meta-analysis

Meta-analysis pools effect sizes across studies using inverse-variance weighting to produce a summary effect with 95% confidence interval. Convert study results to common metrics like Cohen's d or odds ratio, calculate standard errors, select a model (fixed or random-effects), and report the pooled estimate along with heterogeneity metrics (I², Q).

Fixed-effect models assume all studies estimate the same true effect and weight by inverse variance alone. Random-effects models account for between-study heterogeneity by adding a variance component, producing wider confidence intervals and more conservative estimates when studies vary substantially.

Publication bias occurs when small, non-significant studies go unpublished. Evaluate it using funnel plots (scatter of effect size vs. precision) and Egger's test (regression asymmetry test). Asymmetry suggests bias; trim-and-fill methods can estimate the adjusted pooled effect after accounting for missing studies.

Yes. Meta-analysis synthesizes evidence across clinical trials, behavioral studies, and social science research by converting diverse outcomes to common effect size metrics. Standardize results (e.g., d, odds ratio, risk ratio) with standard errors, then pool within homogeneous subgroups or explore sources of variation via meta-regression.

Heterogeneity is variation in effect sizes across studies beyond chance. Quantify it with I² (percentage of variance due to heterogeneity) and Q statistic (chi-square test). High heterogeneity signals true differences between studies; use random-effects models and subgroup or meta-regression analysis to explore sources.

Extract effect sizes and standard errors from each study: means and standard deviations for standardized differences, or event counts and sample sizes for odds and risk ratios. Organize as study-level data with outcomes, sample sizes, and study characteristics needed for subgroup and meta-regression analyses.