This
site is under construction. Please check back every few weeks for
updates
COMMON MISTEAKS
MISTAKES IN
USING STATISTICS: Spotting and Avoiding Them
Comparisons
of treatments applied to people, animals, etc.
(Intent to Treat; Comparisons
with Drop-Outs)
In many forms of comparison of two treatments involving human
subjects, there are subjects who do not complete the treatment. They
may die, move away, encounter life circumstances which take priority,
or just decide for whatever reason to drop out of the study or not do
all that they are asked. It is tempting to just analyze the data for
those completing the protocol, essentially ignoring the drop-outs. This is usually a
serious mistake, for two reasons:
1. In a good study, subjects should be randomized to treatment. Just
analyzing data for those who complete the protocol destroys the
randomization, so that model assumptions are not satisfied. To preserve
the randomization, outcomes for all subjects assigned to each group
need to be compared. This is called intent-to-treat
(or intention-to-treat, or ITT) analysis.
2. Intent-to-treat analysis is usually more informative for consumers
of the research. For example, in studying two drug treatments, dropouts
for reasons not related to the treatment can be expected to be, on
average, roughly the same for both groups. But if one drug has serious
side-effects that prompt patients to discontinue use, that would show
up in the drop-out rate, and be important information in deciding which
drug to use or recommend.
Reason 1 (and sometimes also reason 2) also applies when treatments are
applied to animals, plants, or even objects.
For more information on intent-to-treat analysis, see the following:
Freedman, DA (2005) Statistical Models: Theory and Practice,
pp.
5, 15
D.A. Freedman. “Statistical models for causation: What
inferential leverage do they provide?” Evaluation Review vol. 30
(2006) pp. 691–713. Preprint
van Belle (2008) Statistical
Rules of Thumb, p. 156 - 157