The costs and benefits of providing feedback during learning

damental goal—is maximizing long-term retention and transfer. Toward achieving that goal, one key manipulation is to induce the retrieval of to-be-learned information, which has been shown in many studies to enhance later recall more than does providing additional study trials (e.g., Roediger & Karpicke, 2006). Thus, in the domain of language learning, for example, students’ frequent use of flash cards to study vocabulary items (Kornell, 2009; Kornell & Bjork, 2007) is wise because it induces retrieval (e.g., retrieving the English word soul when cued by the Swahili word roho). From a principles-of-learning standpoint, flash cards would seem to be a particularly effective study tool for another reason as well: They provide feedback in a natural way. After trying, successfully or unsuccessfully, to retrieve soul, for example, a student needs only to flip the card over to see roho–soul. Intuitively, feedback should be universally beneficial. However, several studies have identified conditions in which feedback provides no benefit. For example, Pashler, Cepeda, Wixted, and Rohrer (2005) taught their participants Luganda–English word pairs by presenting the pairs once and testing them twice. For example, their participants first saw leero–today and later had two separate opportunities to retrieve today when provided with leero–___. Feedback (e.g., leero–today) was provided for some items but not for others. Providing immediate feedback yielded no benefit to long-term retention when the retrieval attempt had been successful. Karpicke and Roediger (2007, 2008) reported similar results. Their participants derived no benefit from re-reading information that they had successfully retrieved (although reretrieving information doubled its later recall). This pattern of results is neither new (Guthrie, 1971) nor limited to particular instruction formats (see, e.g., Anderson, Kulhavy, & Andre, 1971). There is no doubt that feedback can be beneficial. In the study mentioned above, Pashler et al. (2005) found that immediate feedback after an incorrect response improved final recall by 494%. Further, feedback after a correct response can be helpful—if the feedback is delayed (e.g., Kulhavy & Anderson, 1972; but see Metcalfe, Kornell, & Finn, 2009). To our knowledge, however, the only evidence that immediate feedback might be beneficial after a correct response comes from Butler, Karpicke, and Roediger (2008). When their participants provided correct responses with low confidence ratings, immediate feedback substantially improved their recall. However, Butler et al.’s participants were responding to multiple-choice questions. As a result, some of their low-confidence correct responses may have reflected some guessing (i.e., after participants’ having eliminated one or two answer choices), in which case it would not be surprising if feedback were to be beneficial. Thus, Butler et al.’s results do not necessarily challenge the idea that feedback has few,