Following a 1994 study that found that errorless learning was better than trial-and-error learning for amnesic patients and older adults, errorless learning has been widely adopted in the rehabilitation industry. Errorless learning involves being told the answer without repeatedly trying to answer the question and perhaps making mistakes. For example, in the 1994 study, participants in the trial-and-error condition could produce up to three errors in answer to the question “I am thinking of a word that begins with QU”, before being told the answer was QUOTE; in contrast, participants in the errorless condition were simply told “I am thinking of a word that begins with QU and it is ‘QUOTE’.”
In a way, it is surprising that errorless learning should be better, given that trial-and-error produces much deeper and richer encoding, and a number of studies with young adults have indeed found an advantage for making errors. Moreover, it’s well established that retrieving an item leads to better learning than passively studying it, even when you retrieve the wrong item. This testing effect has also been found in older adults.
In another way, the finding is not surprising at all, because clearly the trial-and-error condition offers many opportunities for confusion. You remember that QUEEN was mentioned, for example, but you don’t remember whether it was a right or wrong answer. Source memory, as I’ve often mentioned, is particularly affected by age.
So there are good theoretical reasons for both positions regarding the value of mistakes, and there’s experimental evidence for both. Clearly it’s a matter of circumstance. One possible factor influencing the benefit or otherwise of error concerns the type of processing. Those studies that have found a benefit have generally involved conceptual associations (e.g. What’s Canada’s capital? Toronto? No, Ottawa). It may be that errors are helpful to the extent that they act as retrieval cues, and evoke a network of related concepts. Those studies that have found errors harm learning have generally involved perceptual associations, such as word stems and word fragments (e.g., QU? QUeen? No, QUote). These errors are arbitrary, produce interference, and don’t provide useful retrieval cues.
So this new study tested the idea that producing errors conceptually associated with targets would boost memory for the encoding context in which information was studied, especially for older adults who do not spontaneously elaborate on targets at encoding.
In the first experiment, 33 young (average age 21) and 31 older adults (average age 72) were shown 90 nouns presented in three different, intermixed conditions. In the read condition (designed to provide a baseline), participants read aloud the noun fragment presented without a semantic category (e.g., p_g). In the errorless condition, the semantic category was presented with the target word fragment (e.g. a farm animal p_g), and the participants read aloud the category and their answer. The category and target were then displayed. In the trial-and-error condition, the category was presented and participants were encouraged to make two guesses before being shown the target fragment together with the category. The researchers changed the target if it was guessed. Participants were then tested using a list of 70 words, of which 10 came from each of the study conditions, 10 were new unrelated words, and 30 were nontarget exemplars from the TEL categories. Those that the subject had guessed were labeled as learning errors; those that hadn’t come up were labeled as related lures. In addition to an overall recognition test (press “yes” to any word you’ve studied and “no” to any new word), there were two tests that required participants to endorse items that were studied in the TEL condition and reject those studied in the EL condition, and vice versa.
The young adults did better than the older on every test. TEL produced better learning than EL, and both produced better learning than the read condition (as expected). The benefit of TEL was greater for older adults. This is in keeping with the idea that generating exemplars of a semantic category, as occurs in trial-and-error learning, helps produce a richer, more elaborated code, and that this is of greater to older adults, who are less inclined to do this without encouragement.
There was a downside, however. Older adults were also more prone to falsely endorsing prior learning errors or semantically-related lures. It’s worth noting that both groups were more likely to falsely endorse learning errors than related lures.
But the main goal of this first experiment was to disentangle the contributions of recollection and familiarity to the two types of learning. It turns out that there was no difference between young and older adults in terms of familiarity; the difference in performance between the two groups stemmed from recollection. Recollection was a problem for older adults in the errorless condition, but not in the trial-and-error condition (where the recollective component of their performance matched that of young adults). This deficit is clearly closely related to age-related deficits in source memory.
It was also found that familiarity was marginally more important in the errorless condition than the trial-and-error condition. This is consistent with the idea that targets learned without errors acquire greater fluency than those learned with errors (with the downside that they don’t pick up those contextual details that making errors can provide).
In the second experiment, 15 young and 15 older adults carried out much the same procedure, except that during the recognition test they were also required to mention the context in which the words were learned was tested (that is, were the words learned through trial-and-error or not).
Once again, trial-and-error learning was associated with better source memory relative to errorless learning, particularly for the older adults.
These results support the hypothesis that trial-and-error learning is more beneficial than errorless learning for older adults when the trials encourage semantic elaboration. But another factor may also be involved. Unlike other errorless studies, participants were required to attend to errors as well as targets. Explicit attention to errors may help protect against interference.
In a similar way, a recent study involving young adults found that feedback given in increments (thus producing errors) is more effective than feedback given all at once in full. Clearly what we want is to find that balance point, where elaborative benefits are maximized and interference is minimized.