Learning by invention is an alternative approach to teaching statistics where students are tasked with attempting to solve a problem before being taught the canonical formula for solving it, often resulting in increased understanding of material compared with traditional instruction. The first study, conducted in a college statistics classroom with mixed-skill groups, extended previous work showing an advantage for learning-by-invention activities compared with a lecture-first control. A second study explored group interactions that led to benefits from learning-by-invention activities. Successful groups were more likely to propose solutions and less likely to propose arbitrary formulas or highlight group members' math skills. The features of small group interactions that may help or harm learning during invention activities are discussed.