Meta-analysis is frequently read from the diamond down. The forest plot’s tidy alignment gives the illusion of certainty, with the pooled diamond suggesting a single definitive answer. Yet the forest is rarely uniform: some trunks lean, others twist, and a few tower or collapse, reshaping the skyline. This metaphor illustrates heterogeneity—the unevenness between studies—that ultimately determines the reliability of pooled estimates. This tutorial recenters interpretation on that variability: Q signals its existence, I² describes the proportion beyond chance, and τ² quantifies its magnitude. At the same time, prediction intervals extend these measures into practice by showing the range that future studies may realistically occupy. In diagnostic test accuracy, hierarchical models such as Reitsma’s bivariate and HSROC are highlighted, as they preserve the correlation between sensitivity and specificity and capture threshold-driven heterogeneity. Beyond numerical measures, visual and analytical approaches provide complementary insights into the underlying sources of heterogeneity, helping to explain why studies diverge in their findings. From these tools emerge practical lessons: the need for transparent reporting, robust estimators, prediction intervals, and caution in interpreting subgroup claims, while routine pitfalls—such as defaulting to DerSimonian–Laird, selecting the model solely based on a heterogeneity statistic, or reporting I² in isolation—are avoided. The message is simple: the diamond is not the compass—meta-analysis earns credibility not by multiplying averages, but by explaining the uneven forest behind them.
Javier Arredondo Montero (Tue,) studied this question.