Scientists studied replication studies to see if they could predict the outcomes. They used a new model that considers inflated results and differences in effects between original and replication studies. The model did well in predicting outcomes in two out of four studies. The results suggest that many original study results were inflated, possibly due to bias. They also found that non-significant replication results can still be valid, even if the original study had significant results. This shows that using statistical significance alone to judge replication success may not be reliable. The statistical methods and data are available in the ReplicationSuccess R package.