Cautions in Interpreting Variance Explained (3 of 3)
Next section: Alternative approaches
In most cases, there is no absolute standard to go by, and there is nothing
much one can do other than measure effect size relative to subject variability.
However, it is important not to overlook instances in which effect size
can be measured more directly. For example,
Tullis
(1981) found that improving the format in which the results of a computerized
test of a phone line are displayed reduces the average response time
by about two seconds. although this effect does not explain much variance,
it is important since computerized testing is done literally millions
of times a year. The cumulative time savings of using the better display
is gigantic.
Finally, the
sampling distribution of
measures of effect size complex, and
confidence
intervals on these measures are rarely computed (although procedures are
available, see
Steiger, 2004). It is difficult to interpret an estimate of an effect size
without knowing the range of plausible effect sizes. When small sample
sizes are used, it is possible for the estimate of effect size to be
very different from the actual effect size. Therefore, it is easy to
make the mistake of thinking a small effect is actually a large effect.
Next section: Alternative approaches