- Statistics Without Statistics: Did They Use the Right Test? (continued)
ANOVA (Analysis of Variance)
An ANOVA table is a "super t-test". An ANOVA can test
several means of several categories
at once. Commonly this test can be seen with stratified data in several groups. Imagine a
categorical variable with 4 values (group I, II, III, IV) each with the continuous variable SBP. The
ANOVA can test the differences in the 4 categories and look for significance among the means. Perhaps
groups I and IV are significantly different but I and III are not. ANOVA's are commonly seen in the larger
studies or meta analyses because of their ability to look at several means at once.
Correlation
Another very common statistical test is the correlation. Several types exist but the Pearson correlation
is usually used.
This test involves two continuous variables plotted against each other two
generate a new line. The desire is to find some type of association between the two variables
in order to build an argument for one variable changing with another variable. The associations are
positive or negative or none and range from a perfect -1 to a perfect +1. Strength of association is
described by the
r-value. Note the slope itself is not significant other than positive or
negative.
It is inappropriate to use the actual line to describe the effect of one continuous variable on
another. Correlations find associations, regressions find equations.
Regression
Regression analysis is one of the fundamental pillars of statistics. The actual math of a regression can
be quite tedious but the general idea is fairly easy to understand. A regression also looks for significant
differences, but also looks at the actual interplay between variables. With regressions authors can build
strong arguments that variables interact or "explain" each other. (The
F-value
is the usual measure of strength of a model.) The principle is based on the equation y = mx + b.
