Understanding who shares academic publications on Twitter is critical to interpretingaltmetrics as signals of scholarly or societal impact. Prior studies have used diverse and oftenincompatible user classification schemes, making synthesis difficult. This study presents asystematic review and meta-analysis of 23 empirical studies (covering 79,014 Twitter users,over 20 million tweets, and more than 5 million tweeted publications) to estimate category-specific engagement across three metrics: user counts, tweets, and tweeted publications. Wedeveloped a harmonized categorization scheme encompassing 11 user types and applied bothrandom effects models (REM) and beta-binomial hierarchical models (BBHM) to estimateproportions, account for study-level variation, and model uncertainty. Across all indicators,individual users were the most active, comprising 66% of users, 55% of tweets, and 50% oftweeted publications. BBHM further enabled in-category vs. out-of-category comparisons andrevealed engagement differences not detected by REM.t-tests on study-level means confirmedsignificant differences between academic individuals and other user types. Despitemethodological heterogeneity, results consistently show that academic and nonacademicindividuals statistically equally dominate Twitter engagement with scholarly content. Ourfindings support the need for standardized user classification schemes and demonstrate thevalue of Bayesian modeling for synthesizing altmetric data in study variation and sparsity.