Controlled-Trial Designs

1
Controlled-Trial Designs

• See Batterham and Hopkins, Sportscience 9, 33-39,
  2005.
• Choice determined by availability of subjects,
  reliability of dependent variable, and time to
  wash out treatment.
• Time Series
• Weakest design
• Any change post treatment could be coincidental.
• Analyze using within-subject modeling
• Fit line or curve to each subject's baseline.
• Extrapolate to the post-test.
• Paired t of observed post minus predicted post.

2

• Posts-only Controlled Trial
• Least disturbance of subjects
• Large sample size, but sample size is smaller
  than for the usual pre-post designs for variables
  with sufficiently poor reliability.
• Analyze via unequal-variances unpaired t.
• Can get estimate of individual responses as an
  SD, but not usefully for individuals.

3

• Pre-post Controlled Trial (or fully-controlled
  trial)
• Most common design.
• Analyze difference of changes via
  unequal-variances unpaired t.
• Pre-post Crossover
• Best design if all subjects motivated to know
  effect of treatment on themselves.
• Analyze change of changes via paired t.

4

• Posts-only Crossover
• Least number of subjects and tests when
  reliability high.
• But can get estimate of individual responses only
  if include a repeat of one of the treatments for
  each subject.
• Analyze via paired t.
• Important to control for the order effect, if it
  is substantial and there are unequal numbers in
  the crossover groups.
• A spreadsheet is on its way.

5
Decision Tree
6
Can you use a control group
or a separate control treatment?
NO
YES
Time-series
trial
n10-???
7
a
Is the measure
reliable
over the intervention period?
NO
YES
Posts
-
only
trial
n2x 150
8
Will the intervention wash out in
an acceptable time for a crossover?
NO
YES
Fully
-
controlled
trial
n2x 10-300
-
9
b
Is the measure
reliable
over
washoutintervention
period?
NO
YES
Are you limited
by subjects
or resources?
NO
YES
Fully
-
controlled
Simple
crossover
crossover
n10-300
n10-75