Eighty-one Spanish-speaking children were recruited and took part in the study. All subjects had normal or corrected-to-normal vision and no history of learning disabilities. Data from 4 participants were excluded due to not being able to learn at least 75% of the to-be-studied word pairs or having outlier values (mean ± 2.5 * standard deviation) in the final test administered 2 days after encoding, leaving a final sample of 77 subjects (M = 11.03 years; SD = 2.12 years; 41 females). The sample was divided into four groups based on the corresponding age group and learning procedure. The age division was made according to the critical improvements observed half way through middle childhood years (e.g., Ghetti and Angelini, 2008; Paz-Alonso et al., 2009; Paz-Alonso et al., 2014), resulting in a Younger children group (i.e., middle-childhood, ages 7–10) and an Older children group (i.e., early-adolescence, ages 11–14). The second division, based on the initial procedure to commit items to memory, separated the participants into Repeated Study group (i.e., learning mainly through re-study) or Repeated Retrieval group (i.e., learning mainly through testing). According to this criterion, our final sample consisted of younger participants (i.e., between 7 and 10 years old) who encoded the items either under repeated study conditions (Younger Repeated Study group: n = 18, M = 9.21 years; SD = 1.16 years; 10 females), or via repeated retrieval (Younger Repeated Retrieval group: n = 20, M = 9.20 years; SD = 1.07 years; 10 females); and older participants (i.e., between 11 and 14 years old) who encoded the items engaging either in repeated study (Older Repeated Study group: n = 20, M = 12.81 years; SD = 1.07 years; 9 females) or repeated retrieval (Older Repeated Retrieval group: n = 19, M = 12.82 years; SD = 1.19 years; 12 females). All participants were recruited from the same school in Vitoria (Spain) at which Spanish, Basque and English are taught from preschool to school years with a similar degree of exposure to these languages across these years. Prior to taking part in the experiment, participants’ parents gave written informed consent in compliance with the ethical regulations established by the Ethics Committee of our research center and the guidelines of the Helsinki Declaration.

Within each age group, repeated study and repeated retrieval encoding groups resulted to be matched in gender and individual difference variables, including: vocabulary knowledge and fluid reasoning (Kaufman Brief Intelligence Test, K-BIT-II; Kaufman and Kaufman, 1990), working memory (Woodcock-Johnson-III, numbers reversed and auditory working memory subtests; Woodcock et al., 2001), speed of processing (Woodcock-Johnson-III, cross-out subtest; Woodcock et al., 2001), short-term visual memory (Cambridge Neuropsychological Test Automated Battery, CANTAB; Delayed Matching to Sample, DMS, and Pattern recognition memory, PRM subtests) and visual memory recognition (CANTAB; Paired Associates Learning, PAL) (see Table 1).

A total of 60 Spanish weakly semantically related cue-target word pairs (e.g., hole-cheese) were used. Thirty new cue words were also selected for the retrieval test that took place 2 days after the encoding happened to ensure that participants were engaging in the task by reducing the frustration they could experience when not remembering studied items. Participants exhibited a near-perfect accuracy at withholding responses to these non-studied novel cues. All word stimuli were classified as words typically acquired by age 7. Moreover, cue words, target words and new cue words were matched for age of acquisition, familiarity and imageability psycholinguistic norms (ps ≥ 0.16).

Participants from both learning groups (i.e., repeated study, repeated retrieval) underwent two experimental sessions 2 days apart from each other. They were randomly assigned to the repeated study and repeated retrieval conditions. In session I (i.e., initial), participants were presented with the 60 Spanish word pairs to commit to memory. Session I was divided into 8 study-test cycles wherein participants studied the to-be-remembered word pairs under either the repeated study or repeated retrieval learning conditions depending on their assigned group. Each cycle consisted of a study phase, where the word pair (e.g., hole - cheese) appeared on the computer screen for 7,000 ms while they attempted to memorize it, and a cued recall test phase, where participants were shown the first word of the word pair for 5,700 ms (e.g., hole - ____) and had to verbally recall the second word of the pair.

Younger and older participants in the repeated study groups studied all of the 60 experimental word pairs in each of the 8 study-test cycles and were only re-tested on those items that they did not remember in the immediately preceding cycle. Younger and older participants in the repeated retrieval groups retrieved the 60 experimental word pairs in all of the 8 study-test cycles. They were only asked to re-study those items they were unable to remember in the immediately preceding cycle. Thus, while both groups underwent a study and a retrieval phase for them to learn the pairs and assess whether they had learned them correctly, each group focused on either re-studying or re-testing to commit the pairs to memory. Participants who were not able to correctly remember at least 45 word pairs by the end of the 8th study-test cycle were excluded (n = 3; the remaining excluded participant exhibited an outlier performance in the final test, 2 days after encoding). The low number of participants who were not able to successfully encode at least 75% of the word pairs indicates that the majority of the children were able to understand and perform the task.

Table 2 shows the encoding procedure in 8 study-test cycles and the average number of stimuli presented in each cycle, as a function of Age (younger children, older children) and Encoding group (repeated study, repeated retrieval). Importantly, the overall amount of exposure to the stimuli did not vary significantly between the repeated study and repeated retrieval conditions, either in the older children group (p = 0.80) or in the younger children group (p = 0.08). Nevertheless, to rule out the possibility that individual differences in the degree of exposition to the word pairs during encoding may determine developmental effects, we tested whether or not controlling the number of expositions in our main analysis had an influence on the results.

In session II, which took place 2 days after session I, participants performed a final cued recall test. Participants were presented with the cue words for the previously studied 60 word pairs (e.g., hole -___) and were asked to respond aloud (e.g., “cheese”) while the cue was on the screen (i.e., 5,700 ms). Together with the studied cue words, participants encountered 30 new Spanish cue words that had not previously been presented to the participants during the encoding phase to reduce the frustration they could experience when not remembering studied items. They were encouraged to say “pass” anytime they did not know the response to a given cue word.

The paradigm we utilized in the present study has been extensively used in the testing effect literature. The backward testing effect paradigm is of special interest in controlling for the items encoded in long-term memory by the end of the learning phase, which is critical in order to level the playing field among developmental groups in this sense. It is also of special interest in making sure that once a memory is encoded, that memory is only reinforced either through repeated retrieval or repeated study, as a function of the encoding group conditions. Across several analyses, our results demonstrate that the effects of testing differ across the middle childhood years. All groups were able to encode most of the to-be-encoded items and performed at ceiling in the first session, but in the second session (2 days after encoding), only older children exhibited a testing benefit. In fact, the only group that differed in the number of correctly retrieved items 2 days after encoding was the older repeated retrieval group. Importantly, this developmental effect holds when controlling for the ability to learn from feedback/study or the degree of exposition to the to-be-encoded word pairs during the learning phase.

Moreover, the group of participants who engaged in retrieval practice during encoding exhibited a strong association between long-term memory retention and age. This suggests improvements in the ability to benefit from testing from early middle childhood to early adolescence. In contrast, the group of participants who engaged in study practice during encoding did not show statistically significant associations between long-term memory retention and age. These age-related changes in the ability to benefit from the testing effect over the middle childhood years appear to be concomitant with well-documented improvements in other cognitive processes that also occur from early to late middle childhood such as binding operations (Ghetti and Angelini, 2008; Lloyd et al., 2009; Sluzenski and Newcombe, 2006), mnemonic control (Ashcraft et al., 1976; Bunge et al., 2002; Emmerich and Ackerman, 1978; Paz-Alonso et al., 2013a; Tang et al., 2018; Yu et al., 2018) and the ability to process semantically related information (Cunningham and Stanovich, 1991; Paz-Alonso et al., 2013b; Paz-Alonso et al., 2008). Developmental cognitive neuroscience research on memory retrieval have also characterized age-related changes in hippocampo-cortical functional coupling from early middle childhood to adolescence (e.g., Ofen et al., 2012; Paz-Alonso et al., 2013a; Paz-Alonso et al., 2013b) that can be at the base of the developmental differences here observed in the ability to benefit from testing effect.

Besides investigating age-related effects on the testing effect, we also sought to examine the role of individual differences in this phenomenon. Our results revealed that the long-term memory retention of participants who encoded information via repeated study was positively associated with their vocabulary knowledge, short-term memory and their ability to retain and manipulate information in working memory. However, long-term memory retention in the repeated retrieval group did not show any associations with these or other cognitive measures. This finding suggests that variables related to the above-mentioned concomitant changes observed during middle childhood years are not specifically related to the benefits of the testing effect. Only information learnt via the repeated study encoding procedure was associated with other individual difference measures. These results are in line with previous evidence showing that individual differences in working memory and fluid intelligence are associated with the ability to benefit from different encoding agendas (Agarwal et al., 2017; Brewer and Unsworth, 2012). For instance, Agarwal et al.’s (2017) study found that differences in long-term memory retention for information encoded via testing versus repeated study was greater for those participants with lower working memory capacity.

The fact that these associations were only present in the repeated study group suggests that only information encoded via the repeated study condition benefits from these individual difference variables. Differing from repeated testing, the repeated study encoding condition did not include specific instructions regarding the cognitive operations to be performed during encoding of the to-be-remembered items. In this context, previous semantic knowledge may be critical for strengthening or forming new semantic relations between weakly semantically related word-pairs. Similarly, holding information in mind at the short-term memory level and performing operations with this information may be strongly beneficial in terms of binding elements in a word pair together or binding a word pair to other external elements during encoding. This could facilitate subsequent retrieval of the information. It has further been shown that these types of encoding strategies are associated with better long-term episodic memory (see Richardson, 1998 for a review). Thus, under the repeated study condition, encoding strategies may arise spontaneously, depending on individual differences in semantic knowledge and short- and working-memory capacity, leading to better long-term memory retention.

In contrast, the repeated retrieval encoding condition capitalizes on actively performing retrieval operations in order to commit the to-be-remembered information to long-term memory. In this sense, participants exercising retrieval practice have already been provided with specific instructions and strategies. They know what cognitive operations they should engage in during encoding. This suggests, in line with the wealth of research evidence demonstrating the robustness of the testing effect, that testing may per se benefit long-term memory retention, without the need for additional encoding strategies.

We confirmed our hypothesis finding age-related improvements in the ability to benefit from the testing effect over the middle childhood years. However, we found these specific testing effects were not related to individual differences in other mnemonic and higher-cognitive functions known to improve over this same developmental period, such as binding operations, relational semantic processing and mnemonic control.

This lack of overlap with concomitant developments in cognitive abilities during middle childhood may be due to the fact that the testing effect differs considerably from other episodic long-term memory paradigms typically used in studies which show improvements in the above-mentioned mnemonic and higher-cognitive functions over the middle childhood years. If so, the testing effect might be governed by different mechanisms. The benefits of testing are typically more robust after a retention interval of at least 24 h (i.e., 2 days in our experiment). These retention intervals, longer than those typically used in paradigms, such as the Desee/Roediger-McDermott (DRM) task (Deese, 1959; Roediger and McDermott, 1995), dual memory paradigms to measure recollection and familiarity (Yonelinas, 1994), the Think/No-Think paradigm (Anderson and Green, 2001), etc., introduce a pivotal new process that also develops over the middle childhood years: memory consolidation. Different studies have shown developmental changes in the sleep-dependent consolidation processes that affect memory consolidation in children from middle childhood through to adolescence. For example, children do not benefit from post-learning periods of sleep in finger sequence tapping and mirror tasks (Prehn-Kristensen et al., 2009; Wilhelm et al., 2008) as adults do (Plihal and Born, 1997; Walker et al., 2003). In fact, using a longitudinal design, Hahn et al.’s (2019) study showed that changes in the distribution of fast spindles toward an adult-like topography (i.e., increased central fast sleep spindle density) occurred between late-middle childhood and adolescence; and that this change was associated with memory consolidation (Hahn et al., 2019).

Although this is just offered as a potential explanation for the age-related changes observed in the ability to benefit from the testing effect in the present study, it is closely related to one of the most important methodological differences between testing effect studies and other more classic episodic memory paradigms. The main paradigms used to examine episodic long-term memory during development, which have demonstrated improvements in binding operations, relational semantic processing and mnemonic control operations, did not involve long retention intervals (e.g., 1 or 2 days) and did not comprise overnight sleep periods. To better understand developmental differences related to the ability to benefit from testing, future research should focus on analyzing differences in sleep efficiency in developmental samples and adults, as well as how sleep cycles and particularly spindle pattern changes during childhood impact long-term memory retention as a function of how information is encoded. Moreover, it would be great if future research can replicate these findings across diverse cultural and educational settings, using alternative materials, and broaden the range of cognitive covariates examined (e.g., attentional control, further executive functions).

Results from the present study could have an impact in terms of implementing the testing effect in applied settings. According to our data, children can further benefit from testing around age 10. However, it is worth mentioning that the younger-aged group did not show any negative effects from the repeated retrieval relative to the repeated study encoding strategy. And other studies have shown the testing effect can be present in children as young as 3–5 years old (Fritz et al., 2007). This leads us to suggest that according to our data repeated retrieval encoding could be implemented as early as 1st grade in elementary school, since it has been shown to benefit other aspects of learning such as anxiety control (Agarwal et al., 2014). Training in encoding through testing from the early stages of school may also lead to an improved capacity to benefit from testing in the future.