Meta Analysis

Yes — a meta-analysis is the statistical stage built on a systematic search and screening. Without a reproducible protocol for finding and selecting studies, pooled estimates aren't credible.

Random effects is the default in most social-science meta-analyses because true effects plausibly vary across contexts and populations. Fixed effect assumes one true effect and suits narrow, homogeneous study sets.

Technically two studies can be pooled, but a useful meta-analysis in the social sciences usually needs at least five to ten comparable studies, and subgroup or moderator analysis needs considerably more. With very few studies, estimates of heterogeneity are unreliable and a structured narrative synthesis may be the more honest choice.

I² estimates the percentage of variation across studies that reflects real differences in effects rather than chance. Values around 25%, 50% and 75% are conventionally read as low, moderate and high heterogeneity. High I² is not a reason to abandon the analysis — it is a reason to investigate why effects differ, through subgroup analysis or meta-regression.

The R packages metafor and meta are the most flexible and are free; Comprehensive Meta-Analysis (CMA) is a widely used point-and-click option; JASP and jamovi offer free menus for standard analyses; and RevMan is common for health-science reviews following Cochrane methods. Choose one your supervisor can review, and report the software and version in your methods.