On Tuesday 13^th April, 2021, the Western Australian branch was pleased to have Professor Luke Prendergast to present at the monthly branch meeting. The title of his talk was “Meta-analysis: common traps and misconceptions”.

Luke is Associate Head of School, School of Engineering and Mathematical Sciences and Director of the Statistics Consulting Platform at La Trobe University. He has worked on Meta-Analysis for the past 6 years in his career. He gave a talk on certain flaws and misconceptions in meta-analysis and provided recommendations on how these flaws could be overcome using examples from published research. He mentioned that some of the flaws may be due to researchers lacking adequate training in meta-analysis.

Using the published examples, Luke identified problems and confusion from authors on whether to use random effects or fixed effects models in meta-analysis. He proposed that random effects models would be appropriate to use, which assumes heterogeneity between the studies. In addition, analysts should avoid allocating very large weights to a small number studies which may bias the results of the analysis and which may happen with fixed effect are used. Another problem relates to the interpretation of the meta-analysis results. Simple tests may lack power, especially when the number of studies is small. Authors should make it clear that the tests are for a mean effect and not effects overall. Prediction intervals which take into account the variance in the estimators and the variance associated with heterogeneity are one good example of how the magnitude of heterogeneity can be assessed. If there are wide prediction and confidence intervals in random effects models, this is to account for the heterogeneity among the studies.

The key message was that “Heterogeneity in studies can be a good thing”. Heterogeneity may be very interesting and may lead to new research directions. He further responded to questions from the audience and recommended that researchers should look for additional information on studies included in the meta-analysis to explain the analysis and conduct sensitivity analysis as appropriate.

Fadzai Chikwava