Overview (UNIANOVA command)
This section describes the use of UNIANOVA
for univariate analyses. The UNIANOVA
procedure provides regression analysis
and analysis of variance for one dependent variable by one or more
factors and/or variables.
Options
Design Specification. You can specify
which terms to include in the design on the DESIGN
subcommand. This allows you to estimate a model
other than the default full factorial model, incorporate factor-by-covariate
interactions or covariate-by-covariate interactions, and indicate
nesting of effects.
Contrast Types. You can specify contrasts
other than the default deviation contrasts on the CONTRAST
subcommand.
Optional Output. You can choose from a wide variety of optional output on the
PRINT
subcommand. Output appropriate to univariate designs includes descriptive
statistics for each cell, parameter estimates, Levene’s test for equality of variance across cells,
tests for heteroskedasticity, partial eta-squared for each effect and each parameter estimate, the
general estimable function(s) matrix, and a contrast coefficients table (L' matrix). The
OUTFILE
subcommand allows you to write out the covariance or correlation matrix,
the design matrix, or the statistics from the between-subjects ANOVA table into a separate data
file.
Using the EMMEANS
subcommand, you can request tables of estimated marginal means of
the dependent variable and their standard deviations. The SAVE
subcommand allows you to save predicted
values and residuals in weighted or unweighted and standardized or
unstandardized forms. You can specify different means comparison tests
for comparing all possible pairs of cell means using the POSTHOC
subcommand. In addition, you can
specify your own hypothesis tests by specifying an L matrix and a K matrix to test the univariate hypothesis LB = K.
You can display robust or heteroskedasticity-consistent (HC) standard errors
by using the ROBUST
subcommand, and you can write the robust covariance matrix
estimates to a new file or dataset. You can also display a second set of output for custom
hypothesis tests, with results based on the specified robust covariance matrix estimator.
Basic Specification
- The basic specification is a variable list identifying the dependent variable, the factors (if any), and the covariates (if any).
- By default,
UNIANOVA
uses a model that includes the intercept term, the covariate (if any), and the full factorial model, which includes all main effects and all possible interactions among factors. The intercept term is excluded if it is excluded in the model by specifying the keywordEXCLUDE
on theINTERCEPT
subcommand. Sums of squares are calculated and hypothesis tests are performed using type-specific estimable functions. Parameters are estimated using the normal equation and a generalized inverse of the SSCP matrix.
Subcommand Order
- The variable list must be specified first.
- Subcommands can be used in any order.
Syntax Rules
- For many analyses,
the
UNIANOVA
variable list and theDESIGN
subcommand are the only specifications needed. - If you do not enter
a
DESIGN
subcommand,UNIANOVA
will use a full factorial model, with main effects of covariates, if any. - At least
one dependent variable must be specified, and at least one of the
following must be specified:
INTERCEPT
, a factor, or a covariate. The design contains the intercept by default. - If more than one
DESIGN
subcommand is specified, only the last one is in effect. - Dependent variables and covariates must be numeric, but factors can be numeric or string variables.
- If more than one
MISSING
subcommand is specified, only the last one is in effect. - If more than one
ROBUST
subcommand is specified, only the last one is in effect. - The following words are
reserved as keywords or internal commands in the
UNIANOVA
procedure:
INTERCEPT
, BY
, WITH
, ALL
, OVERALL
, WITHIN
Variable names
that duplicate these words should be changed before you run UNIANOVA
.
Limitations
- Any number of factors can be specified, but if the number of factors plus the number of split variables exceeds 18, the Descriptive Statistics table is not printed even when you request it.
- Memory requirements depend primarily on the number of cells in the design. For the default full factorial model, this equals the product of the number of levels or categories in each factor.