What can be learned from replication in the social sciences? While practitioners of the credibility revolution advocate replication -- conducting a new study based upon an existing study -- as a means to probe the robustness and generalizability of findings, there is limited conceptual clarity about what constitutes a replication or what can be learned from comparing the results of these studies. Most replication studies in political science are "conceptual replications," rather than "direct replications." We argue that greater specificity is needed to understand whether conceptual replications provide any information about robustness or generalizability. Specifically, we employ a theoretical implications of empirical models approach (TIEM) to consider the implications of changes in setting (or population), instruments (treatment conditions), and measurement strategies for what can be learned from the comparison of results of a study and a replication. Our results provide new guidance for the application and interpretation of replication studies in empirical social science.