Recently, John Hattie put out what amounts to a correction paper on his meta-analysis of feedback with B Wisniewski, and K Zierer. Previously, Hattie had placed the effect size of feedback as .70. However, as is admitted in the 2020 paper, Hattie and his original co-author accidentally included some outlier data twice, when he included both the original outlier studies and meta-analyses that included this outlier data within his own secondary meta-analysis. For whatever reason, there does appear to be a significant amount of outlier data within this specific topic of education. While most studies appear to put feedback with an ES of .40 - .60, we have several studies with both negative effect sizes and extreme effect sizes higher than 2.0. Of course, this is precisely why meta-analysis is such an imperative tool to use within the context of education and why we cannot rely on just individual studies.
Overall, in this new meta-analysis, we saw that feedback had a generalized effect cohen’s d size of .55. However, if we remove outliers, we have an effect size of .48. Giving feedback a moderate effect size overall. The authors also noted that feedback with studies with control groups had an ES of .42, and feedback studies that used pre and post-test data had an ES of .63. I have previously discussed this on my podcast, but I do not think pre and post-test data studies are valid for education, as they are so affected by time. Ultimately, we should want education studies to be always using control groups. However, as most meta-analyses in education are not currently controlling for this, I, unfortunately, think we cannot exclude this data when making comparisons between education factors and their effect sizes.
In general, it appears from this new analysis that feedback should be considered a moderate yield teaching factor and not a high yield one. That being said, their meta-analysis also showed that different types of feedback were more and less effective; moreover, feedback was more and less effective for different types of learning. Please see the graph below for these results.