The compelling intuition in learning to abstract categories and concepts from examples is that such inductive learning is enhanced when the examples are blocked by category, such as solving addition problems first before moving on to subtraction problems, or learning painters' styles by comparing multiple paintings by one artist before moving on to the paintings of another artist. Recent research, however, suggest that interleaving examples of separate to-be-learned categories enhances such inductive learning. Learners typically fail to appreciate the benefit of interleaving (as compared to blocking study by category), both before study experience (Tauber, Dunlosky, Rawson, Wahlheim, & Jacoby, 2013) and even after completing a test in which their performance demonstrates better learning after interleaved study (e.g., Kornell & Bjork, 2008; Kornell, Castel, Eich, & Bjork, 2010). Across ten experiments, I investigated learners' metacognitive beliefs as to what optimizes inductive learning and how those beliefs, when in error, can be modified. In Chapter 2, I report on my efforts, using theory- and experience-based de-biasing procedures (Koriat & Bjork, 2006), to dislodge learners' beliefs in the superiority of blocking. Learners' intuitions favoring blocking proved remarkably difficult to dislodge, despite my taking multiple steps to direct participants' attention to the link between initial study schedule and subsequent test success, and providing explanations of the interleaving benefit. In Chapter 3, I report on my efforts to extend the cognitive literature by investigating whether hybrid schedules of study--that is, schedules that combine elements of blocking and interleaving--are effective. While learners remained insensitive to the benefits of purely interleaved study, they demonstrated some appreciation of hybrid schedules that were about as effective as a purely interleaved schedule. Finally, in Chapter 4, I report on my efforts to examine learners' a priori beliefs by analyzing self-scheduled study orders in light of the cognitive results found in Chapter 3, which revealed that learners can demonstrate some metacognitive sophistication with respect to category learning, when given the right opportunity to do so.

Considerable research attests that practice testing (sometimes referred to as retrieval practice) is a potent enhancer of memory (Bjork, 1975; Roediger & Karpicke, 2006; Rowland, 2014). Practice testing is thought to enhance learning because the act of retrieving information makes it more recallable in the future, possibly by strengthening or adding retrieval pathways to that information (Bjork, 1975) or by increasing the integration of recalled content with other information stored in long-term memory (Bjork & Bjork, 1992). A particularly useful feature of practice testing is its flexibility: Learners can engage in practice testing using a variety of unstructured (e.g., free-recall) and structured (e.g., multiple-choice) item formats. Much of the research on the efficacy of practice testing for learning, however, has centered on practice testing individually, despite evidence that working with others on other types of tasks (e.g., problem solving) can potentiate learning (Johnson & Johnson, 2009). Collaboration might also facilitate processes, behaviors, and cognitions which could impact the efficacy of practice testing for learning, such as offering and receiving explanations (Lou et al., 2001), engaging in overt retrieval of information (Tauber et al., 2018), and correcting errors (Barber et al., 2010). I therefore explored if practice testing together (collaborative practice testing) would yield different patterns of learning than practice testing alone. I did so across three learning contexts—a large undergraduate STEM course, an online laboratory setting, and an in-person laboratory setting—and across three different structured test formats: multiple-choice, true-false, and flashcards. I elected to use structured practice test formats as they tend to more clearly guide learners’ retrieval than unstructured test formats and therefore offer the possibility to more clearly interpret differential patterns in learning as evidence of differential patterns of retrieval during the practice testing event. Overall, I find evidence that collaborative practice testing can result in different patterns of learning than individual practice testing (Chapters 2 and 3) and learning-relevant outcomes (i.e., monitoring of one’s learning; Chapter 4). Taken together, these findings suggest that learners may have much to gain by practice testing with others.

Most laypersons assume that remembering and forgetting occur along a single continuum. That is, to remember is to avoid forgetting. To the contrary however, over 20 years of research suggests that forgetting can actually be incredibly adaptive, and indeed quite essential for the efficient functioning of the memory system. Forgetting serves as an updating mechanism, allowing for the successful retrieval of current information that is relevant, at the expense of irrelevant or outdated information. One very illustrative example of such adaptive forgetting is a phenomenon known as retrieval-induced forgetting (Anderson, Bjork, & Bjork, 1994). While information that is retrieved is strengthened in memory and thus becomes more recallable than it would have been otherwise, related items in competition with that which is retrieved become less recallable. This decrement in recall for related, competing information is referred to as retrieval-induced forgetting.

Investigations of retrieval-induced forgetting typically employ a standard retrieval-practice paradigm that consists of three phases. During the initial learning a phase, a series of category-exemplar pairs, drawn from a number of different categories, are presented to the participant for study (e.g., fruit: orange, drinks: whiskey, trees: elm). Subsequently, during the retrieval-practice phase, a subset of the pairs are retrieved in response to category-plus-two-letter-stem cues (e.g., fruit:or, drinks: wh). After a brief distractor task, participants are prompted to retrieve all of the initially presented pairs in response to category-plus-one-letter stem cues (e.g., fruit; o, drinks:w). This retrieval-practice paradigm creates three types of items: Rp+ items that were retrieved during the retrieval-practice phase (e.g., fruit: orange), related Rp- items that share the same category cue but were not retrieved during retrieval-practice (e.g., fruit: apple), and baseline or Nrp items from nonpracticed categories that did not receive retrieval-practice (e.g., trees: elm). Not surprisingly, Rp+ items that were given retrieval practice after study are recalled better than related Rp- items from the same categories that were not practiced. Interestingly however, relative to nonpracticed items from nonpracticed categories, (i.e. Nrp items), Rp- items are impaired. Said differently, on average, participants recall fewer Rp- items than Nrp items. This decrement in recall for Rp- items relative to Nrp items is referred to as retrieval-induced forgetting.

Such retrieval-induced forgetting is assumed by many to reflect inhibitory mechanisms recruited during retrieval of target items to decrease the accessibility of related, nontarget items. In Chapter 1, I will briefly review the literature on retrieval-induced forgetting.

Recently however, Jonker, MacLeod, and Seli (2013) proposed an alternate account that emphasized the role of context in producing retrieval-induced forgetting. In this framework, the study and retrieval-practice phases are represented as two disparate contexts. While the study context includes all of the studied items – that is, the Rp+, Rp-, and Nrp items – the practice context only contains the Rp+ items. Critically, while the Nrp or baseline items are only present in the study context, the Rp- cues are associated with both the study context and the retrieval-practice context. Thus, at test, participants may search the retrieval practice context first which then makes it more difficult to recall items from the study context. The Nrp cues, however, are only associated with the study context, and are therefore very effective at reinstating the initial study context.

In evidence of this account, Jonker et al. (2013) showed that even restudy practice—which is assumed by the inhibitory account to be insufficient to cause forgetting —can cause forgetting when a mental context change is inserted between study and restudy. In Chapter 2, I present a series of experiments designed to replicate this finding while also testing the possibility that a far mental context change would cause more forgetting than a near mental context change.

An important assumption of the contextual-cuing account is that if only the practice context is reinstated at final test, retrieval-induced forgetting will occur because participants are unable to access Rp- items. In contrast, if the study context is reinstated at test, because all items are available, retrieval-induced forgetting should be eliminated. To test this critical assumption, in Chapter 3, several experiments are presented in which subjects were encouraged to reinstate the appropriate context regardless of whether a given item was practiced or unpracticed. Finally, in Chapter 4, a summary and interim conclusions are given, as well as future directions that may be fruitful in this domain of research.

Retrieving information is an effective learning strategy to promote the long-term retention of materials (see, e.g., a meta-analysis by Rowland, 2015), so tests can foster learning as well as assess learning. Multiple-choice testing, however, has often been criticized as not engaging retrieval processes to the same extent as do other test types, such as cued-recall and free-recall (see, e.g., a meta-analysis by Hamaker, 1986). Recent research, however, suggests that as a pedagogical tool, multiple-choice tests can in fact trigger productive retrieval processes, provided the incorrect alternatives are competitive enough to induce the retrieval of why they are incorrect, and can even have benefits over other formats. More specifically, multiple-choice tests can boost the recall of non-tested, related information as learners retrieve and reject information connected to those alternatives (e.g., Little, Bjork, Bjork, & Angello, 2012), especially when the test format requires that learners make confidence judgments about the alternatives (Sparck, Bjork, & Bjork, 2016). Additionally, multiple-choice testing might be an efficient way to study when one is tasked with learning a large amount of information. The experiments reported in this dissertation were designed to explore further the potential of multiple-choice testing as a tool for learning.

The results of Experiments 1 and 2 suggest that taking a confidence-weighted multiple-choice test, which requires that a test taker consider more carefully than does a standard multiple-choice test why a given alternative is correct or incorrect, can lead a learner to transfer that behavior to a subsequent standard-format multiple-choice test. The results of Experiment 3, on the other hand, in which the focus was shifted to the possible benefits of multiple-choice tests as pre-tests presented before a study phase, found no significant benefits of the confidence-weighted format over the standard format.

Experiments 4 and 5 were designed to examine whether multiple-choice testing might be beneficial in the learning of vocabulary words, especially when the to-be-learned words are difficult and confusable. Again, the results demonstrated that multiple-choice testing can have benefits that go beyond the benefits of cued-recall testing, especially for words that are incorrect alternatives on an initial test, but are correct alternatives on a subsequent test. The results have major implications for the design of flashcards.

In summary, the results reported in this dissertation demonstrate that a multiple-choice test, when well designed, can be a powerful pedagogical tool, one that can contribute to optimizing educational practices.

The internet has revolutionized how we access information, allowing us not only to find new knowledge, but also to retrieve information that we may already have stored in our memory but does not readily come to mind. As a consequence, we are vulnerable to googling before even trying to retrieve such information (e.g., Storm et al., 2017; Sparrow et al., 2011), thus bypassing the benefits of “retrieval as a memory modifier” (Bjork, 1975): that is, making the recalled information more memorable in the future. Googling before thinking can have metacognitive costs as well, such as impairing judgments as to whether information will be recallable if needed in the future. In addition, even a failed attempt to retrieve yet-to-be-learned information can potentiate learning, versus simply being presented with that information, typically referred to as the pretesting effect (e.g., Grimaldi & Karpicke, 2012; Kornell, Hays & Bjork, 2009; Little & Bjork, 2016). To investigate the potential benefits of pretesting with the internet, I examined whether an attempt to guess the answer to a question or problem before using the internet to find the solution (aka thinking-before-googling) can make the subsequently googled information more memorable compared to simply googling the answer right-away. Additionally, I wondered as to whether thinking about yet-to-be-learned information can also strengthen the memory for relevant information learned prior to the Google search. (Chapter 2 & 3). Furthermore, I explored ways to enhance the attractiveness of engaging thinking-before-googling attempts without reducing its efficacy (Chapter 4). Overall, attempting to think, before consulting the internet, about a question or a solution to a problem promotes better memory for searched-for information found on the internet as well as relevant information learned prior to the Google search. Furthermore, the findings indicate that the idea of hint support may result in a higher self-reported willingness to engage in thinking-before-googling attempts, without compromising the benefits of pretesting in the context of internet-based learning. Taken together, these findings have implications for how one could use the internet as a more effective tool for learning.