Skip to main content Skip to navigation

Event Diary

Show all calendar items

CRiSM Seminar - Freedom Gumedze (University of Cape Town)

- Export as iCalendar
Location: A1.01
Freedom Gumedze (University of Cape Town)
 
An alternative approach to outliers in meta-analysis
 
Meta-analysis involves the combining of estimates from independent studies on some treatment in order to get an estimate across studies. However, outliers often occur even under the random effects model. The presence of such outliers could alter the conclusions in a meta-analysis. This paper proposes a methodology that detects and accommodates outliers in a meta-analysis rather than remove them to achieve homogeneity. An outlier is taken as an observation (study result) with inflated random effect variance, with the status of the ith observation as an outlier indicated by the size of the associated shift in the variance. We use the likelihood ratio test statistic as an objective measure for determining whether the ith observation has inflated variance and is therefore an outlier. A parametric bootstrap procedure is proposed to obtain the sampling distribution for the likelihood ratio test and to account for multiple testing. We illustrate the methodology and its usefulness using three meta-analysis data sets from the Cochrane Collaboration.

Show all calendar items