fac.aov
Model
fac.aov
model. In the special case of reduced models of an
orthogonal saturated model, this retains the saturated effects and
corresponding error estimates.
update.fac.aov(object, formula.new, ...)
fac.aov
object.
update.default
.
fac.aov
object, fitting the formula given.
In the case where
object
is a saturated model
(
object$df.residual == 0
) from an orthogonal design matrix, and
formula
specifies a true
submodel of
object
,
update
keeps the effects from the full model
of
object
, along with estimates of
pse
,
tse
and
ase
. Thus, the
reduced model retains estimates from the full model for use in
functions that examine the model, e.g.,
pareto
,
summary
. However
by default, the
mse
of the reduced model is used for significance
tests, unless otherwise specified.
In all other cases, the value returned by
update
is the same as that
returned by
fac.aov
.
In the reduced model case, the values returned by
fac.aov
and
update
, for the same fitted model will have different values for
pse
,
tse
,
ase
and
seffects
. Estimates in both cases are valid,
however those from the
update
function are based on the effects of
the saturated model and are more easily interpretable.
buffer.fac <- fac.aov(buffer.df) summary(buffer.fac) pareto(buffer.fac) # now compare the results of update and fac.aov # for fitting a reduced model buffer.rmod <- update(buffer.fac, ~ pH*thimer + pH*gent) summary(buffer.rmod) pareto(buffer.rmod) buffer.fac2 <- fac.aov(rate~pH*thimer+pH*gent,buffer.df) summary(buffer.fac2) pareto(buffer.fac2) # look at diagnostic plots for the reduced model plot(buffer.rmod) # refit using a transformation of the response buffer.sqrt <- update(buffer.fac, sqrt(rate) ~ .)