Participants

The participants were 90 undergraduate students enrolled in psychology courses at the University of Waterloo (33 males, 57 females) with a mean age of 19.34 years. All participants took part in exchange for course credit.

Apparatus and stimuli

The experimental program was written using Python (www.python.org) and presented using PsychoPy software (Peirce,2007). The experiment was run on a Mac Mini computer with a 2.40-GHz processor connected to a 24-in. Phillips 244E LCD monitor, placed at an approximate viewing distance of 57 cm. All of the stimuli were nouns, and the stimulus list was composed such that half of the words represented things that were bigger than the computer monitor, and half represented things that were smaller than the computer monitor. The stimuli were 360 English words obtained from the MRC Psycholinguistic Database. The words representing things bigger than the computer screen had a mean Kučera and Francis (1967) frequency of 45.3 (SE = 3.4) and a mean word length of 5.4 letters (SE = 0.4). The words representing things smaller than the computer screen had a mean frequency of 31.1 (SE = 2.3) and a mean word length of 5.2 letters (SE = 0.4). The stimuli did not differ statistically on either of these dimensions. All of the stimuli were presented centrally and subtended a visual angle of 1º (for lowercase words) or 1.5º (for uppercase words) vertically, and 3º–5.5º horizontally. All stimuli and instructions were presented in white Arial font on a black background.

Procedure

Prior studies examining the effects of depth of processing on memory performance have employed either intentional-encoding instructions (in which participants know that the encoding task will be followed by a memory test) or incidental-encoding instructions (in which the memory test is not known in advance). In the present study, we employed intentional-encoding instructions for important methodological reasons: Because we were combining thought probes into a standard study–test memory paradigm (which, to our knowledge, has never been done before), a sufficiently long study phase was required for us to be able to insert enough thought probes to obtain a reliable metric of mind-wandering. In the present context, however, an overly long encoding phase would be detrimental to subsequent memory performance. For this reason, we had to retain short encoding phases (of 30 items), which meant that participants had to perform several of them, since only two thought probes can be inserted into an encoding phase of this length without contaminating our measure of mind-wandering (see Seli, Carriere, Levene & Smilek,2013). Therefore, the repeated study–test design necessitated by these constraints means that after the first block of study and test sessions, participants become aware that they will be required to “remember” the presented items. For this reason, encoding was necessarily intentional. We therefore told participants prior to the first encoding phase that a subsequent recognition memory test would follow each encoding phase. This requirement is not of particular concern, however, since Craik and Tulving (1975), in their seminal work on depth-of-encoding benefits, showed that such benefits do not depend on whether encoding was conducted under intentional or incidental instructions.

For the encoding phases of the experiment, participants were informed that they would either judge whether the words presented represented things that were bigger than the computer monitor, or judge whether the words presented were in uppercase. At the beginning of each encoding phase, either the phrase “Bigger than monitor?” or “Uppercase?” was presented for 1 s. Following a fixation cross that was also presented for 1 s, the encoding phase for that block began. Each encoding phase consisted of the presentation of 30 words, one at a time. Words appeared centrally for 200 ms, followed by a fixation cross that remained upon the screen until a response was made. For both encoding tasks, “Yes” responses were made by pressing the “z” key, and “No” responses were made by pressing the “m” key. Following responses, the fixation cross remained on the screen for an additional 1,200 ms before the presentation of the next word.

In addition to the size and case judgments during the encoding phase of each block, participants were informed that at various points throughout the encoding phases that their thoughts would be probed. These thought probes required participants to indicate whether, to the best of their knowledge, their attention was focused on or off the encoding task. Prior to the experiment, participants were provided with the following definition of on-task versus off-task thought (mind-wandering), adapted from Smallwood et al. (2007): “During this experiment you will be asked at various points whether your attention is firmly directed towards the task, or alternatively you may be aware of other things than just the task. Occasionally you may find as you are performing the task that you begin thinking about something completely unrelated to what you are doing; this is what we refer to as ‘mind-wandering’.” Participants were also provided with written definitions of on- and off-task thought that they were free to consult throughout the experiment. These written definitions were stated as follows: “On-task: Just prior to the probe, your attention was firmly directed towards the task. Off-task: Just prior to the probe, you were aware of things other than the task; you were thinking of something completely unrelated to what you were doing.” Participants indicated that they were on task by pressing the “z” key on the keyboard, and indicated that they were off task by pressing the “m” key (this assignment was reversed for half of the participants). For each encoding phase (three size blocks, three case blocks), two thought probes were presented, resulting in a total of six thought probes for each encoding type (shallow and deep). Thought probes were presented pseudorandomly throughout each encoding phase, with the constraints that no probes were encountered within the first ten judgments of each encoding phase and that probes were separated by a minimum of 12 intervening judgments. This yielded an experiment-wide total of six thought probes during shallow encoding and six during deep encoding.

In each recognition phase, the 30 studied items were randomly presented along with 30 novel items (all items were presented in lowercase font). The participants’ task was to indicate whether each item had been encountered in the immediately preceding encoding task. “Yes” and “No” responses were mapped to the same keys as in the encoding phase. Therefore, each experimental block consisted of making either a size (deep-encoding) judgment or a case (shallow-encoding) judgment to each of 30 words, followed by a recognition memory test with 50 % old items and 50 % new items. Three blocks were completed in which size judgments were made during encoding, and three blocks were completed in which case judgments were made during encoding. These blocks were interleaved, and order was counterbalanced across participants. The experiment took approximately 30 min to complete.

Encoding phase

Overall, the observed frequency of off-task reports collapsed across encoding condition was 2.08 out of six thought probes (34.5 %). Mind-wandering rates were no different in the deep-encoding condition (M = 2.01,SE = 0.18) than in the shallow-encoding condition (M = 2.14,SE = 0.18),t < 1. However, reaction times for the size judgment task were significantly slower than those for the case judgment task (1,019 vs. 661 ms),t(89) = 15.96,p < .001,d = 1.69, as had been reported previously (i.e., Craik & Tulving,1975). Finally, accuracy was significantly lower for the size judgment task than for the case judgment task (78/90 and 82/90, respectively),t(89) = 2.692,p = .008,d = 0.29.

Recognition performance

Recognition performance was computed for items encountered in the deep-encoding (size judgment) conditions (collapsed across block), and separately for items encountered in the shallow-encoding (case judgment) conditions (collapsed across block). For each type of encoding, recognition sensitivity was computed using the following formula:d' = z(hits) –z(false alarms). Mean recognition performance was higher following deep encoding (d' = 2.22,SE = 0.12) than during shallow encoding (d' = 1.12,SE = 0.08),t(89) = 12.39,p < .001,d = 1.31—a replication of the classic “depth-of-encoding” benefit on memory performance (Craik & Lockhart,1972; Craik & Tulving,1975). The mean benefit of deep relative to shallow encoding wasd'_deep –d'_shallow = 1.10,SE = 0.09.

Recognition performance as a function of mind-wandering rates

Our primary interest here was the relation between mind-wanderingwithin each type of encoding condition and recognition memory performance. To that end, Pearson product–moment correlation coefficients were computed separately for the shallow and deep encoding conditions. For the deep encoding condition, we found a significant negative relationship between mind-wandering and recognition performance,r = −.225,p = .033 (partials with order covariate:r = −.235,p = .027), whereas for the shallow encoding condition, no significant relation was apparent between these two variables,r = −.033,p = .761 (partials with order covariate:r = −.034,p = .749). Therefore, in line with our hypothesis, the effects of mind-wandering on memory performance are asymmetric with respect to deep and shallow encoding. Scatterplots of the relationships between mind-wandering frequency and memory performance for the two encoding conditions are depicted in Fig. 1a andb. We also present the scatterplot representing depth-of-encoding benefits (d'_deep –d'_shallow) as a function of differences in mind-wandering rates (MW_shallow –MW_deep) in Fig. 1c. In Table 1, we include the full correlation matrix assessing the relations among mind-wandering at encoding, judgment task accuracy at encoding, and recognition memory performance, for both the shallow- and deep-encoding blocks.

a Scatterplot depicting recognition sensitivity (d') following deep encoding as a function of the number of “off-task” reports during encoding.b Scatterplot depicting recognition sensitivity (d') following shallow encoding as a function of the number of “off-task” reports during encoding.c Scatterplot depicting depth-of-encoding benefits, as measured by the difference between recognition performance following deep (size judgments) relative to shallow (case judgments) encoding, are shown as a function of the difference in “off-task” reports during the deep- and shallow-encoding phases. The line of best fit indicates that greater mind-wandering differences (i.e., more mind-wandering during shallow than during deep encoding) positively predict bigger depth-of-encoding effects (i.e., largerd' following deep than following shallow encoding)

Full size image

As can be seen from Table 1, although mind-wandering during deep encoding correlated with subsequent memory performance (our primary effect of interest), it also correlated with encoding accuracy during deep encoding,r = −.211,p = .046, indicating that thought probe responses were diagnostic of participants’ online focus of attention to the task, and not merely representative of trait-level mind-wandering or differences in individuals’ willingness to report being off task, as has been demonstrated in prior work using thought probes (see McVay & Kane,2009; Seli, Cheyne & Smilek,2013). Furthermore, we observed a strong correlation between encoding accuracy and subsequent memory performance in the deep-encoding blocks,r = .460,p < .001. Indeed, this was to be expected, since if accuracy on the judgment task is a metric of the degree of attentional focus, greater attention (hence, greater accuracy) should result in better encoding into memory and better consequent memory performance. In fact, the likely mechanism by which increases in mind-wandering hinder memory performance in the deep-encoding block is a coopting of attention away from the judgment task, thus reducing the effectiveness of encoding. This was confirmed via a step-wise regression, in which mind-wandering on its own significantly predicted memory performance, whereas when mind-wandering rates as well as judgment task accuracy for the deep-encoding task were taken into account, the relation between mind-wandering and memory performance was not significant. This indicates that the effects of mind-wandering on memory performance in the deep-encoding condition were, as expected, mediated via the effects of mind-wandering on judgment task accuracy. The results of this analysis can be seen in Table 2.

The purpose of the work reported here was to assess the hypothesis that the effects of mind-wandering on shallow- and deep-encoding tasks would be asymmetric with respect to subsequent explicit memory performance. We reasoned that a semantic-encoding task would place higher demands on executive resources than would a simple perceptual task, so that instances of mind-wandering during such tasks would be more likely to impair semantic than to impair perceptual encoding. Consistent with this hypothesis, we observed a significant negative relation between mind-wandering rates and encoding task accuracy for words encoded under semantic processing conditions. Importantly, the negative effects of mind-wandering on encoding had a detrimental impact on subsequent recognition memory performance, as was evidenced by our step-wise regression analysis, in which the relationship between mind-wandering and recognition performance in the semantic-encoding condition was mediated by encoding task accuracy (as per Table 2). Crucially, however, we found no relation between mind-wandering rates and either encoding task accuracy or subsequent recognition performance for the perceptual-processing condition. Although it may be tempting to attribute the lack of effects in the perceptual-encoding condition to floor effects on memory performance, one must remember that shallow encoding, by definition, results in extremely poor memory performance. Nonetheless, these results demonstrate that memorial gains afforded by so-called “deep” encoding, under a levels-of-processing framework, were severely offset by instances of off-task thought.

The results of the present study have important implications for both the study of mind-wandering and the study of human memory. First, we have provided the first demonstration, to our knowledge, of differential consequences of mind-wandering on performance across conditionswithin the same task context, a pattern of results that is wholly consistent with (and indeed, is predicted by) attentional-resource accounts of mind-wandering (Smallwood & Schooler,2006). Specifically, it is argued that as the executive/attentional demands of a task increase, so should the observed consequences of mind-wandering on primary-task performance. Second, we have provided the first online measure of mind-wandering in a standard study–test memory paradigm, by inserting thought probes directly into the study phase. This methodology revealed that mind-wandering decreases attention to the encoding task itself, thus resulting in impoverished memorial representations. Although these findings are well-explained under an attentional-resource account of mind-wandering, such a theory would also predict thatoverall differences in off-task reports should be observed across our shallow- and deep-encoding conditions, to the extent that these conditions differed in their attentional demands. Although this lack of difference in the present study is not problematic for our primary findings (which concern individual-difference measures), it is noteworthy and deserves further comment.

The similarity of mean mind-wandering rates across the shallow- and deep-encoding conditions in the present study occurred despite the fact that the deep-encoding task placed greater demands on executive/attentional resources. And although this may seem at odds with attentional-resource accounts of mind-wandering, it may actually be perfectly compatible with such a theory. Specifically, in a prior study, Thomson, Besner and Smilek (2013) put forth a resource distribution account of task-unrelated thought, in which mind-wandering-related deficits in primary-task performance are argued to derive from an inefficient distribution of attentional resources between task-related and task-unrelated thought. In the present study, it may simply be the case that individuals failed to devote sufficient resources to the deep-encoding task (as evidenced by the slower and less accurate performance in this condition). In this way, similar proportions of available resources may have been devoted to both the shallow- and deep-encoding tasks, thus resulting in similar rates of reported mind-wandering across tasks, but poorer performance on the deep-encoding task. The precise factors that determine the efficiency of attentional resource distribution will be a fruitful avenue of future study.

In summary, we conclude that the effects of mind-wandering on the encoding of information into memory depends on the nature of the encoding operations being performed. To the extent that encoding operations place high demands on attentional/executive resources (which almost all known effective encoding strategies do), mind-wandering will be particularly detrimental. In the context of the levels-of-processing manipulation, mind-wandering during a semantic-encoding task coopts attentional resources in a way that hinders one’s ability to create a durable episodic representation. Furthermore, the dependence of these effects on the attentional demands of the encoding task itself is consistent with attentional-resource theories of mind-wandering.