A Meta-Analysis and Critical Review of Prospective Memory in Autism Spectrum Disorder

For time-based PM, a total of 118 participants with ASD and 118 NT control participants from six studies were included using a fixed-effects model. The weighted effect for the between-group difference in performance was −0.87 (SE 0.14, 95% CI −1.14 to −0.60; z = 6.38, p < .001). The homogeneity test was non-significant (Q = 1.22, p = .94) indicating that the variance across the included effect sizes was not greater than expected by sampling error. These findings indicate a large and consistent impairment of time-based PM in ASD across studies (see Fig. 3).

For event-based PM, a total of 311 participants with ASD and 287 NT control participants from 11 studies were included in the fixed-effects model. The weighted effect for the between-group difference in performance was −0.41 (SE 0.08, 95% CI −0.57 to −0.24; z = 4.83, p < .001). However, the test for homogeneity of effect sizes was highly significant (Q = 25.14, p = .005), which suggests that the variance among the included effect sizes was greater than expected by sampling error. Subsequently, the data was re-entered into a random-effects model that includes random effects variance (due to random differences between studies) in the weighting of the individual effect sizes in the model. This revealed a significant weighted effect of −0.43 (SE 0.13, 95% CI −0.69 to −0.17; z = 3.20, p < .01). Hence, both models indicate a small impairment of event-based PM in ASD, although the underlying effect sizes are heterogeneous across studies (see Fig. 4).

In order to interpret the results of case-control studies of PM ability with confidence several issues should be considered. Table 3 presents an overview of all studies included in the meta-analysis, as well as an indication of whether any of the potential methodological problems described below apply to them.

In the first study of event-based PM in ASD, Altgassen, Schmitz-Hübsch, and Kliegel (2010) compared 19 children/adolescents with ASD to 19 age- and ability-matched NT peers. The ongoing task tapped visuospatial working memory. In a study phase of this task, participants viewed a number of geometric shapes, and had to encode and store the configuration of shapes. After a short delay, a second set of geometric shapes appeared on the screen and participants had to decide whether the shape configuration was same or different to the first one (recognition phase). The background colour on which the shapes were presented changed randomly after each trial. For the PM component, participants were instructed to press a pre-specified keyboard key whenever they noticed a change in background colour to yellow. Participants performed the ongoing task alone for 10 trials (single-task block), followed by the PM condition (dual-task block). The results revealed no between-group differences in ongoing or PM task performance. The authors concluded that event-based PM, which depends on cued retrieval, is unimpaired in ASD.

Altgassen and colleagues were also the first to investigate time-based PM performance in individuals with ASD. Altgassen et al. (2009) assessed 11 children/teenagers with ASD and 11 age- and ability-matched NT control participants. The ongoing task required participants to perform a visuospatial working memory task similar to the one used by Altgassen et al. (2010). In the PM condition, participants were instructed to press a pre-specified keyboard key at 2-min intervals throughout the ongoing task. During this condition, participants could, at any time, bring up an on-screen clock that displayed the time elapsed by pressing a specified key. Importantly, the ongoing task was carried out twice - once as single-task block (ongoing-only condition), and once as dual-task block together with the PM instruction (PM condition). The ASD group performed significantly worse in the ongoing task during the PM condition, but not in the ongoing-only condition. More importantly, the results revealed significantly better PM performance, as well as a more adaptive time-monitoring curve, in NT children. Therefore, the authors concluded that the diminished PM performance in ASD might originate from difficulties with self-initiated processing, as reflected by a less-than-optimal pattern of time monitoring.

Despite some methodological concerns regarding Altgassen et al.’s (2009) study (see Williams et al. 2013), the conclusions drawn both from this study (that time-based PM is impaired in ASD) and the study by Altgassen et al. (2010) (that event-based PM is unimpaired in ASD) are supported by the results from two studies by Williams et al. (2013, 2014). Williams et al. (2013) examined both time-based and event-based PM in a sample of 21 children with ASD and 21 NT children matched on age, and IQ. Time-based vs. event-based PM were assessed separately as two within-subject conditions carried out within the context of a computer-based driving game (the ongoing task). The ongoing task required participants to collect tokens and avoid obstacles while driving down a road. For the time-based PM task, participants were told that their car had only a limited amount of fuel, which would run out after 80 s unless they remembered to refuel it. The fuel level could be monitored at any time by pressing a particular keyboard key, which caused a fuel gauge to be displayed on screen temporarily. Importantly refuelling was only possible after the fuel level dropped to a critical level (between 60 and 80 s). For the event-based PM task participants had to press a specific keyboard key whenever they passed a truck. Results revealed a significant Group (ASD/control) × Condition (event-based/time-based) interaction effect on PM task performance, reflecting preserved event-based PM performance but impaired time-based PM in the ASD group. More adaptive time monitoring (i.e. a greater number of fuel checks prior to the period where refuelling was possible) was related to fewer time-based PM failures in both groups. Importantly, groups did not differ in ongoing task performance, nor time-monitoring frequency or pattern.

The results of Williams et al. (2013) were replicated precisely in a subsequent study by Williams et al. (2014) of 17 adults with ASD and 17 age-, and verbal and performance IQ-matched NT adults. The same ongoing task (which tapped verbal short-term memory) was used for both PM conditions, which were carried out separately in counter-balanced order. In the ongoing task, participants studied sequences of seven words across 40 trials. After each study trial, a test list of seven words appeared on-screen and participants had to decide whether all seven had been present on the immediately-preceding study list. The event-based PM instruction required participants to press a specific key when one of the test list words represented a musical instrument. For the time-based PM task, participants had to press a specific key every two minutes throughout the ongoing task. Participants could bring up a clock displaying the elapsed time via key press. Williams et al. (2014) found a significant Group × Condition interaction, reflecting diminished time-based but spared event-based PM performance in the ASD group. Again, groups did not differ in ongoing task performance or time-monitoring frequency.

Firstly, Yi et al. (2014) studied the role of executive functioning in event-based PM in a sample of 25 children with ASD and two NT comparison groups. One comparison group was reported to be matched with the ASD group for chronological age (NT_CA, n = 25), whereas the other comparison group was reported to be matched with the ASD group for verbal mental age and nonverbal IQ, (NT_MA, n = 28). In Yi et al.’s paradigm, the ongoing task involved naming pictorial items on a series of cards. The PM task was to hand the experimenter a “target” card that had a red heart-shaped sticker on it. The ASD group performed significantly worse than both comparison groups on the PM task. Although this is an interesting study, there are two potential methodological issues with Yi et al.’s procedure that might lead to caution when interpreting the results. Firstly, ongoing task performance was not reported and memory for the PM task instruction was not checked. It is not clear whether participants with ASD either noticed the target sticker on the relevant cards or even that they encoded the instruction to hand the cards with such a sticker to the experimenter. Secondly, based on the data provided, we believe that the groups were not equated on baseline cognitive abilities. The NT_CA group was not equated for verbal mental age or nonverbal IQ, whereas the NT_MA group was not matched on nonverbal IQ or chronological age. Although the authors stated that the NT_MA group was matched with the ASD group for nonverbal IQ this was not accurate. Rather, the groups were matched for raw scores on the Combined Raven’s Matrices test of nonverbal ability, but not for the standardised scores (i.e., not for nonverbal IQ). Crucially, the standardised score among ASD participants was an average of 11 or 12 points below that of the comparison groups and was in the “below average” range (M = 79.17; SD = 21.83). In general, it is not clear why Yi et al. adopted this matching strategy. Ideally, case and control groups are matched for age and IQ, as other studies have shown is possible when investigating PM in ASD (see Williams et al. 2014).

The second such study also examined solely event-based PM in a similar age group. Brandimonte et al. (2011) studied event-based PM and response inhibition among 30 primary school-aged children with ASD, as well as 30 age, and full-scale IQ-matched and NT comparison participants. All participants completed a computerised ongoing task that involved sorting pictorial items into one of two categories (food and animals) via key press. The participants of each experimental group (n = 30 per group) were assigned to one of three between-subject conditions of the ongoing task. In a “PM condition”, n = 10 ASD and n = 10 NT participants completed the ongoing task as described, but had the additional requirement to press a particular keyboard key whenever pre-specified images appeared. In other words, participants had to encode and retain a PM instruction while completing the ongoing categorisation task (hence, this was a standard PM task). In a “response inhibition” condition, n = 10 ASD and n = 10 NT participants completed the ongoing task, but had the additional requirement to not respond (i.e., not make a categorisation judgement) when pre-specified images appeared. The requirements of this condition resemble a classic “Go/No-Go” task (Verbruggen and Logan 2008). Finally, in an “ongoing-only” condition, n = 10 ASD and n = 10 NT participants completed the ongoing task as described, but with no additional secondary requirements. This ongoing-only condition could be considered a control condition to establish how able participants with ASD are to perform the ongoing task independent of their PM or response inhibition skills. Brandimonte et al. (2011) performed two ANOVAs on their data. First, a 2 (Group: ASD/NT) × 3 (Condition: ongoing/PM/response inhibition condition) was conducted with ongoing task accuracy (percentage of correctly categorised images) as the dependent variable. There was a significant main effect of group, indicating that the comparison groups performed better than the ASD groups (\(\eta _{p}^{2}\) = 0.07, p = .05) in the image categorisation task.

Second, a 2 (Group: ASD/NT) × 2 (Condition: PM/response inhibition) ANOVA was conducted using secondary task accuracy (percentage of PM successes/percentage of responses correctly inhibited) as the dependent variable. In this ANOVA, there was a significant main effect of group (\(\eta _{p}^{2}\) = 0.10, p < .05) indicating better performance in both secondary tasks (PM and response inhibition) in the comparison group than in the ASD group. Crucially, the Group × Condition interaction was reported as non-significant and associated with a negligible effect size (\(\eta _{p}^{2}\) = 0.02, p = not reported). Despite the minimal interaction effect, Brandimonte et al. (2011) nonetheless broke it down using planned comparisons. These comparisons suggested that the ASD group was less accurate (\(\eta _{p}^{2}\) = 0.11, p < .05) and responded slower (\(\eta _{p}^{2}\) = 0.42, p < .01) than the NT group only in the PM task, whereas no between-group differences were evident in the Go/No-Go task (no effect sizes reported). Therefore, the authors concluded that individuals with ASD have a deficit in event-based PM, but are less affected in their response inhibition. However, there are some methodological issues that need to be addressed. First, breaking down non-significant interaction effects is problematic both statistically and conceptually, and may lead to erroneous conclusions being drawn (see Gelman and Stern 2006). Second, groups were clearly not equated for ongoing task performance and the difficulties with this aspect of the experiment for participants with ASD could account for their difficulties with the secondary tasks (i.e., PM and response inhibition). Finally, although Brandimonte et al. (2011) matched the participant groups for age and FSIQ overall, it is not clear that this was the case for each of the between-subjects experimental conditions (i.e. PM, inhibitory control and the ongoing-task-only conditions). It is quite possible (but not reported in the paper) that diagnostic groups for each between-group condition were not comparable on these baseline variables, which could have generated artificial group differences in PM performance.

Third and most recently, Sheppard et al. (2016) investigated the effect of autism symptom severity on event-based PM in 28 children with ASD and 26 NT controls. Severity of autism symptoms in the ASD group was assessed with the Child Autism Rating Scale (Schopler et al. 1980). Scores were used to divide the ASD sample into “severe ASD” and “mild ASD” groups (n = 14 each). In a game-like procedure participants interacted with a hand-puppet wolf (Wally, acted out by the experimenter) playing a distractor task game alternating with the completion of three different PM tasks, and a retrospective memory task. Thus, there was no ongoing task as such. The PM tasks required the child to remember to (a) clap when they heard music being played (two trials, PM clapping task); (b) remove a toy food item out of Wally’s view because he could not eat them (two trials, PM feeding task); and (c) collect their reward at the end of the session (signalled by the experimenter saying “The games are now finished, time to go back to class.”) (PM reward task). For the PM clapping and the PM reward tasks, participants were prompted once at the end of each trial if they did not spontaneously remember the PM activity (children were asked: “Can you hear the music?” for the PM clapping task, and “Have you forgotten anything?” for the PM reward task). Collapsed across all three PM tasks, a significant main effect of group (\(\eta _{p}^{2}\) = 0.13) was found. Post-hoc tests revealed that this was driven by the significant difference between the severe ASD and the NT group (d = 1.39). When examining the results separately, a similar pattern was found for the PM feeding task (\(\eta _{p}^{2}\) = 0.14). Further, although there were no group differences on the first trial of the PM clapping task (similar performance without prompting, as well as similar performance improvement for all groups after being prompted), both ASD groups performed significantly worse than the NT group (who significantly improved their performance from the first to the second unprompted trial) on the second unprompted trial. No significant group differences emerged for the PM reward task. The authors concluded that severely autistic children can succeed on certain event-based PM tasks if task characteristics are adjusted to their needs (i.e., rewarding circumstances, specific PM cues). Further they suggested that, although children with ASD benefitted from prompts in the individual trials of the PM clapping task, this did not positively affect their performance from trial 1 to 2 in contrast to the NT group. The authors suggested this could be explained by information processing deficits in ASD when the ability to connect and integrate information across tasks is required (Olu-Lafe et al. 2014).

Although Sheppard et al.’s (2016) study is interesting, it suffers from a major methodological issue in terms of matching procedure. Groups were not matched for age or gender, and no standardised measure was employed to equate groups for cognitive ability. Instead groups were matched only for national curriculum point scores for reading, writing, and number skills. The authors argue that “the demands of standard IQ tests such as WASI (Wechsler Abbreviated Scale of Intelligence, Wechsler 1999) or even BPVS (British Picture Vocabulary Test, Dunn et al. 2009) make this type of matching unsuitable for children with severe ASD” (pp. 6–7). However, it is counterintuitive that, if a child is not capable–because of their severe social-communication deficits–of completing standard IQ tests, they could complete an experimental task as complex as the task(s) used in this study (some of which were inherently social).

Fourth and finally, Altgassen and Koch (2014) studied the contribution of inhibitory control demands on event-based PM performance in 22 adults with ASD and 22 age- and nonverbal ability-matched NT adults. For this purpose, they used a triple task within-subject design. Participants completed two experimental conditions in counter-balanced order. The ongoing task involved a word categorisation paradigm (deciding which of two words belongs to the same category as a third word). The colour of all three words changed randomly every trial. The PM instruction was to press a pre-specified key whenever all three words were printed in blue font. Simultaneously, inhibitory load was manipulated using an auditory mental arithmetic task. In a “low inhibitory load” condition, participants were presented with a sequence of numbers via headphones and were required to add five to each number and state the resulting sum. In a “high inhibitory load” condition, the procedure was the same but participants had to withhold saying the sum aloud when it equalled 8 or 15. Contrary to the hypothesis of the authors, there were no between-group differences in PM in either condition (i.e., regardless of inhibitory load), or in the extent to which responses were correctly withheld in the high-inhibitory control condition. These findings are striking and suggest that event-based PM is not impaired in ASD even when demands on executive resources associated with successful performance are high. However, in both inhibitory load conditions, performance on the ongoing categorisation task itself was significantly superior in the comparison group than in the ASD group (\(\eta _{p}^{2}\) = 0.18). As such, it remains a possibility that participants with ASD were allocating relatively more of their cognitive resources to completion of the PM task and the ongoing inhibition task than were comparison participants. Therefore, it may be that participants with ASD were using alternative, compensatory strategies to succeed on the PM component of the task at the expense of performance on the ongoing activity. Problems with multimodal integration (Baum et al. 2015) in this particular triple-task design might have led individuals with ASD to focus more on one of the three tasks. That is, it might be that when attentional demands of the environment are high, people with ASD may need to prioritise carrying out the planned PM action at the expense of other activities to an extent that NT individuals do not. Either way, the fact that participants with ASD in Altgassen and Koch’s study performed comparably to NT individuals on a PM task that had very high executive demands suggests that this ability cannot be grossly impaired in adults with this disorder.

Overall, with the exception of study of Altgassen and Koch (2014), the studies discussed above, suggest that event-based PM is impaired in ASD contrary to the initial landmark studies reviewed. However, based on the critical analysis of the studies’ methodology, such a conclusion should be drawn tentatively at best. Below we will discuss a final set of studies exploring PM in ASD in more naturalistic settings.

To explore both time- and event-based PM in a naturalistic setting Altgassen et al. (2012) tested 25 adults with ASD and 25 NT participants matched for chronological age and intellectual abilities. The ongoing task consisted of preparing breakfast (using props) for four people following a simple set of rules. Two time-based (taking the tea bag out of the tea after 3 min; putting butter on the table 6 min before guests arrive) and two event-based (preparing tea immediately after kettle went off, which was indicated by the kettle changing colour; turning off the egg cooker when it beeped) PM components were embedded the breakfast preparation routine. The aim of this complex paradigm was to mirror real life PM demands. In addition, participants also completed a standard measure of event-based PM (Red Pencil Test; Salthouse et al. 2004), in which participants were asked to repeat the words ‘red pencil’ whenever the experimenter said ‘red pencil’ throughout the experimental session (which happened twice). Based on previous research, the authors predicted that only time-based PM would be diminished in the ASD group. In fact, however, participants with ASD showed diminished performance (i.e., more failures to complete the PM action) across all PM tasks. In relation to the time-based PM components of the breakfast task, the ASD group also monitored the time less often than the NT group.

Two other studies aimed to explore PM in ASD in a more “real-life” setting. Both used the Virtual Week paradigm (see Rendell and Henry 2009 for full details of the paradigm). The Virtual Week is a computerised single player board game that simulates 5–7 days of a week. Depending on the version of the task, the day and time of the virtual day are either always visible, or appear only after a particular keyboard key is pressed by the participant. The player rolls a die and moves a token around a board of 121 squares that represents one virtual day. On each virtual day, the player will pass ten squares, which require them to pick up an action card. Each action card poses a question about an activity for that time of the day that the participant has to make a decision about (e.g., choosing between three options for that day’s breakfast). Importantly, each choice determines whether the player has to roll a specific, an odd, or any number with their die on their next move to be allowed to continue moving their token around the board (which is revealed to them after selecting an activity option). Participants are unable to move on and have to repeat rolling the die until they have rolled the specific die number required. Hence the demands of rolling the die, moving the token around the board and making decisions about the activities to participate in, serve as the ongoing activity. Due to this board game nature of the task, the Virtual Week does not provide a measure of ongoing task performance per se, unlike more laboratory PM tasks. Additionally, each day, a total of four time- and four event-based PM tasks have to be carried out, in addition to the ten activity-related decisions that are inherent to the game. The event-based PM tasks have to be carried when a particular event occurs (as indicated by the action cards; e.g., take medication at breakfast would be triggered by the breakfast card). The time-based PM tasks have to be carried out at a set time of the virtual day (e.g., phone the plumber at 5 pm). Some PM tasks require execution on a regular basis whereas others are irregular. The regular ones are the same on each day of the Virtual Week (two time-based, two event-based). The irregular ones are one-off tasks that are instructed at the beginning of, or during, a new virtual day (two time-based, two event-based). Thus, the retrospective memory load for the irregular PM tasks is higher compared to the regular ones. At the appropriate moment to execute a PM task, the participant has to press a “perform task” button to bring up a list of possible tasks, and then choose the correct one.

One of the two studies which have used this Virtual Week paradigm with individuals with ASD was that of Henry et al. (2014). This study explored how differential levels of task absorption (i.e., the level of engagement in the ongoing task) affected PM in 30 children with ASD and 30 NT children matched for age and IQ. Tasks in the Virtual Week were adjusted to reflect children’s everyday life. Further, to reduce/eliminate time-monitoring demands, the time of each virtual day was always present in the centre of the screen. Participants completed the Virtual Week game under two conditions (each lasting three virtual days) in counterbalanced order. In the high task absorption condition, participants could only continue moving their token around the board if they rolled a specific number after an event card (see description above), whereas in the low absorption condition, the outcome of the next die roll was not restricted. The authors predicted that high task absorption may lead to greater PM impairment in the ASD group. A 2 (Group: ASD/NT) × 2 (Task absorption: low/high) × 2 (Type of PM task: event-based/time-based) × 2 (Regularity: regular/irregular) mixed ANOVA was conducted to explore effects on PM performance. Most importantly, a significant Group × PM task interaction (\(\eta _{p}^{2}\) = 0.21) was found indicating that the ASD group was only impaired in time-based but not event-based PM, which ties in with the consistent pattern found by the landmark (tightly controlled) studies summarised above. Contrary to author predictions, high task absorption did not affect the ASD group to a greater extent. Further, the Group × Regularity interaction approached significance (p = .06, \(\eta _{p}^{2}\) = 0.06). The authors broke this marginally non-significant interaction down using post-hoc t-tests. These t tests showed no performance difference between regular and irregular PM tasks within the ASD group (\(\eta _{p}^{2}\) = 0.02), whereas NT individuals performed slightly better on regular PM tasks (p = .04, \(\eta _{p}^{2}\) = 0.07). However, in comparison to the NT group, participants with ASD were still less accurate on both regular (\(\eta _{p}^{2}\) = 0.21) and irregular (\(\eta _{p}^{2}\) = 0.16) PM tasks. Therefore, authors concluded that PM difficulties in children with ASD are not a result of retrospective memory processes. Instead, they suggested that a monitoring deficit might underlie their PM deficits as time-based PM requires more self-initiated monitoring processes.

Another study which used the Virtual Week set-up was that of Kretschmer et al. (2014). Rather than focusing on the degree of absorption (which presumably affects retrieval of the original intention to carry out an action), Kretschmer et al. (2014) used this paradigm to investigate the effects of different encoding strategies on PM in a sample of 27 adults with ASD and 27 NT adults matched for age, verbal, and non-verbal abilities. The Virtual Week setup and tasks were as described above, but participants had to complete only three virtual days, and needed to press a specific keyboard key to display the time of the day. PM encoding was compared across two between-subject conditions, the first being a ‘standard’ condition and the second requiring implementation intentions, which is an encoding strategy that requires to form ‘if-then’ statements; that is, creating a specific situation when, where, and how to perform one’s intention (Gollwitzer 1999). Implementation intentions have been shown to improve PM in NT studies (Chen et al. 2015) and are thought to support episodic future thinking (Atance and O’Neill 2001). Hence, this was the first study exploring strategies to enhance PM in ASD. Specifically, in the implementation intention condition, participants had to form such an ‘if-then’ statement for each irregular PM task instruction of that day after the instructions had been presented on-screen (e.g., say out-loud, “when it is 5 p.m., then I will press the ‘perform task’ button and select ‘phone the plumber’”; Kretschmer et al. (2014), p. 3112). The logic here was that forcing participants with ASD to form implementation intentions would support their PM and, thus, raise their performance level to one commensurate with that among NT participants. No additional instructions were given in the standard condition. To analyse experimental task performance, the authors ran a 2 (Group: ASD/NT) × 2 (Encoding condition: standard/implementation intention) × 2 (Type of PM task: event-based/time-based) × 2 (Regularity: regular/irregular) mixed ANOVA to analyse PM performance across all three virtual days. They found a main effect of group (\(\eta _{p}^{2}\) = 0.14), as well as a significant Group x Regularity (\(\eta _{p}^{2}\) = 0.12) interaction. Interestingly, post-hoc test results revealed no within-group differences for NT adults, but ASD participants performed better on regular than irregular PM tasks (\(\eta _{p}^{2}\) = 0.17). Because no Group × PM task interaction emerged, the authors concluded that individuals with ASD have a general deficit across both time- and event-based PM. The Group × Encoding condition interaction was non-significant with a small to medium effect size (p = .08, \(\eta _{p}^{2}\) = 0.06). Nonetheless, the authors broke down the interaction effect. Post-hoc between-participant tests revealed that, relative to comparison participants, individuals with ASD showed diminished performance in the standard condition only; whereas in the implementation intentions condition, the between-group differences in performance were non-significant. Based on these results, the authors concluded that implementation intentions might present a strategy to support PM in individuals with ASD. However, a closer inspection of the results suggests that this conclusion is not entirely warranted. Kretschmer (personal communication, October 2016) provided the group means and SDs, which indicated that the ASD group only benefitted from employing implementation intentions for event-based PM [ASD: M_{ImplementationIntentions} = 0.81 (SD = 0.22), M_Standard = 0.62 (SD = 0.33); TD: M_{ImplementationIntentions} = 0.84 (SD = 0.22), M_Standard = 0.80 (SD = 0.29)]. However, rather than implementation intentions improving time-based PM performance of ASD, they instead decreased the performance of comparison participants (relative to the standard condition performance) [ASD: M_{ImplementationIntentions} = 0.53 (SD = 0.32), M_Standard = 0.49 (SD = 0.32); TD: M_{ImplementationIntentions} = 0.58 (SD = 0.25), M_Standard = 0.79 (SD = 0.24)]. Further, in contrast to Henry et al. (2014), the authors concluded that retrospective memory demands are important to understand PM deficits in ASD as participants only showed significant impairments in the irregular (one-off non-routine) PM tasks (p < .001, \(\eta _{p}^{2}\) = 0.27), which place particularly high demands on retrospective memory. Unfortunately, the authors could not check whether participants actually remembered the irregular PM task instructions after each virtual day; although participants had to repeat the PM instruction three times aloud at the stage of encoding (which was supposed to ensure later remembering), there is no way of knowing whether the instruction was stored for the duration of the Virtual Week task. As such, retrospective memory limitations in ASD might explain entirely the group difference in the number of times irregular PM tasks were completed.

The Virtual Week is an interesting approach to study PM in ASD. The game format makes it easily accessible and it attempts to mirror everyday PM demands in several respects. However, the PM demands of the task are arguably much greater than (and of a different quality to) those in real life; participants have to remember 24 PM tasks (requiring the execution of both time- and event-based tasks and changing retrospective memory load) during a short period of time and without the use of any external reminders. Further, in the version used for both studies on ASD, the Virtual Week version did not offer the possibility to check whether participants actually remembered their PM tasks for each virtual day, which would have been particularly important for irregular PM tasks. However, Henry et al. (2014) pointed out that this feature is now part of the newest version of Virtual Week.

The results of the two studies that employed the Virtual Week paradigm differ in one major respect. Henry et al. (2014) found only time-based PM to be diminished in children with ASD, which is in line Williams et al.’s findings (2013, 2014). In contrast, Kretschmer et al. (2014) observed impairments of both time- and event-based PM in adults. This is surprising and requires further exploration. Kretschmer et al. (2014) employed a version of the Virtual Week that was equivalent to Henry et al.’s (2014) high-absorption condition. A possible explanation for the differing pattern of results could be the aforementioned retrospective memory difficulties in the ASD group. Kretschmer et al. (2014) found that the ASD group performed worse on the irregular PM tasks that posed the highest retrospective memory demand. This result may reflect the viable possibility that participants with ASD simply forgot the PM instruction more frequently. In general, the possibility that event-based PM deficits in ASD are observed only when demands on retrospective memory are high is brought into focus by the findings from a very large study of “everyday memory” by Jones et al. (2011).

Jones et al. (2011) investigated everyday memory in 94 adolescents with ASD and 55 age-, and IQ- (verbal, performance, and full scale) matched NT peers. Jones et al. used the Rivermead Behavioural Memory Test (Wilson and Baddeley 1991) to assess everyday memory across multiple subtests, three of which tested event-based PM. These sub-tests involved (a) reminding the experimenter about the location of a pen upon the occurrence of a particular verbal cue; (b) asking the experimenter a question when an alarm went off; and (c) remembering to pick up an envelope before walking a route as demonstrated by the experimenter. The ASD group achieved a significantly lower PM composite score (across the three subtests) than the comparison group, indicating significant event-based PM impairments in this very large sample of ASD participants. However, Williams et al. (2013) noted that Jones et al.’s analysis included participants who had failed to remember the PM task instruction at all. When the data from Jones et al. (2011) were re-analysed excluding participants who completely failed to recall the PM instruction, there was no hint of any between-group differences in PM task performance (Williams et al. 2013, p. 1564). This re-analysis underscores the importance of controlling for retrospective memory demands when drawing conclusions about PM ability in ASD.