The developmental influence of primary memory capacity on working memory and academic achievement.

Immediate Memory and Academic Performance

Working memory tasks measure memory storage in the face of competing distraction and are thought to reflect a set of abilities, including immediate storage capacity, speed of processing, and executive control processes ( Bayliss et al., 2003 ; Kane, Conway, Hambrick, & Engle, 2007 ; Unsworth, Redick, Heitz, Broadway, & Engle, 2009 ). Developmental improvements can be seen in general speed of processing ( Kail & Ferrer, 2007 ) and executive control tasks ( Alloway, Gathercole, Willis, & Adams, 2004 ). Furthermore, immediate storage capacity has been shown to increase between the ages of 4 and 15 years ( Alloway et al., 2004 ). The growth observed in these domains has been linked to increases in working memory task performance during development, as well as increasing academic aptitude and general intelligence ( Fry & Hale, 2000 ). Executive control in particular has received a great deal of attention, as working memory tasks are typically more predictive of academic performance than are measures of immediate storage or speed of processing ( Bayliss et al., 2003 ; Swanson, 1994 ; Swanson & Alloway, 2012 ). However, attempts to fragment working memory tasks into their component parts have shown that processing speed is a predictor of classroom behavior ( Jarrold, Mackett, & Hall, 2014 ), while immediate storage capacity predicts unique variance in reading ability ( Bayliss et al., 2003 ) and mathematics ( Bull et al., 2008 ). Indeed, Colom et al. (2008 ; see also Shahabi, Abad, & Colom, 2014 ) have suggested that immediate storage capacity alone underpins the link between working memory and academic attainment.

Reading is likely to tax the developing immediate memory system as multiple pieces of information must be managed online in order to decode words while building mental models of sentences for comprehension ( Engel de Abreu & Gathercole, 2012 ; Wang & Gathercole, 2013 ). Similarly, mathematical problems require the concurrent storage of task instructions and running totals ( Andersson, 2008 ; Kyttälä, Aunio, Lepola, & Hautamaki, 2014 ). Immediate memory capacity undoubtedly plays a role in successful use of working memory for this purpose, and valid measures of this construct must therefore be used to clarify the nature of any suggested relationship between academic performance and working memory.

Many studies examining this link have used immediate serial recall (ISR) span tasks to measure storage capacity in developing populations (e.g., Bayliss et al., 2003 ; Bull et al., 2008 ; Gathercole, Pickering, Knight & Stegmann, 2004 ); however, we argue that span tasks are not ideal measures of immediate memory, as they are inherently impure. It is probable that multiple systems underpin ISR performance, with competing theories implicating active portions of long-term memory, or the use of strategic skills and metamemory (cf. St. Clair-Thompson, 2007 ), all alongside any temporary memory storage system. Indeed, in work intended to isolate the “focus of attention” ( Cowan, 2001 ; Oberauer, 2003 ) from any contribution of rehearsal or long-term memory when measuring immediate memory, Cowan, Elliott et al. (2005) observed that the number of items that could be held within the focus of attention was strongly related to aptitude measures. In the current work we therefore assess whether our novel measures, designed to specifically estimate and characterize immediate memory in the absence of the above confounding factors, are more predictive of academic ability than are standard ISR tasks in young children. ISR and novel tasks will be compared and contrasted, with the aim of determining whether better predictions of academic performance can be made if immediate memory can be isolated from strategic influences and long-term memory contributions.

In this study, we elected to focus solely on recall of verbal information. Immediate recall in the verbal domain has been heavily studied in relation to academic achievement in children ( Bayliss et al., 2003 ; Bull et al., 2008 ; Gathercole, Alloway, Willis, & Adams, 2006 ), and may, potentially, rely on separate systems to those involved in visual or spatial immediate recall ( Alloway, Gathercole, & Pickering, 2006 ; Smith & Jonides, 1997 ). Indeed, verbal and visuospatial immediate recall measures are dissociable from one another throughout development from age 4 to 15 ( Alloway et al., 2004 ). Perhaps as a result of this, the associations between visual immediate memory and academic measures differ somewhat from those seen with verbal immediate recall ( Holmes & Adams, 2006 ; Titz & Karbach, 2014 ). However, we note that while there is evidence for the separability of these two immediate memory systems, there is also support for the view that they share common processes and features ( Chuah & Maybery, 1999 ; Cortis, Dent, Kennett, & Ward, 2014 ; Jones, Farrand, Stuart, & Morris, 1995 ), and it was these processes that we sought to capture in this work.

Primary Memory

The terms short-term memory (STM) and long-term memory bring with them assumptions about the temporal properties of the memory system. However, this focus on duration (short and long) carries with it some ambiguity about the processes involved in retaining memory material. The current work therefore uses in preference Waugh and Norman’s (1965 , see also James, 1890 ) definition of primary memory to characterize the “pure” capacity of immediate memory in the absence of additional contributions from rehearsal processes or long-term memory. In contrast to STM, primary memory carries with it theoretical assumptions about the processes involved in, and phenomena associated with, immediate memory recall. It therefore provides a potentially more informative framework in which to study and characterize immediate memory recall. The characteristics of primary memory include the fact that it accommodates the concurrent maintenance of a fixed number of items ( Waugh & Norman, 1965 ; Unsworth & Engle, 2007 ), and that it is open to conscious awareness ( James, 1890 ). Primary memory therefore maintains the subset of items that fall within some form of focus of “current attention” ( Broadbent, 1958 ). However, recall from primary memory is also characterized by spontaneous serial ordering of the output, even when accurate serial order is not required ( Broadbent, 1958 ; Bryden, 1971 ; Sperling, 1967 ). Secondary memory (which may be seen as a complement to long-term memory) is, on the other hand, characterized by recall that is not spontaneously serial ordered and is likely to be the product of a more controlled or probed search of items not within the current focus of attention. Shelton, Elliott, Matthews, Hill, and Gouvier (2010) have used structural equation modeling to show that primary memory, secondary memory, and working memory are indeed separable latent factors, which contribute differently to variation in fluid intelligence (see also Unsworth & Engle, 2007 ; Unsworth, Brewer, & Spillers, 2010 ).

This conceptualization of primary memory, therefore, obviously shares many similarities with more recent notions of the focus of attention in models of working memory (e.g., Cowan, Elliott et al., 2005 ; Cowan, Nugent, Elliott, Ponomarev, & Saults, 1999 ; Oberauer, 2003 ), which is the items held active in immediate memory, independent of sensory information ( Cowan, 2011 ). However, we prefer the term primary memory because it has clearly defined processes and characteristics, that include the additional claim that recall from this system is characterized by spontaneous and accurate serial order output ( Broadbent, 1958 ; Bryden, 1971 ; Sperling, 1967 ).

There are also two reasons why a focus on primary memory is timely and conceptually significant (see also, Unsworth et al., 2007 ). First, few studies have directly addressed the development of primary memory, with those that have done so recently producing conflicting results ( De Alwis, Myerson, Hershey, & Hale, 2009 ; Jarrold et al., 2013 ; Roome, Towse, & Jarrold, 2014 ). Second, we chose to focus on primary memory to adapt existing (adult-based) paradigms that capture significant elements of this concept. We modified a dichotic listening paradigm, which was originally used by Broadbent (1958) to reveal the differences between primary memory and perceptual attention. Following Waugh and Norman (1965) , immediate and probed free-recall tasks were also used to provide potential indices of primary memory capacity. As such, the novel tasks in this paper have been developed to specifically index spontaneous serial ordering of material that is within the focus of current attention.

In a typical dichotic listening experiment, participants are presented with items to both ears simultaneously but are required to attend to only one stream. Bryden (1971) showed that when adult participants were required to freely recall from the attended ear (with four items presented to each ear), recall was equally good at all serial positions, with high probability of spontaneous serial order output. However, when participants were required to recall from the unattended ear (whether before or after the attended items), recall followed a steep recency curve, with a clear advantage for the final item. This suggests that the attended items were held within primary memory, and that the unattended items were held within a separate store, such as perceptual memory (cf. Bhatarah, Ward, & Tan, 2006 ; Broadbent, 1958 ; Lachter, Forster, & Ruthruff, 2004 ). Indeed, findings from earlier, similar, dichotic listening studies formed the basis of Broadbent’s (1958 ; see also Sperling, 1967 ) theory of perception and memory, as the characteristics of recall from different streams evidenced separate stores (in his terms, S-system [sensory] and the P-system [limited capacity channel]). A focus on the processes involved in primary memory, therefore, provides a means of determining which items are held within primary memory and which are held within a separate system. For example, there are individual differences in performance on dichotic listening tasks, with adult participants who have low to average digit spans showing the pattern observed by Bryden (1971) , while participants with higher spans show equally good recall across items heard in either ear ( Parkinson, 1974 ). This suggests that successful recall from an unattended stream is possible when the number of attended items does not exhaust primary memory capacity (see also Colflesh & Conway, 2007 , for links between working memory capacity and unshadowed speech perception).

We therefore developed a new selective free-recall memory task, drawing on the logic of these dichotic listening studies, to provide a potential index of primary memory in our sample. This involved sequentially presented verbal items, rather than simultaneous presentation of items to each ear, in order to make the task suitable for use with young children

Measuring Primary Memory Capacity Using Probed Free Recall

Waugh and Norman (1965) and others (e.g., Murdock, 1968 ) have used recall of the final list items in immediate free recall to estimate primary memory capacity. However, a potential problem with this approach is that it relies on participants beginning their recall with these list-final items, which is not always the case (see Howard & Kahana, 1999 ). One way of enforcing this response pattern is by setting memory probes at various points from the end of the list. For example, Waugh and Norman (1965) gave participants a probe digit to request recall of the last two, three, four, or five items of a 10-item list and examined the difference in recall of these probed items versus items from the earlier section of the list. Raymond (1969 ; see also Aslan, Bäuml, & Grundgeiger, 2007 ; Murdock, 1968 ; Waugh & Norman, 1965 ) suggested that once any one set of memory items has been probed from a section of a list, the resulting output interference from that recall set would result in all subsequently probed items from the same list being drawn from secondary memory. Therefore, there is a potential difference in the characteristics of items that are recalled from the last section of a list and those that are recalled from the first section (cf. Tulving & Colotla, 1970 ), particularly when the last section is probed first. In addition, if items from the last section of a list are maintained in primary memory, one would expect a high degree of spontaneous serial ordering of these items when the last section of the list is probed first. The difference in performance on items recalled from the last section when probed first, as opposed to when probed second, can therefore potentially index which items are drawn from primary memory.

This assumption formed the basis of our second new measure, the split span task, which was again piloted prior to use in the current experiment. The key findings from the pilot data were that items from the last section of the list, when probed first, were recalled more accurately and with a greater degree of spontaneous serial ordering than were either items from the last section when probed second, or items from the first section whether probed first or second. There was also some evidence of an increase in primary memory capacity in older as opposed to younger children, and a clear increase in the likelihood of older children starting their recall from the start of a probed list (these data can be found in the Supplementary Materials). These data suggested that it was appropriate to assume that items from the “last” portion of the list, when probed first, were drawn from primary memory.

Estimating Primary Memory From Free Recall

An alternative method for estimating primary memory from immediate free recall follows from recent work by Ward, Tan, and Grenfell-Essam (2010) . These authors tested adults on free recall of unpredictable list lengths of between one and 15 items. When participants recalled lists of one to five items, there was no large change in probability of item recall across serial positions. On longer lists, the serial position curve became clearly bowed, with increasingly steep recency portions of the curve from six-item lists onward. In addition, Ward et al. found that participants were likely to begin recall with the first item on the list until lists exceeded five items in length, at which point they were more likely to begin recall with items from the end of the list. One reading of these data (cf. Farrell, 2010 ), is that the capacity of primary memory can be referenced from immediate free-recall data both by the point at which the serial position curve becomes nonflat, and the point at which a participant stops recalling the first list item first as list length increases. 1 Bowing and flatness of serial position curves are unlikely to be observed fully using ISR span methods, and using varying list lengths in a free-recall methodology allows us a greater insight into cognitive processes underpinning performance at short and long list lengths (see Gibson et al., 2013 ; Unsworth & Engle, 2006 ). In the current work, we therefore examined these characteristics using Ward et al.’s (2010) varying list length methodology with children.

The Current Work

The present experiment was designed with three main objectives in mind, which combined methodological and theoretical issues: (a) to triangulate performance on our three novel measures of primary memory capacity to determine whether a consistent estimate of primary memory could be achieved in children, (b) to examine the developmental change in these measures so as to derive potential indices of primary memory from the same set of tasks in two age groups of children, and (c) to test whether these novel measures of primary memory capacity can better and/or separately predict storage contributions to working memory tasks and academic achievement than traditional ISR-based span measures of STM.

Following the previous discussion, we assumed that the defining characteristics of primary memory are (a) flat, comparable performance across items in the serial position curve (cf. Unsworth & Engle, 2006 ); and (b) evidence that participants begin recall (probability of first recall) with the first item on the list, which we take as a marker of an intention to recall in forward serial order ( Broadbent, 1958 ; Farrell, 2012 ; Sperling, 1967 ; Ward et al., 2010 ). Given this, and to properly examine the extent to which participants spontaneously elected to recall the items in serial order, all three novel tasks were based on free-recall methodologies. In addition, list length was unknown prior to presentation for any trial. This manipulation was put in place to ensure that children could not selectively attend to list items, with a view to minimizing strategic contributions to primary memory estimation which may be a further potential issue with ISR span tasks ( St. Clair-Thompson, 2007 , though see Jarrold & Hall, 2013 ).

To that end, in this experiment modified versions of the previously piloted interleaved lists and split span tasks were used alongside a free-recall task with unpredictably varying list lengths, as well as traditional simple span and complex span tasks. Standardized assessments of reading and mathematics were also given to participants, who were a cross-sectional sample of children in the early primary school years. Specifically, groups of Year 1 (aged 5 to 6) and Year 3 (aged 7 to 8) children were tested. Evidence for a developmental change in primary memory capacity was expected to be observed in each of the new memory measures in this study, as indexed by the average number of items recalled in a task and the probability of recalling the first item first (as an index of spontaneous serial ordering).

Design

Each child completed three individual testing sessions lasting approximately 30 min each. In each of the first two sessions, children completed two memory tasks, and in the final session, they were tested on one memory task; these tasks were presented to all children in the order in which they are introduced next. In addition to the memory measures, all children were tested on the Sentence Completion Forms of the NFER-Nelson (1998) Group Reading Test II Form A (6–14) and the age-appropriate test from the NFER-Nelson (1994) Mathematics 6–14 series in separate sessions. The Group Reading Test spans a wide age range and was administered to all children in both year groups. The Mathematics 6–14 test uses different test questions dependent on children’s level of education, and the appropriate tests were given to each age group (Progress in Maths 6 was given to Year 1, and Progress in Maths 8 was given to Year 3 children). Both reading and mathematics assessments give a fairly broad overview of skill in each area. The reading assessment indexes word reading and sentence comprehension, and the same questions are given to both age groups. The mathematics assessment taps proficiency in facts and procedures, concepts, problem solving, and reasoning, with identical subscales for both age groups.

Tasks and Procedure

All memory tasks were programmed using Runtime Revolution software and presented on Macintosh Powerbook and MacBook computers. A total of 348 words were used in the memory tasks, which were single syllable concrete nouns, with age of acquisition of under 6.2 years (statistics from the Medical Research Council database, Wilson, 1988 ). Each word was paired with a color cartoon image. No words were repeated within or between tasks in a single testing session. All audio material was presented through the internal laptop speakers using male voices.

Children were presented with increasing lists of words (two to eight) with five trials at each list length. If children successfully remembered any one list in correct serial order from the five trials at a given list length, they moved on to the next list length. If they failed to recall all lists within a given list length, testing was terminated at that point. The predictable list lengths and continuation rules used in this study enabled a direct comparison of the span tasks against existing literature on the link between working memory and academic achievement (see, e.g., Bayliss et al., 2003 ). Children were presented with a digital audio recording of words in a male voice and 3-cm-high color illustrations of those words for 1,000 ms in the center of the computer screen. A blank screen appeared briefly between each word. At the end of each trial, a cartoon giraffe appeared alongside a question mark and children were prompted to recall the words in the order they had heard them.

Complex span

This task followed the span procedure used in the simple span task (with the same number of trials at each of the same list lengths, and the same continuation and stopping criteria), but using digits. Digits were presented in a male voice as digital audio recordings, simultaneously with the appearance of the item in black in the center of the computer screen measuring approximately 2 cm high for 1,000 ms. Between each digit, children were presented with a large colored circle measuring around 4 cm high (either brown, pink, or blue) in the center of the computer screen and told to name the color of the circle as quickly as they could, similar to a complex span task used by Camos and Barrouillet (2011) . As soon as the participant had named the circle, the experimenter tapped the spacebar of the computer, and the computer moved on to another colored circle. This processing task automatically ended after 3,000 ms, regardless of the number of circles that the child had named in that time. This task was therefore designed to fill the fixed 3-s processing window with near-continuous verbal distraction. The child was then presented with the next digit in the list. A cartoon dinosaur with a question mark over his head appeared when the participant was required to begin recall, and they were told to recall the digits in the order they had heard them.