5th graders (n=527) were randomly assigned to self-explain worked examples (WE) or solve problems during a yearlong study. WE students learned more algebra skills when given algebra worked examples on an end-of-year transfer measure. Comparing attempts, conceptual, and mechanical correctness by condition, results demonstrated that the condition difference in accuracy was attributable to better understanding of conceptual ideas in the WE condition, not to greater willingness to try or fewer mechanical errors. [This paper was published in: J. Kay and R. Luckin (Eds.) "Rethinking learning in the digital age: Making the Learning Sciences count: The International Conference of the Learning Sciences (ICLS)," London, UK.]