The article revisits the concept of interaction in statistics, focusing on comparing two estimates of the same quantity derived from separate analyses. The authors discuss the general method for comparing means or proportions, emphasizing the importance of independence and large sample sizes. They provide a formula for the standard error of the difference between two estimates and explain how to test the null hypothesis using a z-test and confidence intervals. The article illustrates these concepts with examples, such as comparing relative risks or odds ratios in meta-analyses. It highlights the complexity of comparing estimates on the log scale and provides a step-by-step guide to performing these comparisons. The example of a meta-analysis on hormone replacement therapy and non-vertebral fractures demonstrates how to test for interaction and estimate the ratio of relative risks. The article concludes by noting the limited power to detect interactions and the importance of targeted statistical analysis rather than comparing P values from separate analyses.The article revisits the concept of interaction in statistics, focusing on comparing two estimates of the same quantity derived from separate analyses. The authors discuss the general method for comparing means or proportions, emphasizing the importance of independence and large sample sizes. They provide a formula for the standard error of the difference between two estimates and explain how to test the null hypothesis using a z-test and confidence intervals. The article illustrates these concepts with examples, such as comparing relative risks or odds ratios in meta-analyses. It highlights the complexity of comparing estimates on the log scale and provides a step-by-step guide to performing these comparisons. The example of a meta-analysis on hormone replacement therapy and non-vertebral fractures demonstrates how to test for interaction and estimate the ratio of relative risks. The article concludes by noting the limited power to detect interactions and the importance of targeted statistical analysis rather than comparing P values from separate analyses.