In immediate serial recall, participants are presented with series of stimuli and are asked to recall them. Direction of recall is known to be an important determinant of performance, with participants typically achieving higher scores when recalling items in their original (forward) order, relative to reverse or backward order (e.g., Li & Lewandowsky, 1995; St Clair-Thompson, 2010; but see Anderson, Bothell, Lebiere & Matessa, 1998). Studies have shown both primacy (advantage for early list items) and recency (advantage for late list items) effects for forward recall but minimal primacy and steeper recency for backward recall (e.g., Bireta, Fry, Jalbert, Neath, Suprenant, Tehan & Tolan, 2010; Li & Lewandowsky, 1995). Recall direction also interacts with the prevalence of traditional short-term memory effects, including those of word length, irrelevant speech, phonological similarity, and concurrent articulation (e.g., Bireta et al., 2010). Evidence suggests that the effects are either absent or greatly attenuated when participants are asked to recall items in reverse order (e.g., Bireta et al., 2010; Madigan, 1971; Tehan & Mills, 2007).

These studies of immediate serial recall have employed various stimuli, including letters (e.g., Li & Lewandowsky, 1995), words (e.g., Bireta et al., 2010; Tehan & Mills, 2007), and digits (e.g., Anderson et al., 1998; St Clair-Thompson, 2010). In the present study, we focus on the recall of digits, because forward and backward digit recall form an integral part of all Wechsler Intelligence Scales and the Wechsler Memory Scales (Wechsler, 1955, 1981, 1997). These are among the most commonly used measures in psychological research and clinical evaluation. However, many theoretical approaches to short-term or working memory (e.g., Baddeley, 2000) assume a domain-specific store for verbal information, including letters, words, and digits. There is no a priori reason to suspect that different patterns of findings would emerge for different types of verbal stimuli.

Several approaches have been taken to account for differences between forward and backward recall. One dominant view explains the differences in terms of attentional demands. According to this view, both forward and backward recall employ short-term phonological storage (i.e., short-term memory). However, backward recall is also considered to require an attention-demanding transformation of the digit sequence, thus classifying this task as a complex span measure of working memory (e.g., Alloway, Gathercole & Pickering, 2006; The Psychological Corporation, 2002, p. 6). Consistent with this suggestion, studies have revealed that backward recall loads onto the same factor as working memory measures such as counting span and listening span, whereas forward recall loads onto to a separable short-term memory factor (e.g., Alloway et al., 2006; Alloway, Gathercole, Willis & Adams, 2004; Gathercole, Pickering, Ambridge & Wearing, 2004). Backward digit recall has also been found to be more sensitive to the effects of aging and brain dysfunction than forward digit recall (The Psychological Corporation, 2002, pp. 201–202), with this often attributed to the involvement of executive control (e.g., Reynolds, 1997). In addition, Gerton, Brown, Meyer-Lindenberg, Kohn, and Holt (2004) used PET to observe greater bilateral activation in the dorsolateral prefrontal cortex during backward, relative to forward, digit recall. It should be noted, however, that any explanation based on phonological representation would predict effects of related manipulations (e.g., phonological similarity) in both recall directions (Rosen & Engle, 1997), a prediction for which there has been some contrasting evidence (e.g., Bireta et al., 2010; Guerard, Saint-Aubin, Burns & Chamberland, 2012).

A second approach assumes that differences result from two different types of representation, with forward recall being more suited to a phonological code and backward recall to a visuospatial code. For example, Li and Lewandowsky (1995) found that backward recall, but not forward recall, was impaired by presenting study items in random spatial locations. Intralist visual similarity was also beneficial to backward, but not forward, recall. Consistent with these findings, there is also neuroimaging evidence for the involvement of visuospatial processes in backward recall (e.g., Hoshi, Oda, Wada, Ito, Yamashita, Oda,… Tamura, 2000; see also Gerton et al., 2004). Although this suggestion may seem to contradict multistore models of memory (e.g., Baddeley, 2000), which assume that information is stored in domain-specific subsystems, there is evidence that recall of verbal information can be improved by the use of visual imagery (e.g., De La Iglesia, Buceta & Campos, 2005), visual representation of a number line (e.g., Dehaene, 1992), or the presentation of digits within familiar visuospatial configurations (Darling, Allen, Havelka, Campbell & Rattray, 2012; see also Mate, Allen & Baques, 2012). It is possible that such visual strategies are employed more so for backward than for forward recall because they can assist with the transformation of the digit sequence.

Other researchers, however, consider forward and backward recall to assess the same cognitive resources. For example, Rosen and Engle (1997) found that forward and backward recall did not differ in terms of predicting performance on standardized tasks. Some studies have also revealed that forward and backward recall load onto the same factor during factor analysis (e.g., Colom, Abad, Rebello & Shih, 2005; Engle, Tuholski, Laughlin & Conway, 1999).

Research investigating the cognitive underpinnings of backward digit recall has therefore yielded mixed results. The present study aimed to further elucidate the cognitive resources involved in backward digit recall using an experimental, dual-task approach. We began by employing an adaptation of the n-back technique (e.g., Jaeggi, Buschkuehl, Perrig & Meier, 2010; Jonides et al., 1997; Owen, McMillan, Laird & Bullmore, 2005), which involves asking participants to track a series of visuospatial stimuli at a lag specified by the parameter n. Participants were tested at 0-back (essentially, simple visuospatial shadowing) and 2-back, as well as under baseline (no-task) conditions. Comparing baseline and 0-back enables an examination of the role of visuospatial processing in backward digit recall. In turn, a comparison of performance under conditions of 0-back and 2-back reveals the extent to which executive control is important. Two-back tasks have previously been shown to be highly demanding of executive resources (e.g., Owen et al., 2005). Furthermore, visuospatial 2-back has been effectively used as an executive-based concurrent task during encoding, showing greater disruptive effects on verbal working memory than 0-back (Baddeley, Hitch & Allen, 2009). In the first experiment, participants completed visuospatial n-back tasks during the auditory presentation of digits to be recalled in either forward or backward order. In the second, the concurrent tasks were performed during the verbal recall stage. This approach allowed us to examine encoding and recall in turn and identify at which stage (if any) visuospatial processing and attentional control become more important in backward recall.

In three additional experiments, we further investigated the role of visuospatial resources in forward and backward recall. We did so by using dynamic visual noise (DVN). DVN was developed by Quinn and McConnell (1996) as a procedure that can interfere with the encoding, maintenance, and retrieval of purely visual information in working memory. It is known to influence visual imagery, but not tasks involving visual memory (e.g., Andrade, Kemps, Werniers, May & Szmalec, 2002; Dean, Dewhurst & Whittaker, 2008). This therefore allowed us to examine the role of visual imagery in backwards recall. In Experiment 3, DVN was presented during the presentation of digits to be recalled in forward or backward order, and in Experiment 4, DVN was presented during recall. Again, this approach allowed us to examine at which stage (if any) visual processes become more important in backwards recall. In Experiment 5, we then replicated Experiment 4 but used the same list length for forward and backward recall.

Experiment 1

Method

Participants

Thirty undergraduate students took part in the study, receiving course credits. Their mean age was 21 years and 8 months (SD = 9 months).

Materials

Sequences were constructed of seven digits for forward recall and six digits for backward recall conditions. In all cases, digits did not repeat within a sequence. Different list lengths were selected for forward and backward recall due to better performance on forward recall tasks (e.g., Engle et al., 1999; St Clair-Thompson, 2010). Visuospatial n-back stimuli consisted of a 3 × 3 grid square, with each of the nine location squares measuring 4 × 4 cm, and a 2-cm diameter black circle occupying one of the locations on each trial. Stimulus presentation and recording of concurrent task responses was performed on a personal computer using E Prime software.

Design and procedure

This experiment followed a 2 × 3 repeated measures design, manipulating recall direction (forward, backward) and concurrent task (no task, 0-back, 2-back). All conditions were blocked, and order was counterbalanced between participants. Each participant was tested in a single session lasting approximately 1 h.

Each session started with 60 practice trials on each of the 0-back and 2-back tasks, performed in a counterbalanced order. On each trial, the black circle appeared in one of the nine locations and remained present for 2 s. The next trial followed immediately, with the circle appearing in one of the eight remaining locations, chosen at random. In the 0-back condition, participants had to respond by pressing a key on the 3 × 3 numerical keypad corresponding to the location presently occupied. In the 2-back condition, the task was to respond to the location occupied two trials previously. In this condition, participants were presented with two preparatory trials with the word “wait” appearing below the 3 × 3 grid. On the third trial, they began by responding to the location occupied on trial one, and so on for subsequent trials.

The baseline (no concurrent task) condition was divided into two phases, with five forward and five backward recall trials performed at the start and a further five trials following completion of the concurrent task conditions. Ten trials for each of the four concurrent task conditions (forward recall with 0-back and 2-back and backward recall with 0-back and 2-back) were presented in between the two halves of the baseline condition. These four conditions were blocked and were fully counterbalanced across participants.

On each trial in the baseline condition, participants were informed of whether they had to recall items in forward or backward order, and then a keyboard press triggered presentation of the series of digits through speakers at either side of the computer. Participants attempted to verbally recall the sequence immediately following its completion, either in their order of presentation (forward recall) or in reverse order. In the concurrent task conditions, presentation of the series of digits was preceded by four concurrent task trials (plus two wait trials for the 2-back conditions). Participants were instructed to continue performing the n-back task as quickly and accurately as possible while listening to the series of digits being presented. The digits were again presented at the rate of one per second. Thus, two digits were presented during each n-back trial, and four concurrent task trials were completed during the presentation of each digit sequence (for backward recall, the last n-back stimulus was presented after the final digit). On completion of digit presentation the screen turned blank, and recall was attempted in either forward or backward order, as appropriate. Thus, no concurrent task was performed during recall. Participants were instructed to state “pass” if they were unsure of a digit, allowing them to continue with recall of the remaining digits in the correct list position. Responses in the primary task were recorded on a digital recording device. Scores for forward and backward digit recall were calculated as the proportion correct for each serial position (see, e.g., Bireta et al., 2010; Li & Lewandowsky, 1995). Since recall direction has previously been shown to interact with serial position, it is important to establish whether the impact of concurrent tasks on forward and backward recall varies with position in the sequence.

Results

Forward and backward digit recall

Figure 1 shows the mean proportion correct across all participants at each serial position for the six conditions. Forward recall is shown in the top panel, and backward recall in the bottom panel. Serial position refers to the presented sequence order; thus, for backward digit recall, serial position 1 is the final item to be recalled. Due to the different list lengths used for forward and backward recall, the serial positions were then collapsed into three groups—early, middle, and late—to allow for further analysis. For backward digit recall, the means from two serial positions were placed in each category, while for forward recall, there were three serial positions in the middle category.Footnote 1 A 2 (direction) × 3 (concurrent task) × 3 (serial position) ANOVA revealed a significant main effect of direction of recall, F(1, 29) = 127.99, MSE = 4.34, p < .01, μ 2 = .82, with higher scores for forward recall than for backward recall (means of .81 and .61, respectively). There was a significant main effect of concurrent task, F(2, 58) = 22.38, MSE = 1.20, p < .01, μ 2 = .44, with pairwise comparisons revealing significant differences between each concurrent task condition (p < .05 in each case). There was also a significant main effect of serial position, F(2, 58) = 109.22, MSE = 3.11, p < .01, μ 2 = .79, There was no significant interaction between direction of recall and concurrent task, F(2, 58) = 2.15, MSE = 1.64, p > .05, μ 2 = .04, or between concurrent task and serial position, F(4, 116) = 1.18, MSE = 0.94, p > .05, μ 2 = .04. There was a significant interaction between direction and serial position, F(2, 58) = 101.06, MSE = 1.17, p < .01, μ 2 = .78, resulting from diminished recency and extended primacy for backward recall. However, there was no significant three-way interaction between direction, concurrent task, and serial position, F(4, 116) = 0.38, MSE = 0.82, p > .05, μ 2 = .02.

Fig. 1
figure 1

Mean proportion correct at each serial position for the six conditions in Experiment 1. a Forward recall. b Backward recall

Full size image

N-back performance

Concurrent n-back performance was based only on trials recorded during digit presentation—that is, excluding the prepresentation trials. Performance and mean reaction times are displayed in Table 1. Using the percentage of correct responses, there was a significant main effect of n-back task, F(1, 29) = 29.82, MSE = 62.48, p < .01, μ 2 = .51, with accuracy being substantially poorer in the 2-back task than in the 0-back task (means of 88.83 and 96.71). There was no significant main effect of primary task—that is, direction of recall, F(1, 29) = 0.44, MSE = 42.24, p > .05, μ 2 = .02. There was also no significant interaction between n-back and primary task, F(1, 29) = 0.01, MSE = 34.59, p > .05, μ 2 = .00. For reaction times, there was a significant main effect of n-back task, F(1, 29) = 13.42, MSE = 10,832.28, p < .01, μ 2 = .32, with longer reaction times in the 0-back than in the 2-back task (means of 681.65 and 612.05 ms). There was no significant main effect of primary task—that is, direction of recall, F(1, 29) = 3.60, MSE = 8,901.18, p > .05, μ 2 = .10. There was also no significant interaction between n-back and primary task, F(1, 29) = 0.09, MSE = 3,911.30, p > .05, μ 2 = .00.

Table 1 Mean scores and reaction times on the n-back task
Full size table

Discussion

The results revealed poorer performance on backward digit recall than on forward digit recall. However, the results also suggest that these differences did not arise due to differential involvement of visuospatial or executive-attentional resources. Performance was poorer in the 2-back conditions than in the 0-back and no concurrent task conditions, presumably because 2-back requires a greater degree of executive control than does 0-back (e.g., Baddeley et al., 2009; Owen et al., 2005). However, backward digit recall was no more affected by the concurrent tasks than was forward digit recall. It is also important to note that differences between forward and backward digit recall were not reflected in performance on the concurrent tasks. Participants made more errors in the 2-back conditions than in the 0-back conditions but also responded more quickly. The latter finding was to be expected because, in the 2-back task, participants could prepare their next response during the 2 s that the stimulus remained on screen. However, there were no differences in performance on the n-back tasks between the forward and backward recall conditions.

The results of the experiment therefore suggest that backward digit recall is not relatively more demanding of visuospatial or executive-attentional resources than is forward recall. In Experiment 1, however, n-back tasks were presented only during encoding of the digits to be remembered. Some studies have shown that prior knowledge of recall direction has no effect upon recall performance (e.g., Surprenant, Bireta, Brown, Jalbert, Tehan & Neath, 2011), suggesting that differences between forward and backward recall are due to retrieval, rather than encoding, processes. It is also well-established that encoding and retrieval phases of memory are not disrupted to the same extent by concurrent performance of a secondary task (e.g., Fernandes & Moscovitch, 2000; Naveh-Benjamin & Guez, 2000). In particular, one possible strategy during backward digit recall is to delay the transformation of the digit sequence until all of the items have been presented (e.g., St Clair-Thompson, 2010). If this were the case, completing the n-back task during encoding would not be expected to be detrimental to performance. In contrast, performing the n-back task during retrieval would allow for any effects of the n-back tasks on sequence transformation to be observed. Experiment 2 was conducted to explore this possibility.

Experiment 2

Method

Participants

Thirty-seven undergraduate students took part in the study, receiving course credits. Their mean age was 20 years and 9 months (SD = 8 months). None of the participants had taken part in Experiment 1.

Materials, design, and procedure

The tasks, structure of testing, and scoring procedures were the same as those used in Experiment 1, with one exception. This time, participants completed the 0-back and 2-back tasks during recall of the digits. Thus, digit sequences were presented in isolation, without concurrent load. In the no concurrent task conditions, each series was followed by a 2-s delay and then a cue to recall the sequence. In the 0-back and 2-back conditions, participants were instructed that they could begin recalling the sequence when the second n-back location was displayed. Thus, they had started responding to the 0-back task or started storing locations for later recall in the 2-back task. Therefore, as in the no concurrent task conditions, this procedure resulted in a delay of 2 s between presentation and recall of digits in each condition. Participants ceased responding to the n-back trials following recall completion and pressed the space bar to hear the next series of digits.

Results

Forward and backward digit recall

Figure 2 shows the mean proportion correct at each serial position for the six conditions. Forward recall is shown in the top panel, and backward recall in the lower panel. Again, due to the different list lengths used for forward and backward recall, the position data were collapsed into three groups: early, middle, and late. A 2 (direction) × 3 (concurrent task) × 3 (serial position) ANOVA revealed a significant main effect of direction of recall, F(1, 36) = 128.47, MSE = 6.16, p < .01, μ 2 = .78, with higher scores for forward recall than for backward recall (means of .84 and.61, respectively). There was a significant main effect of concurrent task, F(2, 72) = 29.56, MSE = 1.55, p < .01, μ 2 = .45, with pairwise comparisons revealing significant differences between each concurrent task condition (p < .05 in each case). There was also a significant main effect of serial position, F(2, 72) = 98.84, MSE = 4.08, p < .01, μ 2 = .73. Of particular interest to the present study, there was also a significant interaction between direction of recall and concurrent task, F(2, 72) = 10.74, MSE = 1.74, p < .01, μ 2 = .23. We explored this interaction further by conducting a one-way ANOVA comparing the concurrent task conditions for forward recall and a one-way ANOVA comparing the conditions for backward recall. For forward recall, there was a significant main effect of task, F(2, 72) = 4.62, MSE = 3.21, p < .05, μ 2 = .11, and pairwise comparisons revealed significant differences between the no concurrent task and 2-back task conditions (p < .05). There was no significant difference between the no concurrent task and 0-back conditions or between the 0-back and 2-back conditions (p > .05 in each case). For backward recall, there was a significant main effect of task, F(2, 72) = 26.92, MSE = 6.73, p < .01, μ 2 = .43, with pairwise comparisons showing significant differences between the no concurrent task and 0-back conditions and between the no concurrent task and 2-back conditions (p < .01). There was no significant difference between the 0-back and 2-back conditions (p > .05). There was also a significant interaction between direction and serial position, F(2, 72) = 77.26, MSE = 1.56, p < .01, μ 2 = .68. This resulted from a diminished primacy effect for backward recall. There was no significant interaction between task and position, F(4, 144) = 1.59, MSE = 0.81, p > .05, μ 2 = .04, and no significant three-way interaction between direction, concurrent task, and serial position, F(4, 144) = 0.99, MSE = 0.77, p > .05, μ 2 = .03.

Fig. 2
figure 2

Mean proportion correct at each serial position for the six conditions in Experiment 2. a Forward recall. b Backward recall

Full size image

N-back performance

Performance and mean reaction times for the n-back tasks are displayed in Table 1. Using the percentage of correct responses, there was a significant main effect of n-back task, F(1, 36) = 25.13, MSE = 292.28, p < .01, μ 2 = .41, with accuracy being substantially poorer in the 2-back task than in the 0-back task (means of 63.15 and 71.68). There was also a significant main effect of primary task—that is, direction of recall, F(1, 36) = 20.46, MSE = 131.34, p < .01, μ 2 = .36, with accuracy being poorer for backward recall than for forward recall (means of 63.15 and 71.68, respectively). There was no significant interaction between n-back and primary task, F(1, 36) = 0.92, MSE = 124.85, p > .05, μ 2 = .03. For reaction times, there was a significant main effect of n-back task, F(1, 36) = 8.83, MSE = 33,823.32, p < .01, μ 2 = .20, with longer reaction times in the 0-back than in the 2-back task (means of 883.28 and 793.42 ms). There was a significant main effect of primary task—that is, direction of recall, F(1, 36) = 7.88, MSE = 25,896.56, p < .01, μ 2 = .17, with shorter reaction times for forward recall than for backward recall (means of 801.23 and 875.47 ms). There was no significant interaction between n-back and primary tasks, F(1, 36) = 0.50, MSE = 6,745.28, p > .05, μ 2 = .01.

Discussion

As in Experiment 1, performance was significantly poorer on backward digit recall than on forward digit recall. Furthermore, this experiment did reveal a significant interaction between recall direction and concurrent task. Further analysis demonstrated that this was due to effects of both visuospatial tasks (0-back and 2-back) on backward recall, relative to baseline (no-task) conditions, but no difference between 0-back and 2-back conditions (illustrating executive control). In contrast, for forward recall, there was only an effect of the 2-back task. This suggests that differences between forward and backward recall occur during recall, rather than during encoding of the items to be remembered (Experiment 1), and also that backward recall may have involved a degree of visuospatial resources.

Further differences between forward and backward recall were reflected in performance on the concurrent tasks. Performance on the n-back tasks was poorer during backward recall than during forward recall. This provides some evidence for additional cognitive resources being employed for backward recall, which then compromises performance on the n-back tasks. Importantly, however, there was no significant interaction between n-back task and direction of recall. Therefore, backward digit recall did not differentially influence performance on the 2-back task, relative to the 0-back task. The additional resources being employed for backward recall therefore do not appear to be executive in nature and, rather, appear to be common to both n-back tasks. This finding therefore provides further evidence for the role of visuospatial resources in backward recall.

The results therefore support previous suggestions that backward recall, but not forward recall, relies partially on visuospatial processing (e.g., Li & Lewandowsky, 1995). It could be the case that when participants are required to recall a sequence in reverse order, they use a visuospatial representation of the sequence from which they can read off the items to recall. However, the equivalent effects of concurrent tasks performed during encoding on forward and backward recall that were observed in Experiment 1 would appear to suggest that participants do not differentially use a visuospatial representation during encoding for forward or backward recall. Therefore, it may be the case that digit encoding and storage are primarily phonological in nature, with some form of visuospatial mental rotation employed at recall in order to reverse the sequence.

It is, however, important to note that although, in the present study, the interaction indicating the involvement of visuospatial resources was significant, the effect size was not large. In addition, the requirements of the visuospatial n-back task make the findings of visuospatial interference somewhat difficult to interpret. For example, when dual-task paradigms are used, there is always a concern about trade-offs in performance. In Experiment 2, accuracy on the n-back task was somewhat poorer during backward recall than during forward recall, and participants also responded more slowly to the n-back task during backward recall. In Experiment 1, there were no differences in accuracy between the forward and backward recall conditions. However, the difference in reactions times between the forward and backward recall conditions was nearing significance. Thus, in Experiment 1, there may have been a hint of a small effect of performing concurrent tasks during encoding on backward digit recall.

It is also worthy of note that the n-back task and both forward and backward recall share a requirement for serial order. It is, however, unlikely that the differential pattern of interference emerged because the n-back task involved a form of backward recall. In the n-back task used in the present study, participants were essentially recalling a spatial sequence in the same order that it was presented, at a lag of either 0 or 2 items. Thus, the task could be considered to be more similar to forward than to backward recall. Previous research comparing n-back with forward and backward recall, and simple and complex span, also reject the notion that the findings in Experiment 2 reflect increased similarity in basic task demands between n-back and backward recall (e.g., Gevins & Smith, 2000; Jaeggi et al., 2010; Miller, Price, Okun, Montijo & Bowers, 2009; Oberauer, 2005; Shelton, Elliott, Hill, Calamia & Gouvier, 2009). Thus, we would claim that the results indicate a greater role for visuospatial support during backward digit recall. Nevertheless, it is important to establish whether the same patterns of differential interference emerge using a different concurrent manipulation that controls possible trade-off and ordering issues. We therefore conducted further experiments to examine the role of visuospatial resources in backward digit recall. We did so by using DVN.

DVN involves presenting patterns consisting of small black and white squares that randomly switch color over time, thus minimizing spatial and temporal components. It was established by Quinn and McConnell (1996; McConnell & Quinn, 2000) to allow selective interference with visual short-term memory. Quinn and McConnell found that DVN had minimal effects on storage of word lists when participants were instructed to use a rote rehearsal strategy but caused substantial disruption when visual imagery of the same items was required. Subsequently, a growing body of evidence has indicated that DVN can interfere with the encoding, maintenance, and retrieval of certain forms of purely visual information (e.g., Andrade et al., 2002; Dean et al., 2008), but not with spatial information (e.g., Darling, Della Sala & Logie, 2007, 2009; Dent, 2010). Furthermore, the passive nature of DVN interference and the selective impact on visual imagery (and not verbal or spatial working memory) indicate that this manipulation places no demands on modality-independent central executive resources. Any disruptive impact of DVN can therefore be attributed to visual processing, and not a more general attentional contribution, or to any requirements of serial ordering also shared by the primary recall task. If participants do indeed employ visual resources during backward recall—for example, to assist with the transformation of the digit sequence—we would expect DVN to interfere with backward recall, relative to forward digit recall. In Experiment 3, DVN was presented alongside forward and backward recall as concurrent interference during encoding.

Experiment 3

Method

Participants

Twenty-five undergraduate students took part in the study, receiving course credits. Their mean age was 22 years and 0 months (SD = 9 months). None of the participants had taken part in Experiment 1 or 2.

Materials, design, and procedure

Participants completed ten trials in each of four conditions (forward recall with no concurrent task, backward recall with no concurrent task, forward recall with DVN during encoding, and backward recall with DVN during encoding). The baseline (no concurrent task) conditions were divided into two phases, with five forward and five backward recall trials performed at the start and a further five trials following completion of the concurrent task conditions. The concurrent task conditions were blocked and were fully counterbalanced across participants.

On each trial in the baseline condition, a keyboard press triggered presentation of the series of digits through speakers at either side of the computer. Participants attempted to verbally recall the sequence immediately following its completion, either in its order of presentation (forward recall) or in reverse order. In the concurrent interference conditions, a keyboard press triggered both the presentation of the digit sequence and presentation of DVN. The DVN consisted of a grid of 80 × 80 cells, each measuring 2 × 2 pixels. At any one time, 50 % of the pixels were white, and 50 % were black. The pixels changed at random at a rate of 50 % per second. The DVN was presented as an 8- or 7-s movie for forward and backward recall, respectively. The digit sequences began after 1 s, with one digit per second. Participants were instructed to look at the computer screen throughout the entire task. On completion of digit presentation, the screen turned blank, and recall was attempted in either forward or backward order, as appropriate. Thus, no DVN was presented during recall. Scores for forward and backward digit recall were calculated as the proportion correct for each serial position.

Results

Figure 3 shows the mean proportion correct at each serial position for the four conditions. Forward recall is shown in the top panel, and backward recall in the lower panel. Again, due to the different list lengths used for forward and backward recall, the position data were collapsed into three groups: early, middle, and late. A 2 (direction) × 2 (concurrent interference) × 3 (serial position) ANOVA revealed a significant main effect of direction of recall, F(1, 24) = 31.06, MSE = 7.42, p < .01, μ 2 = .56, with higher scores for forward recall than for backward recall (means of .82 and.64, respectively). There was no significant main effect of interference, F(1, 24) = 0.42, MSE = 0.39, p > .05, μ 2 = .02. There was a significant main effect of serial position, F(2, 42) = 57.01, MSE = 1.98, p < .01, μ 2 = .70.. However, there was no significant interaction between direction of recall and interference, F(1, 24) = 0.03, MSE = 0.05, p > .05, μ 2 = .00. There was a significant interaction between direction and serial position, F(2, 48) = 81.27, MSE = 2.20, p < .01, μ 2 = .77. This resulted from a diminished primacy effect for backward recall. There was no significant interaction between interference and position, F(2, 48) = 0.14, MSE = 0.66, p > .05, μ 2 = .01, and no significant three-way interaction between direction, interference, and serial position, F(2, 48) = 2.11, MSE = 0.64, p > .05, μ 2 = .07.

Fig. 3
figure 3

Mean proportion correct at each serial position for the four conditions in Experiment 3. a Forward recall. b Backward recall

Full size image

Discussion

The results of Experiment 3 again revealed poorer performance on backward digit recall than on forward digit recall. However, backward digit recall was no more affected by DVN than forward digit recall. The results are therefore consistent with Experiment 1, which suggested that backward digit recall is not relatively more demanding of visuospatial resources than is forward digit recall. The findings of Experiment 2, however, suggested that backward recall was more demanding of visuospatial resources during the recall phase, rather than during encoding of the items to be remembered. If this is the case, we would not expect DVN to interfere with backward digit recall when presented during encoding, but rather we would expect it to interfere when presented during recall. Experiment 4 explored this possibility.

It is, however, worthy of note that in Experiment 2, although the interaction indicating the involvement of visuospatial resources in backward recall was significant, the effect size was not large. There are a number of possible reasons for this finding. One is that the extent of visuospatial involvement is dependent upon participant’s strategy use. Visuospatial imagery may be just one successful strategy for backward recall (e.g., Hoshi et al., 2000), with some participants preferring verbal strategies such as articulatory rehearsal. It seems reasonable to assume that participants who employ visual strategies may be more affected by a concurrent visual task or DVN-based interference. Therefore, in addition to exploring the effects of DVN during recall of digits, in Experiment 4 participants were also asked to report the strategies they employed during backward recall, with these reports later classified as involving either visual or nonvisual processes. The effects of DVN during recall of digits were examined overall and then explored separately in two groups defined on the basis of strategic approach.

Experiment 4

Method

Participants

Thirty undergraduate students took part in the study, receiving course credits. Their mean age was 20 years and 5 months (SD = 9 months). None of the participants had taken part in Experiment 1, 2, or 3.

Materials, design, and procedure

The tasks, structure of testing, and scoring procedures were the same as those used in Experiment 3, with one exception. This time, participants were presented with DVN during recall of the digits. Thus, digit sequences were presented in isolation, without concurrent interference. In the no-interference conditions, each sequence was followed by a cue to recall the digits. In the DVN conditions, each sequence was followed by presentation of DVN, and participants were instructed that they could begin recalling the sequence as soon as the DVN was displayed. Participants stopped the DVN following recall completion and pressed the space bar to hear the next series of digits.

At the end of the testing session, participants were then asked to describe the strategies they had employed during backward digit recall. Responses were recorded on a digital voice recorder. These were then classified by an independent researcher, who was blind as to the aims of the study, as involving either visual or nonvisual processes. To be classified as using visual strategies, participants reported that they had engaged in practices such as “picturing the numbers in my mind” or “imagining the numbers written down on a piece of paper.”

Results

Figure 4 shows the mean proportion correct at each serial position for the four conditions. Forward recall is shown in the top panel, and backward recall in the lower panel. Again, due to the different list lengths used for forward and backward recall, the position data were collapsed into three groups: early, middle, and late. A 2 (direction) × 2 (concurrent interference) × 3 (serial position) ANOVA revealed a significant main effect of direction of recall, F(1, 29) = 120.33, MSE = 3.03, p < .01, μ 2 = .81, with higher scores for forward recall than for backward recall (means of .83 and.62, respectively). There was no significant main effect of interference, F(1, 29) = 1.98, MSE = 1.31, p > .05, μ 2 = .06. There was a significant main effect of serial position, F(2, 58) = 107.08, MSE = 1.30, p < .01, μ 2 = .78. Of particular interest to the present study, there was also a significant interaction between direction of recall and interference, F(1, 29) = 6.97, MSE = 1.22, p < .05, μ 2 = .19. Further analysis revealed a significant effect of DVN on backward recall, F(1, 29) = 4.15, MSE = 1.33, p = .05, μ 2 = .13, but not on forward recall, F(1, 29) = 1.80, MSE = 1.32, p > .05, μ 2 = .06. Returning to the main analysis, there was a significant interaction between direction and serial position, F(2, 58) = 118.41, MSE = 1.66, p < .01, μ 2 = .80. Again, this resulted from a diminished primacy effect for backward recall. There was no significant interaction between task and position, F(2, 58) = 2.80, MSE = 0.34, p > .05, μ 2 = .09, and no significant three-way interaction between direction, interference, and serial position, F(2, 58) = 0.47, MSE = 0.68, p > .05, μ 2 = .02.

Fig. 4
figure 4

Mean proportion correct at each serial position for the four conditions in Experiment 4. a Forward recall. b Backward recall

Full size image

Further analyses then examined the pattern of findings as a function of participant’s strategy use. Of the 30 participants, 11 were classified as having used visual strategies, with the other 19 being classified as using nonvisual strategies. A 2 (direction) × 2 (interference) × 3 (serial position) ANOVA was then conducted separately for each group of participants. The analyses revealed a similar pattern of findings for the two groups, with significant main effects of direction of recall and of serial position in each group (p < .01 in each case), but no significant main effects of interference (p > .05 in each case). However, regarding the interactions between direction of recall and interference condition, this was significant only in the group who had used visual strategies, F(1, 10) = 9.35, MSE = 0.49, p < .05, μ 2 = .48. In this group, DVN caused greater interference for backward recall (means of .68 and .60 for no concurrent task and DVN conditions, respectively) than for forward recall (means of .86 and .85). For the group who used nonvisual strategies, performance on backward digit recall was still impaired somewhat by DVN (means of .62 and .59 for no concurrent task and DVN conditions for backward recall and means of .79 and .80 for forward recall), but the interaction between direction of recall and interference condition was not statistically significant, F(1, 18) = 2.45, MSE = 1.68, p > .05, μ 2 = .12. In both groups, there were also significant interactions between direction and serial position, resulting from diminished primacy effects for backward recall, but no significant interactions between interference and serial position and no significant three-way interaction between direction, interference, and serial position.

Discussion

Consistent with the findings of Experiment 2, this experiment did reveal a significant interaction between recall direction and concurrent task. This provides further evidence that different processes are involved in forward and backward recall during the recall phase, rather than during encoding of the items to be remembered, and also that this process may involve a degree of visuospatial resources. The finding that backward recall is affected more than forward digit recall by DVN further suggests that the visuospatial processes involved in backward recall are at least in part visual in nature and may reflect a contribution of visual imagery (e.g., Andrade et al., 2002; Dean et al., 2008).

Experiment 4 also revealed that one important factor determining the role of visuospatial resources in backward digit recall is participants’ strategy use. When all of the participants were considered together, the interaction indicating the involvement of visual resources in backward recall was significant. However, the effect size was relatively small (.19). When the participants were separated into two groups on the basis of their strategic approach to backward recall, the interaction indicating the involvement of visuospatial resources was significant in the visual strategy group, with a large effect size (.48). However, it was not significant in the nonvisual strategy group (effect size of .12). This finding will be revisited in the General Discussion section.

It is, however, important to note that in the present experiment (and in Experiments 1, 2, and 3), different list lengths were used for forward and backward recall (in order to minimize risks of floor and ceiling effects constraining performance in each direction condition). It is possible that this use of different sequence lengths (seven-item sequences in forward recall and six in backward) may have influenced particular grouping or chunking strategies used in the two recall directions. Although there was no existing evidence leading one to suspect that this particular difference in sequence lengths would make backward recall any more amenable to visuospatial strategies, a further experiment was conducted to evidence a role for visuospatial resources in backward, but not forward, recall when the same list lengths are used.

Experiment 5

Method

Participants

Twenty-eight undergraduate students took part in the study, receiving course credits. Their mean age was 19 years and 2 months (SD = 8 months). None of the participants had taken part in Experiment 1, 2, 3, or 4.

Materials, design, and procedure

The tasks, structure of testing, and scoring procedures were the same as those used in Experiment 4, with one exception. This time, participants were presented with series of seven digits for recall in both forward and backward directions. Again, digit sequences were presented in isolation, without concurrent interference. DVN was then presented during recall. Thus, in the no-interference conditions, each sequence was followed by a cue to recall the digits. In the DVN conditions, each sequence was followed by presentation of DVN, and participants were instructed that they could begin recalling the sequence as soon as the DVN was displayed. Participants stopped the DVN following recall completion and pressed the space bar to hear the next series of digits.

Results

Figure 5 shows the mean proportion correct at each serial position for the four conditions. Forward recall is shown in the top panel, and backward recall in the lower panel. For consistency, again the data were collapsed into three groups: early, middle, and late. For both recall directions, three serial positions were in the middle category. A 2 (direction) × 2 (concurrent interference) × 3 (serial position) ANOVA revealed a significant main effect of direction of recall, F(1, 27) = 90.13, MSE = 5.89, p < .01, μ 2 = .77, with higher scores for forward recall than for backward recall (means of .84 and.59, respectively). There was a significant main effect of interference, F(1, 27) = 21.84, MSE = 1.48, p < .01, μ 2 = .45, with higher scores in no concurrent task than in concurrent task conditions (means of .74 and .68, respectively). There was a significant main effect of serial position, F(2, 54) = 127.67, MSE = 1.86, p < .01, μ 2 = .83. Of particular interest to the current study, there was also a significant interaction between direction of recall and interference, F (1,27) = 10.70, MSE = 1.04, p < .01, μ 2 = .28. Further analysis revealed a significant effect of DVN on backwards recall, F (1,27) = 23.02, MSE = 29.50, p < .01, μ 2 = .46, but not on forwards recall, F (1,27) = 2.97, MSE = 14.43, p > .05, μ 2 = .09. Returning to the main analysis, there was a significant interaction between direction and serial position, F(2, 54) = 146.01, MSE = 1.77, p < .01, μ 2 = .84. Again this resulted from a diminished primacy effect for backward recall. There was a significant interaction between task and position, F(2, 54) = 6.42, MSE = 0.46, p < .01, μ 2 = .19, indicating that lower performance in concurrent task conditions was restricted to early and middle serial positions, and a significant three-way interaction between direction, interference, and serial position, F(2, 54) = 7.72, MSE = 0.68, p < .01, μ2 = .22, indicating that lower performance in the DVN condition in the early and middle serial positions occurred only in the backward, and not in the forward, recall conditions.Footnote 2

Fig. 5
figure 5

Mean proportion correct at each serial position for the four conditions in Experiment 5. a Forward recall. b Backward recall

Full size image

Discussion

Consistent with the findings of Experiment 4, Experiment 5 revealed a significant interaction between recall direction and concurrent task. Participants performed significantly poorer on backward digit recall as a result of DVN, but there was no significant difference in performance on forward recall in the single and concurrent task conditions. This again suggests that different processes are involved in forward and backward recall during the recall phase and that these processes involve a degree of visuospatial resources. The differences between forward and backward recall that were observed in Experiments 2 and 4 were, therefore, not simply a consequence of longer list lengths being used for forward recall than for backward recall. In addition, Experiment 5 revealed a significant three-way interaction between direction, concurrent task, and list position, with larger DVN effects, only on backward recall, at the start and middle of presented lists (i.e., the end of the recalled sequence for backward recall). This would indicate that visuospatial resources are more important at this phase of the sequence when reverse recall is required. This will be discussed in more detail, and in the context of the prior experiments, in the General Discussion section.

General discussion

The aim of the present study was to investigate the cognitive underpinnings of backward digit recall. Experiments 1 and 2 revealed that, at least in adult participants, backward digit recall places minimal additional demands on executive-attentional resources, relative to those involved in forward recall. The basic effects of 2-back indicate a role for the central executive in encoding short sequences of verbal information (e.g., Aleman & van ’t Wout, 2008; Baddeley et al., 2009), but this was no larger for backward than for forward recall. Experiment 1 further revealed that performing a visuospatial task during encoding did not impair backward recall, relative to forward recall, suggesting that it does not particularly rely on visuo-spatial resources either. Experiment 2, however did provide some evidence for a role of visuospatial processing resources in backward recall. Since this effect was apparent only when concurrent load was added during recall (Experiment 2), and not during encoding (Experiment 1), this suggests that participants use visuospatial processes during backward recall during the response phase, rather than during digit encoding.

Experiments 3–5 further explored the role of visuospatial resources in backward recall by using DVN. Experiment 3 revealed that presenting DVN during encoding did not impair backward recall, relative to forward recall, suggesting that backward recall does not particularly rely on visuospatial resources during the encoding phase. However, Experiment 4 then revealed an interaction between direction of recall and concurrent task, which, consistent with Experiment 2, provides evidence for a role of visuospatial resources during the response phase of backward recall. This was replicated in Experiment 5, using equal sequence lengths for forward and backward recall.

These results support previous suggestions that backward, but not forward, recall tends to draw partially on visuospatial processing (e.g., Li & Lewandowsky, 1995). The finding that DVN interferes with backward recall further provides evidence that visual strategies are involved . DVN does not interfere with spatial information (e.g., Darling et al., 2007, 2009; Dent, 2010), and it is known to influence visual imagery, but not tasks involving visual memory (e.g., Andrade et al., 2002; Dean et al., 2008), verbal memory (Quinn & McConnell, 1996), or central executive resources. Therefore, it seems that the role of visual resources in backward digit recall may well reflect participant’s use of visual imagery during the recall phase of the task, in order to assist with transformation of the digit sequence. This is not to claim that visual imagery can only be used for backward and not forward recall. Instead, it appears that participants do not generally utilize such a strategy when required to recall digit sequences in their original order. It is, however, important to note at this stage that the exact role of visuospatial processes in backward recall is not well understood. It could be that participants create a visual image to perform some kind of mental rotation of the digit sequence, or form a visual image of the sequence and then retrieve items by scanning the image starting from the final item that was presented.

The three-way interaction (between direction, DVN, and list position) observed in Experiment 5 may shed some light on this. This indicated that DVN had a larger effect on backward recall at the start/middle of presented lists (i.e., the middle/end of recalled sequences). One possibility is that, for backward digit recall, participants rely on phonological memory to recall the final few items in the sequence (as they were heard most recently). However, as reversed recall progresses toward the start of the presented sequence, this becomes a more difficult and unreliable strategy to use, and participants instead switch to a visual-based code to reverse the digits. While this is clearly a tentative post hoc account, patterns of interference from Experiments 2 and 4 would fit with this, since they indicate somewhat larger effects in the same sections of the sequence. The three-way interactions in these experiments were not significant, but this may reflect the slightly shorter sequences used for backward recall than in Experiment 5. Thus, the final experiment replicated the critical interaction between direction and concurrent load from Experiments 2 and 4, and furthermore, the use of seven-item sequences enabled the three-way interaction to more clearly emerge in this experiment. Further research would, however, benefit from further examining the role of list length or task difficulty on the role of visuospatial processes in backward recall and from examining the effects on the likelihood of visuospatial processes being employed during other phases of the task—for example, during encoding as well as recall.

Experiment 4 also indicated that visual imagery is just one possible strategy that can be used during backward digit recall (see also Hoshi et al., 2000). Eleven of the 30 participants in Experiment 4 reported using visual strategies for backward recall. Furthermore, for these participants, there was a much greater decrement in backward recall, relative to forward recall, as a result of concurrent DVN. Participants who did not report using visual strategies commonly reported using the strategy of subvocal rehearsal, ceasing each rehearsal cycle when reaching the next digit to be recalled (see Li, Qin, Zhang, Jiang & Yu, 2012, for recent evidence of contribution of auditory-phonological short-term memory to backward digit recall). These participants did not show a significant decrement in backward recall as a result of DVN. This finding suggests that the variable use of strategies across participants may have been responsible for the relatively small effect size of the interaction indicating a role for visuospatial resources that was observed in Experiment 2. This finding also suggests, more generally, that it is important for researchers to consider strategy use during short-term and working memory tasks.

It is also worthy of note that the design of the present experiments may have minimized the impact of concurrent visuospatial tasks The present experiments used auditory presentation of digits (due to the concurrent tasks being visuospatial in nature). In contrast, previous studies examining backward digit recall (e.g., Bireta et al., 2010; Li & Lewandowsky, 1995) have used visual presentation of digits. The format used in the present study is likely to have increased the role for phonological coding, yet we still obtained some evidence for a role for visuospatial resources. Larger effects of concurrent visuospatial tasks may emerge with visual presentation, which may also encourage the use of visual codes and strategies (see Logie, Della Sala, Wynn & Baddeley, 2000). Further research is therefore needed to explore the role of visuospatial resources in backward recall using visual presentation.

These findings have implications for both theory and practice. They suggest that participants employ different processes, or strategies, during forward and backward recall (although not during encoding of these sequences). Models of immediate serial recall such as the primacy model (Page & Norris, 1998, 2003), assume that forward and backward recall are performed in the same manner. The findings of the present study suggest that such models require adjustment to explain backward recall, by assuming that participants often employ different strategies or representations depending on the direction of recall. The findings also fail to support the suggestion by Bireta et al. (2010) based on SIMPLE (Brown, Neath, & Chater, 2007) of differential item/order trade-offs between forward and backward recall. Under this approach, participants focus more on order processing in the temporal dimension during backward recall and are, thus, less affected by manipulations that impact on item than on order. However, while 0-back may involve an ordering requirement (although not to the same extent as 2-back, which did not show an interactive interference effect), DVN has no such load, with its impact previously shown to be independent of the serial ordering requirement of primary memory tasks (Darling et al., 2009; Zimmer, Speiser & Seidler, 2003). Instead, DVN directly affects visuospatial working memory representations (Dean et al., 2008). Therefore, the item/ order trade-off theory cannot account for the greater effects of DVN in backward than in forward recall. The results also suggest that researchers and practitioners need to consider the cognitive resources underlying forward and backward digit recall. It is reported that backward digit recall is a complex span task requiring executive resources (The Psychological Corporation, 2002, p. 6). However, the findings of the present study suggest that backward digit recall, for the most part, behaves more like a simple span task, with relatively minimal additional processing required only at the recall stage, and only as a result of visual strategies. Therefore, at least in young adults, comparison of forward and backward digit recall appears to not be an appropriate method of examining central executive resources and may, instead, at least partly reflect the strategic utilization of visual imagery.