Content area
Abstract
The mean odds ratio was 0.96 with a 95% confidence interval of 0.78–1.18. [...]there is no evidence of a relationship on average. On that basis, most readers would assume that the distribution corresponds to panel A or B. That would be a mistake since the distribution actually corresponds to panel C (as indicated by the check-mark). [...]in either analysis, I2 did not provide useful information about the dispersion in effects. Specifically, the variance of the observed effects is equal to the variance of true effects plus the additional variance due to sampling error. [...]the distribution of observed effects tends to be wider than the distribution of true effects [5]. [...]a small proportion of a large number can still be a large number.4 The prediction interval If the I2 statistic does not tell us how much the effect size varies, what statistic does provide that information?