Research

Cognitive psychology - the scientific study of the mind - has the potential to drive advances in education, especially in terms of instruction/pedagogy, and contribute to educational practice that is grounded on rigorous scientific evidence. Our research focuses on human memory and learning, often with an explicit consideration for how the findings can be applied to enhance education. Our overarching goal is to promote the connections between basic cognitive research and educational practice. Listed below are some of our key research topics. PDFs of the cited papers can be obtained by going to our Publications webpage.

Other topics of research:

Test Enhanced Learning

An abundance of evidence from basic memory research has demonstrated that taking a (memory) test not only assesses one's knowledge, but also changes the state of that knowledge. In particular, taking a test after an initial learning experience typically enhances retention of the information, relative to not taking a test (or even restudying the information). This testing effect has direct implications for educators and students--tests should not be viewed solely as a means to assess learning, but instead utilised as tools to improve learning and retention.

Our first foray into this area of research examined the role of test format and corrective feedback (Kang, McDermott, & Roediger, 2007). Does the benefit of taking an initial test depend on the format of that test? Does the answer further depend on the format of the final test and on the provision of feedback? Comparing short-answer and multiple-choice test formats, we found that the former was more effective than the latter, regardless of the format of the final test. This outcome, however, was contingent on feedback being given on the initial test, suggesting that the provision of feedback is crucial to optimising the testing effect (especially when initial test performance isn't high). Also, the superiority of the short-answer initial test (regardless of final test format) implicates retrieval processes as the underlying causal mechanism of the testing effect.

In other projects, we have examined the efficacy of open-book vis-à-vis closed-book tests (Agarwal, Karpicke, Kang, Roediger, & McDermott, 2008), as well as the impact of testing on the learning of novel visuo-spatial forms (Kang, 2010). In recent projects, we found that a training procedure that involves testing (with feedback) produces better learning of a bilinear function than just observing the input-output pairings defined by the function (Kang, McDaniel, & Pashler, 2011), and that retrieval practice improved spoken vocabulary acquisition in a foreign language more than drills that require learners to repeat or imitate a native speaker's utterance (Kang, Gollan, & Pashler, 2013).

Based on these and other similar findings, my collaborators and I contend that there should be more (not less) testing in education. We do not, however, mean high-stakes standardised testing (see Roediger, Agarwal, Kang, & Marsh, 2010, for evidence-based suggestions on how testing should be implemented to enhance learning).

Spaced / Distributed Practice

It is well-established within the domain of learning and memory that long-term retention is enhanced when repetition of the to-be-remembered information is spaced apart in time rather than massed (i.e., the spacing effect) and when practice involves retrieval instead of re-presentation (i.e., the testing effect). A combination of the two manipulations--spacing and testing--has been shown to yield potent benefits for learning and retention. However, there is an unresolved question concerning the relative efficacy of two different temporal schedules of spaced retrieval practice--expanding vs. equally spaced retrieval.

Is it better for retrieval practice to be attempted soon after initial study, followed by gradually increasing intervals between successive retrieval attempts, or should the retrieval attempts be equally spaced apart? Prior research on this issue has yielded mixed findings: Landauer and Bjork (1978) found an advantage for expanding retrieval, Karpicke and Roediger (2007) and Logan and Balota (2008) found an advantage for equally spaced retrieval, while Cull (2000) found no difference between the two. In all these previous studies, criterial performance was assessed on a final test that was administered after a delay (ranging from 30 min to 8 days after practice). Also, for most studies, the practice/training phase consisted of a single session, with spacing operationalised in terms of the number of intervening items between repetitions of a particular item.

In a recent project, we compared the efficacy of expanding and equally spaced retrieval practice over a longer time scale (practice sessions were separated by days/weeks). Importantly, efficacy was assessed not just in terms of performance on a delayed final test, but also in terms of performance during the practice sessions. We found that an expanding schedule of practice not only yielded quicker acquisition and higher accessibility to the information (Japanese-English word pairs) over the lengthy training period, but it also resulted in less forgetting over the long term (Kang, Lindsey, Mozer, & Pashler, 2014).

Another ongoing project (in collaboration with Hal Pashler & Doug Rohrer) seeks to assess the effects of rereading (cf. testing/retrieval practice) on the retention of prose material when the rereading is done after a delay. Prior studies showing negligible benefits of rereading typically had the rereading occur immediately after the initial reading (i.e., massed, instead of spaced, rereading). We are investigating how delayed rereading compares against delayed retrieval practice in terms of later retention.

A recent review paper by Carpenter, Cepeda, Rohrer, Kang, and Pashler (2012) provides an overview of how spacing of practice can enhance various forms of learning. Also, this Q&A blog post on distributed practice may provide answers to some of your questions.

Interleaving and Inductive Learning

Research by Nate Kornell and Robert Bjork (2008, Psychological Science) has suggested that the benefit of spacing may extend to inductive learning--i.e., learning to generalise from relevant prior examples. They presented subjects with paintings by several artists and asked them to learn to identify each artist's style. They found that relative to "massed" presentation (a given artist's paintings were blocked and presented consecutively), subjects that viewed the paintings in a "spaced" fashion (paintings by different artists were interleaved) were better able to identify correctly the artists of previously unseen paintings (that were painted by artists whose paintings the subjects had viewed earlier).

We followed up on this research by addressing the question of whether the advantage of "spaced" presentation was due to increased temporal spacing between paintings by the same artist (during study) or due to the interleaving of paintings by different artists (Kang & Pashler, 2012). We found that increasing temporal spacing between paintings while maintaining a blocked presentation sequence produced no better inductive learning than massed presentation. On the other hand, presenting paintings by different artists simultaneously produced inductive learning that was as good as interleaved presentation and superior to massed presentation. Our findings support the idea that the spacing benefits found in perceptual induction are not because of increased temporal spacing per se but rather because of the interleaving that facilitates discriminative contrast between the categories (or artists' styles, in this case).

A recent review paper by Doug Rohrer (2012, Education Psychology Review) summarises the evidence showing that interleaving helps learners differentiate similar concepts or categories.

Metacognition and Study Strategy

While research on the testing effect has tended to focus on potential pedagogical application in the classroom, we need to consider that learning in real life often takes place outside of the classroom, in which case it is usually self-regulated--i.e., it is up to the learner to choose what information to study, how long to study, the type of strategies or processing to use when studying, and so on. These decisions depend on the learner's goals (e.g., desired level of mastery), beliefs (e.g., that a particular type of strategy is advantageous), external constraints (e.g., time pressure), and online monitoring during the learning experience.

Some questions that our research aims to answer include: Do students utilise self-testing as a strategy when they are allowed to regulate their own learning? If they do choose to self-test, what test format do they prefer, and does the answer depend on their expectancy of the format of the final test? Does prior experience with the memorial benefit of testing encourage future self-testing?

Consequences of Guessing / Errors of Commission

When you're stumped by a question on a test, should you make a guess or withhold a response? The right choice might seem trivially obvious, until one considers the possibility that making an incorrect guess might impede subsequent learning of the correct information. After all, prominent behaviourists (e.g., Skinner, Guthrie) have long warned against the commission of errors, because the errors become engrained and tend to be repeated in the future. Also, based on interference theory in memory research, one might predict that exposure to incorrect information (i.e., when one makes a wrong guess) should hinder later remembering of the correct answer.

We have conducted a series of experiments to test whether erroneous guessing affects subsequent learning from corrective feedback. By and large, we have not found any harmful effects of wrong guesses (Kang, Pashler, Cepeda, Rohrer, Carpenter, & Mozer, 2011). We are currently pursuing follow-up research to assess whether the willingness to guess (even though the guess turns out to be wrong) may be a predictor of subsequent learning from feedback.

Visual Word Recognition

Although not our main focus, we have on occasion carried out research at the intersection of memory and reading processes.

  1. There is a debate in the visual word recognition literature as to whether readers can strategically bias the degree to which lexical (i.e., whole word) and sublexical (e.g., letter level) information contributes to word pronunciation. According to the pathway control hypothesis, attentional control can be exerted over the processing pathways involved in reading: pronouncing nonwords (e.g., flirp) will increase reliance on the sublexical pathway (letter-to-sound conversion, since nonwords are not represented in one's lexicon), whereas pronouncing exception words (e.g., pint) will increase reliance on the lexical pathway (since using a letter-to-sound conversion process will yield a regularisation error). However, an alternative time criterion model is also a viable explanation of past studies that examined how list context affects pronunciation or lexical decision latency. Our study introduced an alternative dependent measure--memory performance--to see if it could provide converging evidence for pathway control. We found that the effects of word frequency and orthographic neighbourhood size were modulated in predictable ways by the list context. Importantly, the effects seen in pronunciation performance were reflected in subsequent recognition memory performance, providing strong support for the pathway control model (Kang, Balota, & Yap, 2009).
  2. Sereno, O'Donnell, and Sereno (2009) reported that words are recognised faster in a lexical decision task when their referents are physically large (e.g., cathedral) rather than small (e.g., cigarette), suggesting that "semantic size" might be an important variable that should be considered in visual word recognition research and modelling. We attempted to but failed to replicate Sereno et al.'s (2009) results--i.e., we found that "big" words were not recognised faster than relatively "small" words. Regression analyses of existing lexical decision data from megastudies (e.g., the English Lexicon Project) also converged on the same conclusion: Semantic size does not influence lexical decision performance (Kang, Yap, Tse, & Kurby, 2011). Why is it that some semantic factors (e.g., word imageability) have robust effects on lexical decision performance, whereas others (e.g., semantic size) do not? We propose that the critical determinant is whether the factor in question helps distinguish a word from a nonword (the goal of the lexical decision task). Imageability, for instance, should facilitate lexical decision, because any activation of an image upon processing of a letter string is useful evidence that the letter string is likely to be a word (and hence should speed up the "word" response). Semantic size, on the other hand, would not appear to be a useful dimension on which to distinguish words from nonwords, since both "big" and "small" words have the same degree of "wordness".

Mnemonic Benefits of Survival Processing

Nairne, Thompson, and Pandeirada (2007) propose that our memory systems serve an adaptive function, and have evolved to help us remember fitness-relevant information. In a series of experiments, they demonstrated that processing words according to their survival relevance resulted in better retention, relative to rating them for pleasantness, personal relevance, or relevance to moving house. We sought to examine if the advantage of survival processing could be replicated, using a control condition that we felt better matched the survival processing task in arousal, novelty, and media exposure: relevance to planning a bank heist. We found that survival processing nonetheless yielded better retention on both a recall and a recognition test (Kang, McDermott, & Cohen, 2008). Our findings provide additional evidence that the mnemonic benefit of survival processing is a robust phenomenon, and support the utility of adopting a functional perspective in investigating memory.

In a subsequent project, we investigated whether congruity between the items and the encoding scenario would modulate the survival processing advantage (e.g., whether words more relevant to survival would be better remembered than words that are less relevant). Instead of using a list of random words (which is typical of most of the survival processing studies done to date), we specifically selected 3 sets of words, with each set being either (i) highly relevant to the survival scenario, (ii) highly relevant to the control/bank heist scenario, or (iii) irrelevant to both scenarios. Surprisingly, when congruity was controlled a priori, we failed to find a survival advantage--subjects who rated the words (all 3 sets) for their relevance to survival recalled just as many words as those who rated the words for their relevance to a bank heist (Butler, Kang, & Roediger, 2009). We hope our uncovering of a potential boundary condition for the survival effect will spur further research into how congruity might interact with survival processing.