COMMON MISTEAKS
MISTAKES IN
USING STATISTICS: Spotting and Avoiding Them
Example of How Low Power and Publication Bias Interact
The plot below shows effect size vs sample size for studies in a
research summary of the effect of supplements on osteoarthritis.
Note the greater variability of effect size and tendency toward
larger effect sizes for samples with lower sample size (hence lower power). Greater variability of effect
sizes is expected with low power; publication
bias then tends to result in mainly those studies with higher
effect size being published.
Note: Most of these
studies also had other mistakes in design, implementation, or analysis.
In particular, only one (the one with sample size 104 and effect size
0.61) used intent-to-treat analysis. None
of the studies were considered to have adequate blinding.
Last updated May 12, 2011
