Coding primary studies

With the final list of primary studies in hand, coding is the process by which each of them is examined to extract the data needed for the analysis, following a coding protocol that specifies which data to extract and how. Coded data typically falls into three groups: characteristics of the study (e.g., population, design), characteristics of the publication (e.g., year, language), and the data needed to compute an effect size — a study can only enter the statistical analysis if this last category is available.

As with study selection, it's good practice to have two or more researchers code each study independently and report their agreement rate, resolving disagreements through discussion or a third reviewer.

Primary studies often report more than one relevant result — several outcomes, subgroups, or time points in the same sample. When that happens, it's worth coding each one as its own record rather than picking a single number per study: the studies then feed separate, outcome-specific meta-analyses (reported together in the same publication) instead of blending distinct effects into one figure that answers no single question well.

The result of this step is a coded dataset — study and publication characteristics, quality ratings, and the raw ingredients for an effect size — for every primary study, ready for Computing effect sizes.