.05). For t-tests (parameter TTESTS) no
confidence intervals may be produced.

If only pairwise mean comparisons are wanted then the form of contrasts
need not be defined. Otherwise the form of contrasts must be specified
by the parameters PAIRWISE, CLEVEL, FGROUP and SGROUP (Not available
for NK and TWELSCH):

CLEVEL   If CLEVEL=i is specified then contrasts of means of the form
         (1,k-1,0), ..., (i,k-i,0) will be tested, where k is the total
         number of group means (k>i). Notation (i,k-i,0) refer to those
         comparisons where the mean of i group means is compared with
         the mean of other k-i means and no group is ignored.
FGROUP   Testing a contrasts specified by FGROUP and SGROUP. With para-
SGROUP   meter FGROUP the group values/labels of the first set of group
         means are given. The second set is specified by SGROUP. Groups
         not specified by the parameters FGROUP and SGROUP are ignored.
PAIRWISE  If both pairwise and more general contrasts are to be tested
         then specify PAIRWISE=1.

Further information:
  1 = More on the definitions of the grouping structure 
  M = Multivariate multiple comparisons 
  A = More on ANOVA operations