We focus
on regression analyses where multiple predictors are considered,
also called "multivariable analysis." Such an analysis is in
contrast to a univariable (or "simple") analysis, where single
predictor variables are considered. A multivariable regression
analysis provides predictions based on the combined predictive
effect of predictors.
In epidemiology,
the correlation between covariables is often referred to as
"confounding", i.e. that the "true" effect of an exposure covariable
is mixed up with that of one or more covariables when a univariable
(or "crude") analysis is performed (Kleinbaum
et al., 1982). In principle, confounding may be removed
by a multivariable
(or "adjusted") analysis.
QUESTION
4.2
In medicine,
predictor variables usually have positive correlations (when
higher values of the predictor indicate higher risk). Consider
the situation of having estimated a univariable regression coefficient
for a sign of atherosclerosis. In a multivariable analysis,
we include other signs of atherosclerosis as well. What can
be expected for the multivariable regression coefficient when
compared to the univariable regression coefficient?
QUESTION
4.3
In medicine,
the term "multivariate analysis" is often used when one is referring
to a multivariable analysis. "Multivariate," however, implies
a statistical analysis with multiple outcomes (in contrast to
multiple predictors, which should be labeled "multivariable").
Which one of the following analyses is not an example
of a true multivariate analysis?