colleagues’60 levels of evidence to weight the studiesincluded in the meta-analysis. If a study had a truecontrol group and reported means, mean changes, andstandard deviations or reported a focused F-statistic ort-test statistic, it received a weighting of 4, the highestscore. If a study used a quasi-control group and reportedmeans, mean changes, and standard deviations orreported a focused F-statistic or t-test statistic, it receiveda weighting of 3. If the study had no control group butreported means, mean changes, and standard deviationsor reported a focused F-statistic or t-test statistic, itreceived a weighting of 2. Finally, if the study onlyreported a P value, it received a weighting of 1 regardlessof the study design used. One study47 had such a smallsample size (n3) that it also was weighted with a valueof 1. For this meta-analysis, 17 studies measuring habitualgait speed were determined to be high quality andweighted with a value of 3 or 4 (Tab. 2). For fast gaitspeed, 12 studies had a weighting of 3 or 4 (Tab. 3).Blocking. When there was a significant effect of therapeuticexercise intervention on gait speed, further analyseswere done to identify possible underlying factors (ie,exercise intensity and dosage) that may have contributedto this finding.61 As previously described, exercise interventionwas blocked into combination or strength training;high, moderate, or low exercise intensity, andfinally, high or low exercise dosage.Data AnalysisThe data for habitual and fast gait speeds were analyzedseparately. Because meta-analysis requires only a teststatistic, some of the reported studies did not provideactual gait speed data, which accounts for some of theempty cells in Tables 2 and 3. Several studies includedsufficient description of their methods to support inclusionof the study in the meta-analysis, although they didnot include specific data on gait speeds of the includedgroups. An attempt was made to contact the authors forspecific data, but no author responded. For these studies,the weighting was lowered to reflect the omission ofdata, although the test statistic and sample size reportedwere sufficient to calculate an effect size for the study.Thus, we were able to include them in the metaanalysis.61 A summary of all relevant statistical datapertaining to each set of analyses is presented in Table 5.Because effect size, measured as Pearson correlationcoefficient (r), is the variable of interest in this metaanalysisand is typically not reported in research articles,we had to calculate the effect size for each study fromavailable data. In addition, standardized r values (FisherZ [Zr]) and 95% confidence intervals (CIs) around theeffect size (r) were calculated for each analysis.62,63