Working Memory and Theory of Mind

A number of other studies have shown that the development of executive function, more specifically working memory, influences the development of first-order ToM (e.g., [13–15]; but see also [16] for evidence that simple working memory is not sufficient without inhibition). On the other hand, to the best to our knowledge and according to Miller’s [12,17] extensive review, there are only two studies that focused on the role of working memory and second-order false belief reasoning in typically developing children and those studies yielded contradictory results.

The first one is Perner et al.’s [18] study with typically developing children and children at risk of ADHD. As a part of their study, to test executive function, they used forward and backward digit span. They found a significant positive relationship between the simple working memory task, i.e., the forward digit span, and the second-order false belief task. However, more recently, Hasselhorn, Mahler, and Grube [19] also tested children around the age of six with a second-order false belief task and with a simple working memory task (a digit span task) and a non-word repetition task together with verbal ability tasks. Their results showed that the significant correlation between the simple working memory span score and children’s second-order false belief score was no longer reliable when vocabulary knowledge and age were controlled for (r(56) = .13, ns).

What could be the role of working memory in the development of second-order false belief reasoning? It has been shown that working memory acts as a bottleneck [20], meaning that people can only hold one chunk of information in working memory at a time. Given this restriction, we invoke the serial processing bottleneck hypothesis [21].

Evidence for a working memory bottleneck in ToM reasoning comes from dual-task paradigms. The general idea of the dual-task paradigm is to find two different tasks and present them simultaneously, in order to compare performance with the two single tasks in which the participants have already performed well. If the performance of the first task decreases when it is presented concurrently with the second task, it can be inferred that the two tasks both require the same cognitive resource [22,23] and that that common resource acts as a bottleneck [20,24]. McKinnon and Moscovitch [25] showed that young adults performed significantly worse on second-order ToM reasoning than on first-order ToM reasoning in a dual-task condition with a demanding secondary executive function task (i.e., the 2-back task). Participants in the single-task condition who only did second-order ToM reasoning did not show any loss in performance. This result was replicated in another study with young adults using a more advanced and naturalistic ToM task, namely the “Reading the Mind in the Eyes” task [26].

The serial processing bottleneck hypothesis can be seen as a procedural explanation of the cognitive process of serializing the hierarchical content of thought into proper chunks along with their relations such that they fit easily through the processing bottleneck. Working memory constrains this process in terms of capacity and efficiency. The order of the processed chunks also reflects the reasoning steps children have to go through in order to solve a second-order false belief task. The serial processing bottleneck hypothesis is tantamount to a computational account under which second-order false belief can be conceived as a social-cognitive reasoning task employing a proper procedure–serialization–and a critical amount of mental resources–working memory–in order to cope with its nested structure.

Because working memory acts as a bottleneck, we propose that children who cannot pass second-order false belief tasks might have a lack of efficiency in reasoning when they have to serially process embedded beliefs. More specifically, when children try to answer second-order false belief questions, e.g., “Where does Mary think that John will look for the chocolate?” but have no efficient reasoning rule such as “if Mary didn’t see that John saw her hiding the chocolate, then she thinks that John thinks that the chocolate is still where he put it before, which is in the drawer”, more reasoning steps are needed to attribute a second-order false belief to Mary. A possible sequence of reasoning steps from the child’s perspective might be as follows: i) “John knows that the chocolate is in the toy box”, ii) “Mary does not know this”, iii) “Mary knows that John put the chocolate into the drawer before”, iv) “Mary thinks that John will look for the chocolate in the drawer”. In line with de Villiers et al. [27], each reasoning step is in the form of a single-embedded sentence at the surface. In addition to de Villiers et al.’s study, the serial processing bottleneck hypothesis allows us to propose a possible explanation for their failures on children’s development of processing these rules to answer the second-order false belief question in terms of working memory.

If children do not have an efficient reasoning rule, they need to go through each of the reasoning steps i-iv, which occupies working memory temporarily. In order to proceed in reasoning, due to the working memory bottleneck, at each step, the information in working memory needs to be sent to long-term memory to be retrieved later, if necessary. Retrieving information from long-term memory also takes time and increases the odds of forgetting and of retrieving wrong information [28]. Therefore, having more inefficient rules instead of one efficient rule means that the process is more prone to errors and takes more time [29,30]. This view is consistent with research showing that children perform better in language comprehension tasks and cognitive tasks when they are given more time [31–34]. Once children have enough experience in applying these reasoning steps sequentially, they are combined to one efficient rule, repeated here for convenience: “if Mary didn’t see that John saw her hiding the chocolate, then she thinks that John thinks that the chocolate is still where he put it before, which is in the drawer” (see [35] for the details of a mechanism that combines rules).

In order to investigate the serial processing bottleneck hypothesis, we looked at the relationships between a complex working memory task, a simple working memory span task, and second-order false belief reasoning. We did not use a dual-task paradigm because children have to be good at second-order false belief reasoning already in order to use a dual-task paradigm and our main focus is on children who are still developing second-order false belief reasoning. Although both simple and complex working memory tasks are related to working memory in a broader sense, they require different strategies. Unlike our simple working memory task that only requires building a representation of a list of words to be remembered, our complex working memory task requires processing information serially as well as cognitive control [36,37]. Therefore, we reason that children who have complex working memory strategies that can overcome the working memory bottleneck will be more successful in applying these in second-order false belief reasoning as well.

As a simple working memory span task, we used a word span task (WST) and as a complex working memory span task, we used a listening span task (LST). Considering the language-based nature of the listening span task, we used the word span task instead of the digit span task in order to keep the modality the same between the simple and complex span tasks. However, our results still can be compared with Perner et al.’s and Hasselhorn et al.’s study, because it has been shown that word span and digit span are closely related, r = .65, p < .001 [38] and have been grouped together with other span tasks which test the same component of working memory [39].

We follow Carlson’s [40] terminology, and refer to our simple word span task as a measure of pure working memory. Note that simple word span tasks have been referred to as short-term memory tasks as well [41]. The details of the tasks are explained in the Methods section.

Syntactic Recursion and Theory of Mind

Similar to the studies showing that language development contributes to the development of first-order ToM [27,42–47] (but see [48] for evidence of false belief understanding in preverbal infants), a number of studies have shown that language is important in children’s development of second-order ToM.

For example, Hollebrandse, van Hout, and Hendriks [49] compared six- to nine-year-olds’ performance on a verbal and a low-verbal version of a second-order false belief task in order to investigate whether language in general helps children to pass second-order false belief tasks. They found that children’s scores were lower in the low-verbal version of the second-order false belief task compared to the verbal version and concluded that language might support explicit reasoning about higher-order beliefs by facilitating tracking different beliefs. Similar to the findings of Hollebrandse et al. [50], Kuijper [51] found that low-verbal second-order false belief tasks are also harder than high-verbal ones for children between six and twelve years of age who have been diagnosed with autism spectrum disorder or attention-deficit/hyperactivity disorder.

Lockl and Schneider [52] found that, at the age of five, children’s general language abilities (i.e., a combined score of sentence comprehension, morphological rule abilities and sentence memory) were strongly correlated with their second-order false belief reasoning. However, they stated that their data were not well suited to separate out the effects of syntactic and semantic abilities (p. 163). These studies indicate that explicit mental state language may support the development of second-order false belief reasoning. Yet, it is still not clear which aspect of language it is that helps.

Therefore, in this study, we investigate the possible relationship between syntactic recursion in the language domain and recursion in the thought domain on the same level of recursion, namely second-order.

The syntactic component of language is found to be related to first-order ToM in terms of its hierarchical embedding structure [27,43,53–55]. Usually, first-order complement clauses, as shown below in Example (1) (adapted from [44]), have been used in the literature to investigate this relationship. Complement clauses such as “that p” may be used to express propositional attitudes (or opinions) towards some state p in the world [27]. They may be preceded by mental state verbs as in “Mary knows that p” or in “Mary believes that p” or by communication verbs as in “Mary said that p”. Complement clauses can be used recursively, as shown below in Example (2). Moreover, complement clauses allow people to represent states that contrast with reality or with other people’s mental states in terms of truth-value. Thus, while in Example (2), “Mary said that there was a flea in her cereal” might be false, the whole sentence “John said that Mary said that there was a flea in her cereal” might be true.

1. First-order complement clause: “Mary said that there was a spider in her cereal. But it was just a raisin”.
2. Second-order complement clause: “John said that Mary said that there was a flea in her cereal. But in fact, she said that there was a spider in her cereal”.

Recently, de Villiers et al. [27] argued that experience with truth-value contrasts in contexts with full tensed complement clauses such as Example (1) opens the door for children to pass first-order ToM tasks and to recognize syntactic recursion. Subsequently, understanding sentence recursion allows children to pass recursive ToM tasks. More specifically, they suggest that recursive complements in contexts in which the truth-values vary, such as in Example (2), are necessary for recursive false belief reasoning. However, they conclude that their predictions need to be explored further.

Like complement clauses, relative clauses can be used recursively; that is, besides first-order relative clauses, there are second-order relative clauses as well as even higher-order relative clauses. At each level of recursion they may refer to a different subject or object. However, unlike complement clauses, relative clauses do not involve propositional attitude verbs such as “knowing that” or “believing that” or communication verbs such as “saying that” and they do not involve truth-value contrasts. Table 1 shows examples of the progression of orders of recursion for ToM attributions and relative clauses. Unlike de Villiers et al. [27], who focused on complement clauses, we used relative clauses, which allows us to specifically focus on the structural parallelism between second-order recursion in the language domain and in the thought domain by excluding the role of truth-value contrasts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Examples of the progression of orders of recursion for theory of mind attributions and relative clauses.

https://doi.org/10.1371/journal.pone.0169510.t001

In the first-order domain, Hale and Tager-Flusberg [56] demonstrated that preschoolers who were trained on first-order complement clauses improved their first-order ToM skills significantly while those trained on first-order relative clauses did not. On the other hand, Smith, Apperly, and White [57] found a positive correlation between first-order relative clauses and first-order false belief tasks in children between the ages of 3 and 4. They concluded that first-order false belief reasoning might not be related to the specific structure of the complement clauses but to the broader category of embedded structures. Another developmental study with Turkish first-order relative clauses supports this positive relationship of first-order relative clauses and first-order ToM [58].

Although the role of first-order relative clauses in first-order false belief reasoning has not been resolved yet, it is worthwhile to investigate the relationship between second-order relative clauses and second-order false belief reasoning. This is because relative clauses share only the syntactic feature of embedded (meta-) representation with second-order false belief reasoning, in contrast to complement clauses, which additionally share semantic features with false belief, namely the fact that the main clause and the embedded complement clause have independent truth-values [56,57,59]. Since we are mainly interested in the structural and representational parallelism between recursion in the language and the thought domain, we chose a second-order relative clause comprehension task (REL_2) as a linguistic predictor of a second-order false belief task, which importantly concerns the same level of recursion. To the best of our knowledge, this is the first time that second-order relative clauses have been studied in relation to second-order false belief reasoning.

Predictions

1. Based on the serial processing bottleneck, we predict that the relationship between the complex working memory task and the second-order false belief task will be stronger than the relationship between the simple working memory task and the second-order false belief task.
2. Because the second-order false belief task and the second-order relative clause task share the same level of recursion, we expect a significant correlation between the two tasks.

Therefore, we expect both the complex working memory task and the second-order relative clause task to figure importantly in the development of second-order false belief reasoning.

Participants

Initially, a sample of 103 children between the ages three and eight was recruited from local kindergartens and primary schools in predominantly middle- and upper middle-class areas of Ankara, Turkey. Our study has been approved by Middle East Technical University (METU) Research Centre for Applied Ethics. A written parent approval form was obtained for every child that participated in the study. All children were monolingual Turkish native speakers. Three children left the study before it was completed. We excluded the youngest 11 children (range: 3;8–4;5) because they had very low scores in all tasks, indicating that the tasks were too hard for them in general. Thus, the results of 89 children were analyzed (37 female, M_age = 6;7 years, SE = 0.13, range: 4;5–8;10). Gender did not show any effect; therefore, the analyses were collapsed over gender. Considering the previous literature which indicates that second-order false belief reasoning starts to manifest itself between the ages five and seven [10,11] and that Turkish children’s development of first-order and second-order false belief reasoning shows a similar pattern with children in Western countries [58,60,61], we divided participants into the following two age groups: children younger than 6;6 were assigned to a younger age group and children older than 6;6 to an older age group:

1. Younger (4–6 years) n = 41; range = 4;6–6;5; M_age = 5;6; SE = 0.10; 17 female,
2. Older (6–8 years) n = 48; range = 6;7–8;10; M_age = 7;6; SE = 0.11; 20 female.

Design

A cross-sectional study design was used with age as a quasi-independent between-subjects variable. The same person tested all of the children in a quiet empty classroom at their school. For each child, all of the tests were completed in one session, which varied from 25 to 35 minutes.

All children participated in the following four tests in the following order: Word span task, second-order false belief tasks, second-order relative clause task, and listening span task. We wanted to keep some temporal distance between the two working memory tasks to prevent any interference effects. For this reason, we used one working memory task at the beginning and the other at the end of the session. Moreover, because we used children’s second-order false belief task scores as dependent variables, we used this task as a second task in our study, in order to prevent a possible fatigue effect. As Carlson and Moses [62] argued, using a fixed order of tasks is standard practice in individual differences research and especially for interpreting correlations between tasks it is important that the task order remains the same for all participants.

Note that we also tested children with a pragmatic understanding task that we constructed. This task was presented after the second-order false belief tasks. In this task, children had to choose correct morphological case markers on an object–definite or indefinite–depending on whether the protagonists in a story had encountered the object before or not. However, this task did not correlate with any other task that we used in the experiment. For this reason, we do not present this task and its results here.

Materials and Procedure

In this section, we present the materials and the procedures in the following order: second-order false belief task, second-order relative clause task, word span task, and listening span task. All of the stimuli can be found in S1 Materials.

Second-order relative clause task (REL_2).

This task concerns the comprehension of relative clauses in Turkish and was adapted from Özge, Marinis, and Zeyrek’s [68] first-order relative clause task with the authors’ permission. The questions and the drawings were modified to second-order relative clauses. Fig 1 demonstrates the drawings for one of the questions in this task. First, introductory pictures were shown to the participants in order to familiarize them with the animals in the actions by telling the name of the animals and the actions (e.g., “This is a pushing sheep, this is a looking monkey and this is a pushing monkey”). After that, the pictures representing the questions were shown one by one. The first and second rows of the picture were pointed out in order to make clear that there were two separate lines of pictures by saying: “This is the first picture and this is the second picture”. In the practice session, the experimenter explained that the participants were required to point out the row with the animals corresponding to their answer. If they could not answer correctly in the practice session, the experimenter pointed out the correct animals and described their actions. However, no feedback was provided during the experimental session. The sentences were repeated up to 4 times. The critical positions for finding the correct answers were equally distributed across the drawings (3 times in the first row and 3 times in the second row) and between right (2 times), left (2 times), and central position (2 times). One practice item and 6 experimental items were used. A child’s total score for experimental items was minimally 0 and maximally 6.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Picture used in the second-order relative clause task (REL_2).

“In which picture is there a sheep that is pushing a monkey that is pushing a sheep?” Adapted from Özge et al. [68] under a CC BY license, with permission from the authors.

https://doi.org/10.1371/journal.pone.0169510.g001

There are other types of relative clauses, e.g., “You show me the sheep that a monkey is pushing that a sheep is pushing”. Because our aim is to examine the relationship between syntactic recursion and second-order false belief reasoning and not children’s different abilities in different types of relative clauses, we used relative clauses of the form “In which picture is there a sheep that is pushing a monkey that is pushing a sheep?”, which are straightforward to understand [68]. Children are exposed to recursion in relative clauses from an early age, for example in well-known nursery rhymes such as “This is the house that Jack built”, also cited in de Villiers and de Villiers [44]: “This is the maiden all forlorn, that milked the cow with the crumpled horn, that tossed the dog that worried the cat, that chased the rat that ate the cheese, that lay in the house that Jack built.”

Note that one might argue that children tend to interpret indirect recursion as conjunction reading. Thus, they might interpret “In which picture is there a sheep that is pushing a monkey that is pushing a sheep?” as “In which picture is there is a sheep that is pushing a monkey and a sheep?”. However, as you can see in Fig 1 and in S1 Materials, none of the pictures allow for such a conjunctive reading. Because all of our subjects pointed out three adjacent animals in one of the pictures throughout the task, we can say that this argument is ruled out.

Testing the Differences between Younger (4–6) and Older (6–8) Age Groups

Table 2 presents the means together with the standard deviations of all the variables for each age group. Note that children between the ages 4;6 and 6;6 were assigned to the younger age group and between the ages 6;7 and 8;10 were assigned to the older age group:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive statistics of the four tasks administered to each age group.

https://doi.org/10.1371/journal.pone.0169510.t002

The analyses showed that while children in the older (6–8 years) group outperformed children in the younger (4–6 years) group in the FBT_2 judgment scores (B = 1.29, SE = 0.42, p = .002), there was no significant difference in the FBT_2 justification scores between the younger and older age groups (B = 0.35, SE = 0.39, p = .38). Moreover, children in the older group outperformed children in the younger group for the REL_2 task, (B = 0.88, SE = 0.39, p = .03), for the WST task, (B = 1.69, SE = 0.45, p < .001), and for the LST task, (B = 1.87, SE = 0.47, p < .001).

Table 3 shows the number of participants and percentages (in parentheses) for each second-order false belief (FBT_2) judgment score (0–2) and justification score (0–4). Consistent with the literature [17], judgment scores for the second-order false belief question were a bit higher for the ‘Birthday Puppy’ story than for the ‘Chocolate Bar’ story for both the younger age group (M_{score.chocolate bar} = 0.46, SD = 0.50; M_{score.birthday puppy} = 0.56, SD = 0.50) and the older age group (M_{score.chocolate bar} = 0.69, SD = 0.47; M_{score.birthday puppy} = 0.85, SD = 0.36).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Number of participants and percentage (in parentheses) of each second-order false belief (FBT_2) score.

https://doi.org/10.1371/journal.pone.0169510.t003

The detailed results about the frequency and percentage of each type of justification answer are shown in Table 4. As can be seen from Table 4, children’s correct justification answers mostly involved implicit second-order answers (e.g., “Because she doesn’t know that Murat saw it”) for both age groups. Moreover, while there were two children in the older age group who gave explicit second-order justification answers (e.g., “Because she believes that Murat doesn’t know that the chocolate is in the box”), none of the children in the younger age group gave any explicit second-order answers. However, children’s justification answers in the older age group (6–8) are clearly not at the ceiling and probably continue to develop after the age of 8.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Frequency (Freq.) and percentage (%) of each type of justification answers.

https://doi.org/10.1371/journal.pone.0169510.t004

Bivariate and Partial Correlations

To inspect the interrelationships among the four tasks, we conducted bivariate and partial correlations. Table 5 shows the bivariate correlation coefficients (Spearman’s r_s) and Table 6 shows the partial correlation coefficients (Spearman’s r_s) for the younger (4–6 years) and older (6–8 years) groups. We used age (in months) as a control variable.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Bivariate correlation coefficients (Spearman’s r_s) for the younger (4–6 years) and older (6–8 years) age groups.

https://doi.org/10.1371/journal.pone.0169510.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Partial correlation coefficients (Spearman’s r_s) for the younger (4–6 years) and older (6–8 years) groups.

https://doi.org/10.1371/journal.pone.0169510.t006

As can be seen from Table 5, for the younger group (4–6), there is a significant correlation between the FBT_2 judgment score and REL_2 (r_s = 0.33, p = .03), and a significant correlation between the FBT_2 judgment score and LST (r_s = .51, p = < .001). However, as shown in Table 6, correlations between the FBT_2 judgment score and REL_2 become insignificant when we control for LST (r_s = —.001, p = .99) but remain significant when we control for age (r_s = .31, p = .04) and for WST (r_s = .30, p = .04). On the contrary, the correlation between the FBT_2 judgment score and LST remains significant when we control for age (r_s = .50, p < .001), REL_2 (r_s = .41, p = .006) and WST (r_s = .52, p < .001). For the older group (6–8), none of the tasks show significant correlations with the FBT_2 judgment score. The lack of significant relationships between LST and older children’s FBT_2 judgment scores (range 0–2) is due to the fact that the older children already performed well in providing judgment answers, so there is a lack of enough variation in the data.

Similar to the FBT_2 judgments scores, for the younger age group, there is a significant correlation between the FBT_2 justification score and REL_2 (r_s = .48, p = .001), and a significant correlation between the FBT_2 justification score and LST (r_s = .59, p < .001). Moreover, there is a marginally significant correlation between the FBT_2 justification score and WST (r_s = .29, p = .07). As partial correlations reveal in Table 6, only the correlation between the FBT_2 justification score and LST remains significant when we control for age (r_s = .58, p < .001), REL_2 (r_s = .41, p = .006) and it even increases somewhat when we control for WST (r_s = .63, p < .001), previewing the results of the subsequent regression analyses.

For the FBT_2 justification scores in the older age group, there is only a significant correlation between the FBT_2 justification score and LST (r_s = 0.37, p = .009). As shown in Table 6, the correlation between the FBT_2 judgment score and LST remains significant when we control for age (r_s = .34, p = .01), REL_2 (r_s = .32, p = .02) and WST (r_s = .35, p < .01).

Moreover, for both FBT_2 judgment and justification scores, we found a significant high correlation between LST and REL_2 (r_s = .66, p < .001). These correlations are still significant when we control for age and the other tasks that we used in this study, indicating that LST and REL_2 share a considerable amount of variance.

Predicting the Second-Order False Belief Task Score (FBT_2)

The results of the regression models that best predict second-order false belief judgment and justification scores for both the younger (4–6 years) and the older (6–8 years) age groups are presented in Table 7. We proceeded as follows with our model construction for the younger group: In model 1, we entered the control variable “age in months” in order to account for any more fine-grained age differences within the younger age group, as well as the second-order relative clause task (REL_2), and the complex working memory task (LST), because we had found significant bivariate correlations between the FBT_2 judgment score and these two tasks. Moreover, even though we did not find strong significant correlations between the simple working memory task (WST) and FBT_2 judgment and justification scores, we kept WST in our models because we had a specific prediction that LST is more related to second-order false belief reasoning than WST is, based on the serial processing bottleneck hypothesis. In Model 1, the effects of age in months (B = 0.05, SE = 0.04, p = .22), REL_2 (B = -0.51, SE = 0.36, p = .16), and WST (B = 0.68, SE = 0.46, p = .14) were insignificant; only the effect of LST was significant (B = 3.44, SE = 1.35, p = .01). As shown in Table 5 and Table 6, considering the strong correlations between REL_2 and LST, and based on the result of Model 1 that REL_2 is not significant, we constructed Model 2 by excluding REL_2 from Model 1. As can be seen from Table 7, the effect of LST on the FBT_2 judgment score is significant (B = 2.22, SE = 0.82, p = .007), and the effect of WST is insignificant (B = 0.39, SE = 0.40, p = .33). The model comparison of Model 1 (Akaike Information Criterion (AIC) = 82.60) and Model 2 (AIC = 82.77) was less than 2. Because a simpler model is preferred over a more complex one, we presented the results of Model 2 in Table 7.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Predicting second-order false belief task judgment and justifications scores for both younger (4–6) and older (6–8) age groups.

https://doi.org/10.1371/journal.pone.0169510.t007

We followed the same procedure as explained above for the prediction of the younger age group’s FBT_2 justification scores. We entered the control variable “age in months”, REL_2, LST, and WST into Model 3. In Model 3, the effects of age (B = - 0.006, SE = 0.04, p = .89), and REL_2 (B = - 0.10, SE = 0.30, p = .75) were insignificant, and the effects of LST (B = 2.05, SE = 0.73, p = .005), and WST (B = 1.18, SE = 0.53, p = .03) were significant. Based on the results of Model 3, we constructed Model 4 by excluding REL_2 from Model 3. As can be seen from Table 7, the effects of LST (B = 1.91, SE = 0.57, p < .001) and WST (B = 1.12, SE = 0.49, p = .02) on the FBT_2 justification score are significant. Because Model 4 (AIC = 90.04) is a simpler model than Model 3 (AIC = 91.94), we presented the results of Model 4 in Table 7.

For the older group (6–8 years), we had found a significant correlation only between LST and the FBT_2 justification score, and for them, none of the tasks were significantly correlated with the FBT_2 judgment score due to the lack of variation. Although none of the tasks were significantly correlated with the FBT_2 judgment score, because the younger group’s (4–6 years) final model to predict the judgment score includes LST and WST, we constructed Model 5 to predict the older group’s FBT_2 judgment score by adding LST and WST. As shown in Table 7, in line with the lack of significant correlations, both LST and WST’s effects were insignificant in the model.

Similar to the previous procedures, in order to predict older children’s FBT_2 justification scores, we constructed Model 6 by entering the control variable “age in months”, WST, and LST. The effect of LST is significant when we control for age (B = 0.79, SE = 0.33, p = .02), and the effect of WST is insignificant (B = -0.05, SE = 0.34, p = .88).

These results suggest that the main predictor of second-order false belief reasoning is not syntactic recursion and word span task but complex working memory.