EMMEANS Subcommand (GLM: Univariate command)
EMMEANS
displays
estimated marginal means of the dependent variable in the cells (with
covariates held at their overall mean value) and their standard errors
of the means for the specified factors. These means are predicted,
not observed, means. The estimated marginal means are calculated by
using a modified definition by Searle, Speed, and Milliken (1980).
-
TABLES
, followed by an option in parentheses, is required.COMPARE
is optional; if specified,COMPARE
must followTABLES
. - Multiple
EMMEANS
subcommands are allowed. Each subcommand is treated independently. - If identical
EMMEANS
subcommands are specified, only the last identical subcommand is in effect.EMMEANS
subcommands that are redundant but not identical (for example, crossed factor combinations such as A*B and B*A) are all processed.
TABLES(option). Table specification. Valid options are the
keyword OVERALL
, factors appearing
on the factor list, and crossed factors that are constructed of factors
on the factor list. Crossed factors can be specified by using an asterisk
(*) or the keyword BY
. All factors
in a crossed factor specification must be unique.
- If
OVERALL
is specified, the estimated marginal means of the dependent variable are displayed, collapsing over between-subjects factors. - If a between-subjects factor, or a crossing of between-subjects
factors, is specified on the
TABLES
keyword,GLM
collapses over any other between-subjects factors before computing the estimated marginal means for the dependent variable. For a multivariate model,GLM
collapses over any other between-subjects or within-subjects factors. - The
TABLES
keyword may be followed by theWITH
keyword and a parenthesized list of covariates and values. Tables are produced with the values of the covariates fixed at the specified values. The keywordMEAN
causes the covariate to be fixed at its observed mean value.
COMPARE(factor) ADJ(method). Main-effects or simple-main-effects omnibus
tests and pairwise comparisons of the dependent variable. This option gives the mean difference, standard error, significance,
and confidence interval for each pair of levels for the effect that
is specified in the TABLES
command,
as well as an omnibus test for that effect. If only one factor is
specified on TABLES
, COMPARE
can be specified by itself; otherwise,
the factor specification is required. In this case, levels of the
specified factor are compared with each other for each level of the
other factors in the interaction.
- The optional
ADJ
keyword allows you to apply an adjustment to the confidence intervals and significance values to account for multiple comparisons. Available methods areLSD
(no adjustment),BONFERRONI
, orSIDAK
. - If
OVERALL
is specified onTABLES
,COMPARE
is invalid.
Example
GLM DEP BY A B
/EMMEANS = TABLES(A*B) WITH(X=1.0) COMPARE(A)
/DESIGN.
- The output of this analysis includes a pairwise comparisons table for the dependent variable DEP.
- Assume that A has three levels and B has two levels. The first level of A is compared with the second and third levels, the second level is compared with the first and third levels, and the third level is compared with the first and second levels, with the value of the covariate X fixed at 1.0. The pairwise comparison is repeated for the two levels of B.