Several studies have found that social science experiments too frequently fail to replicate. We discuss a reanalysis of two widely-cited replication studies [Open Science Collaboration, Science, 349(6251), 1–8 (2015) & Camerer et al., Science, 351(6280), 1433–1436 (2016)] that showed theory-driven experiments replicated, but Fisher-designed experiments did not. More generally, we explain why replications succeed when theory designs tests. Consequently, hybrid experiments that blend theory and Fisher design elements present one possible avenue to improve the replicability of test results.