**2.5 Data analysis**

Results were analyzed using the comprehensive meta-analysis software, version 3. Since the studies included in this review varied widely, a random effects model was used. Many of the studies had small sample sizes, in particular polysomnographic studies. Hedges' g was consequently chosen for effect size because of its correction for small samples [32]. Q and I2 were used to assess heterogeneity. For each analysis of significant results including three studies or more, Orwin's fail-safe N was used to investigate publication bias. In these cases, a funnel plot was visually examined.
