What is test expectancy?

According to some descriptions, (e.g. Craik and Lockhart, 1972) memory is nothing more than a by-product of processing. The depth at which information is processed determines the strength of the memory trace and how easily we can recall it later. Information can be viewed as the raw material for creating knowledge, but the operations we apply to this raw material determine the type of knowledge that will take up relatively permanent residence in memory. This is the action we generally describe as information processing. Knowledge formation is hypothesised to lead to schema generation. Schemas can, therefore, be viewed as collections of knowledge. 

Levels of Processing Theory

But what are the operations that lead to knowledge formation? Craik and Lockhart’s Levels of Processing theory (see fig.1) suggests that information entering the memory cycle (encoding – consolidation – retrieval – re-consolidation) can be processed at different levels, from shallow (e.g. rote rehearsal) to deep (e.g. elaborative). Different encoding strategies will determine this depth or level, while consolidation will result in the formation of schemas that provide flexible packets of knowledge that can be applied to different, yet similar, situations to the ones that created them. This means that how often you visit the same piece of information has less impact than the way in which you visit it – it’s how and not necessarily how often

Levels of Processing

A Reason for Learning

But does utility matter? That is, does the reason for learning the new information impact the encoding strategy?  If we accept the process of learning is behaviourally undetectable, then we also accept the view that to know learning has taken place requires some kind of evidence after the fact. If I learn a series of questions and answers in a new language then I can only be sure that I have learned these sentences by, for example, engaging in a conversation where they can be used or perhaps offering answers when interrogated on my understanding. Such assessments take many forms, including completing essays or short answer type questions in formal examination settings and multiple-choice style quizzes.  Research has discovered that when people are told the type of test they’ll receive, they adapt their encoding strategies to fit the format, a behaviour known as test expectancy. This behavioural change is generally referred to as the encoding-strategy adaptation hypothesis. Tests, within the confines of the laboratory at least, tend to be either recognition-based or recall-based. The former is best described as multiple-choice, although we may ask participants to identify learned words from a list including words not learned (a recognition test). Interestingly, people can often identify the words in a recognition test even when they failed to do so in a recall test, so it would appear the words have still been learned. 

Does this mean that relatively permanent changes in long-term memory (a common definition of learning) depends, to some extend, on how this learning is processed? 

Cabeza noted that the act of recognising information appears to be qualitatively different to recalling it (Cabeza et al., 1997) and it would therefore seem plausible that participants would engage different encoding strategies to best fit the type of assessment expected. Furthermore, in laboratory studies (but not necessarily classroom studies) participants who study with the expectation of a recall test outperform those who study with expectation of a recognition test, regardless of the test format. Why might this be the case?

Beliefs about the type of test seem to be an important factor. Recall tests are seen as more demanding than recognition tests so may well encourage intense study and more varied strategies. According to Rivers and Dunlosky (2020) these differences fall into either quantitative or qualitative. The former may lead to learners using the same strategy, just using it to a greater degree. Alternatively, the latter may lead to the adoption of different study strategies. At the very least, learners make quantitative changes to their encoding (Lundeberg and Fox, 1991). 

Testing test expectancy

In order to experimentally study test-expectancy, participants are required to study material (perhaps a list of words) and led to expect a particular test format (recall or recognition). This can be done by directly telling participants the format or by running early trials using one format and then switching the format for the critical trial. Performance on the final critical test is compared between participants led to expect a different test format. In many studies, these lists include word pairs. Word pairs are particularly useful because they result in different kinds of recall: cued or free. Say, for example, I present you with a series of unrelated word pairs: accident – shovel, plant – bike, house – tree. I can then test you by either giving you the first word of the pair and ask you to recall it’s partner word (cued recall) or ask you to write down as many of the second words of the pair in any order that you can remember (free recall). So with cued recall I would say ‘accident’ and hopefully you would reply ‘shovel’. With free recall, you would simply be required to answer ‘shovel, bike, tree.’ We could also manipulate the relatedness of the word pairs, such as worm – apple, animal – lion to see if these changes made any impact on the results.

But what can this tell us about test expectancy? We can assume that if you were expecting a cued recall test, you might make more effort to learn the word pairs because the first word would act as a cue (a nudge) and help you recall the second half of the pair. However, would you still learn the word pairs if you were expecting a free recall test where the first word of the pair might appear somewhat superfluous? What if the words were presented in pairs but the pairing made no difference because all the words could be recalled in any order regardless of their pairings? What would then happen if you were expecting a free recall test but I gave you a cued recall test? Ultimately, it’s the same information, but would you treat it differently if you knew how your ability to recall it was going to be tested? 

In a study from 2020, Michelle Rivers and John Dunlosky ran three experiments, all with differing degrees of word manipulation and expectancy. Participants performed best when the critical test matched their expectations, but the manipulation of the associations between the words had a greater effect on cued recall tests (participants were using the stronger associations between the words to help them with the test). These results are consistent with some previous studied that found expectancy to be related to associative encoding (deeper processing), that is, making links between discrete pieces of information.  

Recall or Recognition?

Recall tests may, therefore, be more effective than recognition tests (e.g. multiple-choice) because the expectancy of a recall tests is going to result in deeper (or more elaborative) processing. But if participants are told to expect a recall test and are then given a recognition test, they should still do better than if they were told to expect a recognition test because they prepared by engaging in much deeper processing anyway. Expectations, therefore, can lead to different encoding strategies.  

Published by

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s