Neurodevelopmental Correlates of Theory of Mind in Preschool Children


  • We thank the families from the Kingston, Ontario community who graciously volunteered their participation. We also thank Phil Zelazo, Don Tucker, Nathan Fox, and Rebecca Saxe for helpful advice and discussion of the methodology, statistical analyses, and interpretation of the data. This research was supported by a CFI New Opportunities award and NSERC Discovery Grant awarded to Sabbagh, and by NSERC undergraduate summer fellowships awarded to Bowman and Evraire. Bowman is now at University of Michigan, and Evraire is now at University of Western Ontario.

concerning this article should be addressed to Mark A. Sabbagh, Psychology Department, Queen's University, Kingston, Ontario, Canada, K7L 3N6. Electronic mail may be sent to


Baseline electroencephalogram (EEG) data were collected from twenty-nine 4-year-old children who also completed batteries of representational theory-of-mind (RTM) tasks and executive functioning (EF) tasks. Neural sources of children’s EEG alpha (6–9 Hz) were estimated and analyzed to determine whether individual differences in regional EEG alpha activity predicted children’s RTM performance, while statistically controlling for children’s age and EF skills. Results showed that individual differences in EEG alpha activity localized to the dorsal medial prefrontal cortex (dMPFC) and the right temporal–parietal juncture (rTPJ) were positively associated with children’s RTM performance. These findings suggest that the maturation of dMPFC and rTPJ is a critical constituent of preschoolers’ explicit theory-of-mind development.

The facility with which we both create and negotiate our social worlds is owed in part to having a theory of mind—the understanding that people’s observable actions are motivated by internal mental states (e.g., beliefs, desires) that are related to, but ultimately distinct from, reality (Perner, 1991; Wellman, 1990). A key entailment of a mature theory-of-mind understanding is that mental states—particularly epistemic mental states such as beliefs and knowledge—are person-specific representations of the world. This understanding is sometimes called a “representational” theory of mind (RTM) and is typically indexed by a canonical battery of tasks, including the “false belief” task in which children are asked to predict how they or another would act on the basis of a belief that does not comport with reality (e.g., Perner, 1991). Recent research using looking-time paradigms has suggested that an implicit appreciation of RTM may be present in infants as young as 12 months old (e.g., Onishi & Baillargeon, 2005; Surian, Caldi, & Sperber, 2008). However, an explicit understanding of RTM seems to undergo a more protracted time course, emerging between the ages of 3 and 5 years old (Wellman, Cross, & Watson, 2001).

The timetable of explicit RTM development is roughly the same across cultures; although there are some subtle and important variations, 3-year-olds typically fail RTM tasks whereas 5-year-olds show near-ceiling performance (Callaghan et al., 2005; Liu, Wellman, Tardif, & Sabbagh, 2008; Sabbagh, Moses, & Shiverick, 2006). One known exception to this stereotyped developmental timetable comes from autism, a neurodevelopmental disorder in which performance on marker tasks of explicit RTM development is particularly impaired (Baron-Cohen, 2001; Peterson, Wellman, & Liu, 2005). These findings of cross-cultural synchrony, combined with the relatively specific impairments seen in autism suggest that specific neuromaturational factors may be associated with explicit RTM development. The goal of this study was to gain an initial characterization of the neurodevelopmental correlates of RTM in preschoolers.

Electroencephalogram Measures of Neurocognitive Development

Electrophysiological measures provide one of the most widely used and reliable windows on young children’s neurocognitive development. Of particular interest has been developmental changes in the “alpha band” (6–9 Hz) of children’s electroencephalogram (EEG). From infancy through the preschool years, alpha gradually becomes the highest amplitude resting rhythm over all regions of the scalp (Marshall, Bar-Haim, & Fox, 2002). Alongside the changes in alpha amplitude (power), there are also important regional changes in the EEG alpha coherence, a measure of the extent to which spectral EEG is correlated at any two electrode sites (Thatcher, 1994; Thatcher, Walker, & Guidice, 1987), In general, increases in EEG coherence reflect increases in synchronized neural firing either within or across neural populations (Nunez, 1995). When applied in the developmental context, increases in resting alpha coherence are generally thought to reflect functional, maturational changes in the organization of neurocognitive systems (Thatcher, 1992).

Recent advances in EEG analysis have made it possible to use the cross-spectral matrix (essentially a matrix of all possible pairwise coherence measures within an EEG recording montage), to estimate the intracerebral sources of spectral EEG power. One such technique is standardized low-resolution electromagnetic tomography (sLORETA) that computes current density throughout the solution space (6,237 5 mm3 voxels within gray matter and hippocampus; Pascual-Marqui, 2002). The sLORETA method distinguishes itself from some other source localization solutions such as brain electrical source analysis, in that it does not seek a fixed number of sources of scalp activity; instead, it estimates the current density at all points in the solution space. The general sLORETA technique has been used extensively and validated in event-related potential (ERP) studies (see Pascual-Marqui, Esslen, Kochi, & Lehmann, 2002, for a review). More relevant to the present article, recent research with adults has used a version of the sLORETA method to show how group differences in regional current density estimated from spectral EEG (including alpha) predicts aspects of cognitive and affective information processing (Pizzagalli, Peccoralo, Davidson, & Cohen, 2006; Pizzagalli, Sherwood, Henriques, & Davidson, 2005).

Because of the previously described relations between resting EEG alpha and functional neural development, we reasoned that regional increases in current density estimates can be taken to reflect increased synchronous activity within associated neural assemblies, which in turn can be attributed to ongoing neurodevelopmental processes (Thatcher, 1992). Thus, in the present study we sought to investigate the neurodevelopmental correlates of RTM development by examining the extent to which individual differences in preschoolers’ regional sLORETA current density estimations (based upon resting EEG alpha activity) were positively associated with their performance on a battery of RTM tasks.

Neurodevelopmental Bases of RTM Development: Predictions

There is now a considerable body of literature investigating the neural bases of RTM reasoning in adults. An in-depth review of this literature is beyond the scope of this report. In brief, this literature has identified several neural regions that are most consistently associated with the kind of RTM reasoning that children acquire during the preschool year (e.g., false-belief reasoning), including the dorsal medial prefrontal cortex (dMPFC), the right and left temporal–parietal juncture (TPJ), and the precuneus (PC; Gallagher & Frith, 2003; Saxe, 2006). Intriguingly, some recent research suggests that these same regions (along with some other regions) provide the neural substrates for RTM reasoning in school-aged children (Kobayashi, Glover, & Temple, 2007). Accordingly, a straightforward developmental hypothesis is that preschoolers’ RTM development is associated with the functional maturation of these regions.

There are a couple of reasons, however, to question whether this straightforward hypothesis would bear out. First, and perhaps most fundamentally, it is not clear whether it is appropriate to assume that the functional neural specializations seen in expert adults (or older children) would be the same as those seen in preschoolers who are only acquiring an explicit RTM reasoning. Karmiloff-Smith (1997) noted that it is not possible to determine solely from research with adults whether observed content-specific functional specializations are present from the outset of development or whether they are the outcome of a more protracted developmental process. A second related concern is that even if RTM reasoning is associated with generally homologous regions in adults and children, it is possible that their contributions would not be as domain-specific in children as they appear to be in adults. These regions may perform more domain-general computations and their specialization for RTM reasoning might emerge with development.

While these concerns might apply broadly, they can currently be most well articulated with respect to the role that dMPFC might play in RTM development. A recent review suggests that medial frontal regions, extending into the dMPFC, are especially involved in some aspects of executive functioning (EF; Ridderinkhof, Ullsperger, Crone, & Nieuwenhuis, 2004). In particular, medial frontal regions are recruited when people are required to negotiate situations in which there is either response conflict or decision uncertainty. What makes this particularly interesting from a developmental perspective is that there is now extensive evidence showing that response-conflict EF tasks are strongly associated with preschoolers’ RTM development (Carlson & Moses, 2001; Hala, Hug, & Henderson, 2003; Hughes, 1998; Perner, Lang, & Kloo, 2002). There are several reasons that response-conflict EF skills associated with medial frontal functioning might be intrinsically connected with children’s abilities to reason about RTM (see Moses, 2001, for a fuller treatment). In short, RTM tasks typically have the methodological task demand of requiring children to inhibit their habitual tendency to refer to the true location of an object to reference the falsely believed location. Furthermore, once children are able to conceptualize that epistemic states can mismatch reality, they have to inhibit their usually correct strategy of assuming that epistemic states match reality. In any case, the fact that response-conflict EF and RTM are associated in development, and both are associated with medial prefrontal cortex, raises the possibility that the contribution of dMPFC to RTM development is mediated though its contribution to response-conflict EF.

Thus, in this first investigation of the neurodevelopmental bases of preschoolers’ RTM reasoning, we included in our research design a battery of response-conflict EF tasks that are typically associated with RTM development in preschoolers. This allowed us to statistically control for children’s EF skills and determine whether any potential association between current density in dMPFC and RTM development can be accounted for by a common association with EF in development.



Eighty-eight typically developing 4-year-olds (range = 48–62 months, = 53.57, SD = 3.71) were recruited to the study. Although no systematic demographic data were collected, participants were from a predominantly European Canadian, middle-class, and military background, reflecting the demographics of the region in southeast Ontario, Canada from which they were drawn. All children were tested in two laboratory visits. The two visits occurred within 2 weeks of each other. Participants’ age in months was calculated from their birth date at the time of the first session. Parents reported that all participants were born within 2 weeks of their original due date, were developing typically, and had no history of neuropsychological disease or trauma.

Our criteria for retaining participants were that they had to complete all of the focal tasks in the battery, and contribute at least 50 s of artifact-free EEG data for use in analyses. The EEG criterion is comparable with that used in previous child EEG studies (Fox, Rubin, Calkins, & Marshall, 1995). Of the initial 88 participants, 10 children were excluded because they failed to complete the behavioral battery. An additional 49 were excluded because they did not contribute sufficient EEG data. To help explain why such a large number of children failed to contribute sufficient EEG, we checked our records and found that of our first 30 participants, 27 did not meet the EEG criterion. This high exclusion rate in the first third of the study was likely attributable to our inexperience collecting EEG data with 4-year-old children. After the first 30 participants, we were more successful on this front and were able to retain about half of all participants tested.

Thus, the final sample consisted of 29 children (9 boys and 20 girls). Ages ranged from 48 to 62 months (= 54.00, SD = 3.70). The large number of exclusions raised questions about whether our final sample might show a selection bias, thereby limiting the generality of the findings. To address this possibility, we report following analyses comparing the included and excluded samples on the behavioral measures.

Measures and Materials

All of the behavioral tasks that we used in this study have been used several times in other studies. Thus, here we provide only a brief description of each task and how it was scored for purposes of the present analyses.

RTM Battery

Knowledge access (Wellman & Liu, 2004).  Children were shown a small wooden box and asked what they thought was inside it. After children took a guess, or responded that they did not know, they were shown that the box contained a toy elephant. Children were then asked if the puppet, “Tiger,” knew what was inside the box given that he had never seen inside. Children passed the task if they answered that Tiger did not know what was inside the box (score = 0–1).

False belief: Contents (Gopnik & Astington, 1988).  Children were shown a Smarties box and asked what they thought was inside. After children responded that they thought the box contained Smarties or chocolates, they were shown that the box contained crayons. Children were then asked what “Mickey Mouse” (a puppet) would think was inside the Smarties box given that Mickey had never seen inside the box. Children passed the task if they answered that Mickey would think the box contained Smarties (score = 0–1).

False belief: Location (Wimmer & Perner, 1983).  Children were shown a scenario in which one puppet, “Heidi,” hid a plane under her bed, and then another puppet, “Andy,” moved the plane from the bed to the toy box whereas Heidi was not looking. Children were then asked where Heidi would look for her plane. Children passed the task if they answered that Heidi would look for her plane under the bed where she left it (score = 0–1).

Appearance–reality (Flavell, Green, & Flavell, 1986).  Children were shown a sponge that was painted to look like a rock and asked what they thought it was. After children responded that they thought it was a rock, they were shown that it was a sponge. Children were then asked again what the object looked like. Children passed the task if they correctly answered that it looked like a rock (score = 0–1).

EF Battery

Grass–snow Stroop (Carlson & Moses, 2001).  Although grass is typically associated with the color green, and snow is typically associated with the color white, children were instructed to reverse these associations and point to a green card when the experimenter said “snow” and to a white card when the experimenter said “grass.” The final score was the proportion (percentage) of correct responses over 16 trials. Sometimes children made multiple responses on a single trial, but only their first responses were scored.

Bear–dragon (Reed, Pien, & Rothbart, 1984).  Children were instructed to follow the commands given by a bear puppet (e.g., “touch your nose”) but to ignore any commands from a dragon puppet. All commands required hand actions only. On the “dragon trials,” children received a score of 0 if they fully complied to the dragon’s command, a score of 1 if they performed a partial movement in response to the dragon’s command, a score of 2 if they performed a movement that did not correspond to the dragon’s command, a score of 3 if they did not respond to the dragon’s command but used a strategy to help them avoid compliance (such as saying “no” to the dragon or sitting on their hands), and a score of 4 if they did not move at all in response to the dragon’s command and did not use any strategy (range = 0–20).

Dimensional-change card sort (Zelazo, 2006).  Children were instructed to sort cards that varied on two dimensions: (a) color (red and blue) and (b) shape (boats and rabbits). First, children were instructed to sort the cards according to shape (i.e., boats in one basket, rabbits in the other). Then, they were asked to switch and sort the cards according to color (i.e., red in one basket, blue in the other). Children were given a score based on the number of postswitch sorts that clearly demonstrated they were sorting based on the second dimension (i.e., color; range = 0–3).

Less is more (Carlson, Davis, & Leach, 2005).  Children were shown two trays: one containing a large amount of candy (i.e., five jelly beans) and one containing a small amount of candy (i.e., two jelly beans). Children were told that when they pointed to a tray, the candy in that tray would go into “Naughty Monkey’s” cup, and they would get the candy in the other tray to put in their cup. Children received a score of 0 if they pointed to the tray with the larger amount, a score of 1 if they pointed directly to the smaller amount. Scores were summed across 16 consecutive trials (range = 0–16).

EEG Recording

Rocket ship–spiral line video.  During EEG recording, participants watched a 6-min video that had two components: (a) a still picture of a rocket ship and (b) an animation of a green line that mapped out a spiral (alternating expanding and contracting spirals with each presentation). The rocket ship and spiraling line components were each 30 s in length and were each presented six times, in alternation. The video began with an expanding spiraling line clip. Because it is difficult for preschoolers to sit still, the alternating video allowed us to instruct children that it was okay to move a little during the spiraling line segments, but to do their best to stay still while the picture of the rocket ship was on the screen. This method was adopted from previous research (Fox et al., 1995) and is the most well-established protocol for collecting baseline or resting EEG alpha with preschoolers.

EEG acquisition and analysis.  Electroencephalogram was recorded from the scalp with a 128-channel Geodesic Sensor Net (EGI, Eugene, OR). The net consisted of 128 carbon-fiber electrodes knitted into an elastic geodesic tension structure that when applied, distributed electrodes evenly over the scalp. Each electrode was positioned relative to the vertex electrode in a geodesic montage that stretches around the sphere of the head. Before application, the net soaked for 5 min in a mild electrolyte solution to aid conductance.

Electroencephalogram activity at all channels was recorded referenced to the vertex electrode (Cz), sampled at 500 Hz, and digitally filtered between 0.01 and 200 Hz (time constant = 1 s). EEG was recorded continuously throughout the video, and any necessary electrode readjustments were made during sections of the video for which no data analyses were planned. Impedances were maintained below 30 kΩ throughout recording.

The raw EEG recordings were filtered (60-Hz notch) and edited to include only recordings that were made while participants quietly viewed the still photograph of the rocket ship. These data were then further divided into smaller 2-s segments, which were in turn submitted to a software algorithmic artifact rejection program (Vision Analyzer; Brain Vision GMBH, Gilching, Germany) that combed the data for evidence of artifact (gradient threshold = ±50 μV in 100 ms, amplitude threshold = ±200 μV, global maximum difference threshold = ±300 μV). Hand-coding of 25% of the EEG records confirmed that these criteria reliably identified artifact due to blink, eye movement, and participant movement. Segments that contained artifact were removed from further analyses. Only participants who contributed at least 25 good segments of EEG (i.e., 50 s of data) were considered for analysis. The artifact-free EEG segments were then transformed to average reference to ensure accurate source-localization.

Cross-spectral matrices were created for the single-subject average-referenced data in 0.5-Hz frequency intervals from 5.5 to 9.5 Hz, thereby ensuring we captured activity in the preschoolers’ alpha band (i.e., 6–9 Hz). The nine resulting cross-spectral matrices were then used to compute three-dimensional distributions of the standardized current density using standardized estimates of the minimum norm inverse solution as applied by sLORETA (see Pascual-Marqui, 2002, for details). The sLORETA transformation results in activation values for 6,237 “voxels” (5 mm3) located within cortical gray matter and hippocampus, as defined by the Probability Atlas from the Montreal Neurological Institute. Preliminary analyses showed that participants’ sLORETA current density estimations across the nine frequency bins were highly intercorrelated at all voxels (Cronbach’s α = .88–.99, = 0.95), and so the sLORETA activation values were collapsed across the nine frequency bins to create a final reliable measure of the regional current-source activation values in the 6–9 Hz (alpha) band at each voxel for each participant.


This study was part of a larger investigation of the neurodevelopmental correlates of children’s social-cognitive development, and thus, in addition to the RTM and EF batteries described above, the larger study included a number of additional tasks that were not analyzed in this study. Most prominently, there were a number of behavioral measures of children’s pretend play including two structured role playing activities, two versions of the “Moe” task (Lillard, 1993), a self-pretense task (Mitchell & Neal, 2005), and a measure of children’s narrative absorption (Rall & Harris, 2000). We also included two tasks that are considered precursors to RTM understanding, including children’s understanding of diverse desires and diverse beliefs (Wellman & Liu, 2004). Finally, we included a standard measure of children’s vocabulary development (PPVT–III; Dunn & Dunn, 1997). Given the number of behavioral measures, children were asked to visit an on-campus laboratory for two sessions. We followed standard practice for individual differences studies and administered the tasks in a fixed order to all participants (see Carlson & Moses, 2001).

Session 1

After obtaining informed consent, we recorded electrophysiological data (EEG) from the scalp using the presoaked sensor net. Participants sat in a stable chair while an experimenter applied the net and adjusted it to ensure correct placement and good contact between the electrodes and the scalp. The final net placement was photographed and later examined to verify that net placement was correct at the outset of the session. Participants were then given the instructions for watching the video described above, the lights were dimmed for optimal video viewing, and EEG recording began. Parents were allowed to remain in the room with their child during recording, and both parent and child were asked to be as quiet as possible throughout the procedure. Net placement, adjustment, instructions, and recording took approximately 20–25 min.

After EEG recording, the experimenter removed the net, and participants moved to a different room for behavioral testing. For the first session, behavioral tasks were administered in the following order: (a) false belief: contents, (b) diverse desires, (c) diverse beliefs, (d) knowledge access, (e) grass–snow, (f) role-play activity, (g) narrative task, and (h) Moe task.

Session 2

Only behavioral data were collected in Session 2. After obtaining informed consent, participants were tested with a single experimenter in the same room as Session 1. The tasks were administered in the following order: (a) bear–dragon, (b) false belief: location, (c) card sort, (d) appearance-reality, (e) role play activity, (f) self-pretense, (g) Moe task, (h) less is more, and (i) PPVT. Session 2 took approximately 45 min.


Behavioral Data

For the RTM battery, preliminary analyses showed that performance was in line with previous research that has used these tasks (see Table 1). Scale reliability analyses showed that the RTM battery had moderate internal consistency (α = .623), which is also comparable with theory-of-mind batteries used in previous research. Children’s performance on the final battery was associated with children’s age in months, r(27) = .444, = .016.

Table 1. 
Children’s Performance on the Theory-of-Mind (ToM) and Executive Functioning (EF) Batteries
 Included (n = 29)Excluded (n = 49)Test of difference
  1. aFigures in this category represent percent passing each task. bFigures in this category represent mean percent trials correct (with standard deviation) on each task.

ToM tasksa
 Knowledge access72.4%94.0%χ2(1) = 8.27, = .004
 False belief: contents37.9%52.0%χ2(1) = 1.51, = .219
 False belief: location72.4%66.0%χ2(1) = 0.36, = .547
 Appearance–reality69.0%66.0%χ2(1) = 0.07, = .785
 Battery (mean total of 4)2.55 (1.30)2.76 (1.11)t(76) = 0.734, = .465
EF tasksb
 Grass–snow78.22% (27.67)78.68% (25.53)t(76) = 0.07, = .944
 Card sort72.41% (42.79)72.67% (42.16)t(76) = 0.02, = .984
 Less is more79.96% (27.11)77.62% (25.71)t(76) = 0.38, = .705
 Battery (mean of 3, with SD)79.91% (22.09)75.84% (22.89)t(76) = 0.768, = .445

For the EF battery, preliminary analyses revealed ceiling level performance and low battery correlation for the bear–dragon task. Upon closer inspection of the preschool EF literature, we discovered that very strong performance on bear–dragon is common for children in this age group (see Carlson, 2005), and so we omitted it from the task battery. Thus, our EF battery was composed of the three remaining tasks: grass–snow, card sort, and less is more. This battery also showed moderate internal consistency (α = .554), but was not significantly associated with children’s age, r(28) = .113, = .559.

The results for each of the behavioral tasks from the final RTM and EF batteries are presented in Table 1, along with statistical comparisons of the groups who were included in the final EEG analyses versus those who were excluded. Apparent from these analyses was that the included and excluded groups differed significantly on only one task—that is, children in the EEG excluded group performed better on the knowledge access task than did children who were included in the final sample. The overall parity between the included and excluded groups was most powerfully demonstrated in the comparison of children’s performance on the battery aggregates, which were nearly identical across groups. These findings provide some confidence that there was nothing particularly special about the included versus excluded group, and thus strengthen our ability to generalize from our findings.

For the final sample, children’s performance on the RTM battery was highly correlated with their performance on the EF battery, r(27) = .575, = .001, and this relation was unaffected by statistically controlling for age, r(26) = .590, = .001.

sLORETA Analyses

Our main goal in these analyses was to determine whether individual differences in regional current-source estimations predicted children’s performance on the RTM battery while controlling for children’s age and performance on the EF battery. Preliminary inspection of the sLORETA data showed that activation values were positively skewed, and so were submitted to a natural log (ln) transformation prior to the focal analyses. All analyses were conducted using purpose-designed scripts that called functions from the general and statistics toolboxes of MATLAB (The MathWorks, Natick, MA).

Whole Brain Analyses

Our analytical approach was to perform voxel-wise partial correlations measuring the relationship between sLORETA activation values and children’s RTM performance while statistically controlling children’s age and EF performance. To establish a significance criterion for the whole brain analyses, we conducted a permutation test in which 2,000 random permutations of the behavioral data were submitted to the partial correlation analyses with the sLORETA activation values. Results of this showed that voxel-wise tests with a relatively lenient p-value criterion (< .01), combined with a cluster-size criterion of 20 contiguous voxels was associated with an acceptable family-wise alpha level (< .05).

Using these criteria, we identified five regions at which current-source density was positively associated with performance on the RTM tasks (see Table 2). The strongest associations were located within the dMPFC and the right TPJ (rTPJ) (see Figure 1). Three additional clusters also met the significance criteria (see Figure 2). One cluster was located in motor and premotor cortex, and the correlation with RTM performance was strongest in the precentral gyrus. A second cluster was within middle occipital regions, and the correlation with RTM was maximal at the cuneus. The third cluster included posterior inferior temporal regions and fusiform gyrus, but the correlation with RTM was maximal at the posterior inferior temporal gyrus.

Table 2. 
Brain Regions in Which sLORETA Current Density Estimates Predicted Theory-of-Mind Performance
Region (MNI coordinates of max. r) L/RIncluded BAskMax. rpartial
df = 25
Agg. rpartial
df = 25
  1. Note. sLORETA = standardized low-resolution electromagnetic tomography; BA = Brodmann area; dMPFC = dorsal medial prefrontal cortex; rTPJ = right temporal–parietal juncture.

  2. **< .01.

dMPFC (5, 55, 40)L/R8/944.585**.580**
rTPJ (55, −55, 30)R4035.552**.537**
Precentral gyrus (50, −15, 45)R3/4/630.526**.521**
Cuneus (20, −95, 15)R17/18/1933.525**.529**
Inferior temporal (50, −70, −5)R19/3728.511**.513**
Figure 1.

 Thresholded statistical map and scatterplot of partial correlations measuring the relation between RTM reasoning and sLORETA estimates of current density at (a) dMPFC and (b) rTPJ. Only significant voxels (< .01) are shown (in white). Voxels in all figures are shown projected onto a template structural MRI to illustrate their neuroanatomical locations.
Note. RTM = representational theory of mind; sLORETA = standardized low-resolution electromagnetic tomography; dMPFC = dorsal medial prefrontal cortex; rTPJ = right temporal–parietal juncture; MRI = magnetic resonance imaging.

Figure 2.

 Additional regions that contained clusters of voxels showing significant relations between sLORETA estimates of current density and RTM reasoning. Only significant voxels (< .01) are shown (in white).
Note. sLORETA = standardized low-resolution electromagnetic tomography; RTM = representational theory of mind.

To further characterize these effects, we averaged across the significant voxels in each region to create one aggregate measure of activity in each region for each subject. Each of these aggregates were strongly correlated with RTM reasoning, with age and EF performance statistically controlled (see Table 2). Furthermore, the aggregates showed were largely intercorrelated—only dMPFC and rTPJ did not show a significant correlation with one another (see Table 3). Because of the high degree of intercorrelation among the aggregates, we conducted a stepwise regression to determine the most parsimonious model of the relation between regional current density and RTM performance. For this analysis, age and EF were entered in the first block, and the aggregate current density estimations were added stepwise in the second block. Age and EF performance each significantly predicted RTM reasoning in the first block, F(2, 26) = 11.825, < .001, Radj2 = .436. The stepwise analysis of current-source density aggregates showed that adding aggregates from the dMPFC and rTPJ made a significant unique contribution, ΔR2 = .235, Fchange(2, 24) = 10.315, = .001. The final model (see Table 4) including age, EF performance and dMPFC and rTPJ aggregates explained 67.1% of the variance in RTM reasoning, F(4, 24) = 15.306, < .001. The additional regions (e.g., cuneus, inferior temporal, and precentral) were excluded from the model in the stepwise analysis because they did not explain unique variance in RTM reasoning, thereby suggesting that the contribution of those other regions could be largely accounted for by their associations with either dMPFC or rTPJ.

Table 3. 
Partial Correlations Between Cluster Aggregates Identified in Whole-Brain Analysis (Controlling for Age and EF)
VariableInferior Temp.dMPFCrTPJPrecentral
  1. Note. EF = executive functioning; dMPFC = dorsal medial prefrontal cortex; rTPJ = right temporal–parietal juncture.

Inferior Temp. .389.779.662
dMPFC  .366.549
rTPJ   .746
Table 4. 
Regression Coefficients for Final Model Predicting RTM Development From Age, Executive Functioning, and Current-Source Density
VariableStandarized βt valuep value
  1. Note. RTM = representational theory of mind; dMPFC = dorsal medial prefrontal cortex; rTPJ = right temporal–parietal juncture.

Executive functioning.5744.994<.001
dMPFC aggregate.3172.707.012
rTPJ aggregate.2922.402.024

Dorsal Medial Prefrontal Cortex and Right Temporal–Parietal Juncture

Of the regions identified in the whole-brain analyses, the dMPFC and rTPJ have been most consistently implicated in several adult and school-aged children’s theory-of-mind reasoning. Thus, we conducted further analyses to determine whether the regions we identified were essentially homologous with those shown in previous research. We began by defining regions of interest in the dMPFC and rTPJ as 2-cm-diameter sphere drawn around the average reported MNI coordinates for each area showing peak activation in recent theory-of-mind studies (dMPFC: 9, 54, 36, rounded to 10, 55, 35 to match our resolution 5 mm3; rTPJ: 56, −54, 19, rounded to 55, −55, 20). This procedure resulted in a 32-voxel region of interest for dMPFC, and a 21-voxel region in the rTPJ. The reason for the difference in the number of voxels included in each region is that the rTPJ has less cortical gray matter than the dMPFC region. We then statistically compared the clusters we identified with these regions of interest. The results of this analysis are summarized in Figure 3. Descriptively speaking, 22/32 (68.8%) voxels in the dMPFC and 12/21 (57.1%) voxels included in the region of interest overlapped with the activation cluster we identified in the whole-brain analysis. In the dMPFC, our cluster extended to slightly more inferior regions. For the rTPJ, our cluster included voxels that were slightly superior and posterior to the region of interest, and did not extend to more inferior regions. Although we did not achieve complete overlap, permutation tests (2,000 iterations) showed that this degree of overlap is an extremely rare result by chance in both the dMPFC and the rTPJ (both ≤ .0005). This suggests that the dMPFC and rTPJ regions in which current-source density was positively correlated with RTM performance were likely homologous with those that have been shown to be active during RTM reasoning in adults.

Figure 3.

 Overlap between observed results and predicted regions of interest in (a) dMPFC and (b) rTPJ.
Note. dMPFC = dorsal medial prefrontal cortex; rTPJ = right temporal–parietal juncture.


Our main finding was that individual differences in estimated current density attributed to the dMPFC and rTPJ were positively associated with preschool children’s performance on a battery of RTM tasks. In addition to these regions, we found some evidence that individual differences in postcentral gyrus, cuneus, and posterior inferior temporal regions also positively predicted RTM performance. We will discuss each of these findings in turn and discuss some implications and limitations of this study.

Dorsal Medial Prefrontal Cortex

The region of dMPFC that we found was critically associated with RTM performance in children is similar to the regions that have been associated with RTM reasoning in functional magnetic resonance imaging with adults and school-aged children. From the outset, we questioned whether any possible contributions of dMPFC to RTM development could be accounted for by the role that dMPFC might play in negotiating response conflict. On this question, we showed that current density attributable to dMPFC was related to RTM performance, even when controlling for performance on an EF battery that was also strongly correlated with RTM performance. Furthermore, other research from our laboratory has provided evidence that when the partial correlation analyses are conducted to assess the relations between current-source density and EF (controlling for age and RTM), the results show that current-source density at dMPFC are not associated with EF performance; instead, EF is associated with individual differences in current-source density attributed to cingulate (anterior and posterior) and ventral-medial prefrontal cortex (Sabbagh, Bowman, Evraire, & Ito, 2009). These findings suggest that increased current density attributable to this region of dMPFC is associated with the emergence of explicit RTM reasoning in young children, independent of the contributions that these increases might also make to response-conflict EF skills.

It is important to clarify that we do not wish to downplay the role of EF in preschoolers’ explicit RTM reasoning abilities. Indeed, our own findings showed that EF was associated with RTM reasoning, and that this association survived the regression analysis in which the EF-RTM relation was assessed while statistically controlling for age, and current-source densities in dMPFC and rTPJ. Clearly, then, EF (and presumably its associated neural substrate) makes some independent contribution to RTM performance in the preschool years. Our finding here helps to clarify the uniqueness of this contribution by highlighting that it is not directly through a common relation with a particular region of dMPFC. Likewise, these findings clarify that the role that the dMPFC plays in RTM reasoning cannot easily be attributed to the role that this region may play in response-conflict EF, even in preschoolers.

Our finding that the dMPFC plays a role in RTM reasoning that is independent of its possible association with EF is consistent with a number of studies suggesting that the neural substrates of RTM reasoning are dissociable from those associated with response-conflict EF, even in closely matched tasks (Kain & Perner, 2005; Saxe, Schulz, & Jiang, 2006). Further, they are consistent with some recent developmental data suggesting that the timetable of EF and RTM can be dissociated in development. For instance, Sabbagh et al. (2006) recently showed that Chinese preschoolers’ EF skills were more advanced than those in a sample of North American preschoolers; however, the two groups showed no differences in their RTM development (see also Oh & Lewis, 2008). Likewise, there are a number of reports now suggesting that RTM skills might vary across groups that are generally equated for their EF abilities (e.g., Liu et al., 2008). These cross-cultural dissociations between the developmental trajectories of EF and RTM may be attributable to the fact that the two skills may have partly dissociable neural substrates during the preschool years.

Yet, more research is required to better understand what the dMPFC does for explicit RTM development. Two different hypotheses about the role that dMPFC might play in RTM reasoning have emerged from recent reviews of the adult neuroimaging literature. Amodio and Frith (2006) suggest that the region of dMPFC that predicted RTM reasoning in the current study (what they call the anterior rostral medial frontal cortex) performs the computations associated with complex metarepresentational, and possibly emotional reasoning. This suggestion is based upon the fact that dMPFC activations are observed as adults perform false belief tasks, and in tasks in which participants make probabilistic judgments about others’ likely beliefs and intentions (e.g., judgments of others’ subjective experience of pain and emotion, economic games). Saxe (2006), in contrast, summarizes the available data by suggesting a more limited role for the dMPFC. Specifically, Saxe argues that the dMPFC is important for representing triadic attentional relations (i.e., representing the relations between self, other, and an object), but finds no conclusive evidence to suggest that the dMPFC generates or represents the propositional contents of mental states themselves. A clear direction for future research is to adapt paradigms like the one we used here to better understand the neural bases of preschoolers’ theory of mind development, based upon insights gleaned from the burgeoning adult literature.

Right Temporal–Parietal Juncture

Our findings showed that individual differences in current density estimations localized to rTPJ predicted preschoolers’ performance on the RTM battery. These findings confirm that the region that appears to be most consistently and specifically associated with RTM reasoning in adults is also implicated in explicit RTM development. Saxe and colleagues (Saxe, 2006; Saxe & Powell, 2006) have presented evidence to suggest that rTPJ performs the highly specific task of reasoning about representational mental states. Our findings are certainly consistent with the notion that rTPJ plays a critical role in RTM development; though, our study was not designed with the aim of licensing fine-grained inferences about this important region.

More generally, the hypothesis of a domain-specific role for rTPJ in RTM reasoning has not been explored in the developmental literature. However, a recent challenge has emerged from work in visual neuroscience noting that the region of rTPJ that is involved in RTM reasoning is isomorphic with regions that are critical for coordinating shifts of visual attention (Mitchell, 2008). When tested in the same subject, the same regions that are associated with attention shifting (in an ostensibly nonsocial context) tasks show considerable overlap with those recruited for RTM reasoning. These findings raise the intriguing possibility that the association of rTPJ with RTM may be attributable, at least in part, to a domain-general rather than domain-specific skill. Were an ontogenetic relation between attention shifting and RTM performance to exist, we suggest that an approach like the one we adopted here would be fruitful in clarifying the functional relation between rTPJ and RTM reasoning in the preschool years.

Additional Regions

We found evidence that variations in current density attributed to the cuneus, inferior temporal lobe or fusiform gyrus, and precentral gyrus were also associated with RTM development. Each of these areas has been linked with performance in social cognitive tasks, if not RTM reasoning more specifically. We will address briefly the possible contributions of each region to RTM development in turn.

Precentral Gyrus

Across several studies, the precentral gyrus has been implicated in tasks that require distinguishing one’s own from another’s visual perspective (Aichhorn, Perner, Kronbichler, Staffen, & Ladurner, 2006; Ruby & Decety, 2001). Although the conceptual links between perspective taking and theory of mind are clear in that both would seem to require reasoning about other’s psychological states (Flavell, 1999), there is relatively scant empirical evidence to suggest that tasks tapping the two constructs are related in development. Nonetheless, it does seem likely that the cognitive operations associated with suppressing one’s own perspective are critical for performance on RTM tasks in children (Birch & Bloom, 2007), and even in adults under circumstances in which one’s own perspective is particularly salient (Keysar, Lin, & Barr, 2003).


Several studies of social-cognitive skills related to theory of mind have noted cuneus activation in at least some of their statistical comparisons (e.g., Vogeley et al., 2001). Even more interesting with respect to the present findings is that Kobayashi et al. (2007) found that activation in the cuneus distinguished theory-of-mind from non-theory-of-mind reasoning in children, but not in adults. These preliminary findings, together with ours suggest that the cuneus may make an especially important contribution to RTM development.

With respect to characterizing the contribution of the cuneus to RTM development, it can be noted that the cuneus is most reliably engaged during tasks that require reasoning about self knowledge (Aichhorn et al., 2006). One possibility, then, is that the relation we observed between RTM performance and current density variations in cuneus may reflect the extent to which RTM performance in children relies on reflecting upon one’s own knowledge states. The cuneus is also recruited in tasks that involve mental imagery (Kosslyn et al., 1999). Thus, a second possibility is that cuneus function is important to RTM reasoning because RTM tasks may require children to reimagine the scenario to answer correctly the question. In some respects, this latter explanation may be preferred because it potentially accounts for why the cuneus is implicated in RTM reasoning in children (but less reliably so in adults), for whom using mental imagery might provide some necessary assistance in recalling the foregoing events.

Inferior Temporal Lobe

In adults, participation in the now common “intentional attribution” task reliably elicits activity in the inferior temporal lobe extending into the fusiform gyrus (Brunet, Sarfati, Hardy-Bayle, & Decety, 2000; Vollm et al., 2006). In children, intention attribution is typically a precursor of explicit RTM reasoning, and is present quite early in development (Woodward, 2003). These precursors are ontogenetically associated with later RTM development. For instance, Wellman and colleagues (Wellman, Lopez-Duran, LaBounty, & Hamilton, 2007; Wellman, Phillips, Dunphy-Lelil, & LaLonde, 2004) have shown that performance in a looking time study designed to measure 12-month olds’ understanding of the link between emotion and intention was associated with those same children’s theory-of-mind development during the preschool years. Although the mechanism underlying this developmental relation is currently unclear, the fact that a developmental relation exists raises the possibility that the development of inferior temporal regions associated with intention attribution may play some crucial role in the development of the regions that are more specifically related to later RTM advances (i.e., rTPJ, dMPFC).


In general, these additional regions have been associated with cognitive abilities that might be either considered necessary for solving RTM tasks (i.e., mental imagery and suppressing first-person perspective), or conceptual precursors to RTM reasoning (i.e., intentional understanding). One possibility, then is that the computations associated with these regions may play a particularly important role in RTM development per se. Preliminary support for this hypothesis comes from our regression analyses in which we found that the contributions of these additional regions were mediated by their relations with either rTPJ or dMPFC. Although there are many possible explanations for this mediation effect, one is that the development of these regions somehow contributes to the functional maturation of dMPFC and rTPJ, which in turn make up the primary neural substrate for RTM reasoning. Longitudinal research may prove fruitful in further understanding the role that these additional regions play in the development of the neural systems critical for RTM development.


From the outset, we took as our main goal establishing the neurodevelopmental correlates of RTM reasoning in preschoolers. Of particular interest was whether regions that are known to play a role in RTM reasoning in adults would also be implicated in children. On this question, our evidence was clear. Individual differences in current density estimates attributable to dMPFC and rTPJ were correlated with RTM skills in preschoolers. Thus, the link between these brain areas and RTM development is present in children just as these skills are explicitly emerging. Further, insofar as variations in current density estimates reflect the ongoing functional maturation of these associated neural regions, our findings provide some evidence that the functional maturation of dMPFC and rTPJ is associated with RTM development.

This finding lays the foundation for the exploration of two interrelated questions in future research. The first concerns whether these functional specializations are a cause or a consequence of RTM development. Some researchers have articulated the hypothesis that humans have evolved a specific neurocognitive system that is dedicated to performing the computations that are associated with constructing and reasoning about representations of others’ mental states (Leslie, Friedman, & German, 2004). Although far from providing conclusive evidence for this hypothesis, the current findings are consistent with the possibility that the dMPFC and rTPJ are constituents of that specially evolved system, and that functional maturation of this system must reach a threshold level to enable explicit RTM reasoning. An alternative hypothesis is that the increased coherence in dMPFC and rTPJ reflect the outcome of a developmental process. Across domains, the neural systems associated with particular computations become reorganized and more circumscribed with expertise (Casey, Giedd, & Thomas, 2000). Thus, the relations we observed may be attributable to children’s RTM performance and neural coherence each reflecting increasing expertise with explicit RTM reasoning. Although we expect that it will be challenging, longitudinal research that takes an approach similar to that presented here may provide crucial insight into the causal relations among regional neural developments and RTM reasoning.

It is important to note that the functional maturation of any particular neural region is likely to be affected by both endogenous and experiential factors. There are reasons to think that both might play a particularly important role in establishing the functional maturation of dMPFC and rTPJ. With respect to endogenous factors, the development of the frontal lobes has been associated with maturational changes in dopaminergic functioning (Diamond, 2002). Insofar as there is some endogenous mechanism underlying the developmental trajectory of dopaminergic expression, it seems possible that this mechanism would in turn set constraints on the functional maturation of the neural systems associated with RTM reasoning. The presence of these kinds of endogenous neurodevelopmental constraints on RTM-related cortical development might help explain why, as mentioned at the outset of the article, the developmental trajectory of RTM looks so similar across cultures, and is specifically impaired in the case of autism.

With respect to experiential factors, there is now substantial evidence to suggest that several experiential variables, including parent–child conversations about mental states (Ruffman, Slade, & Crowe, 2002), family size (Lewis, Freeman, Kyriakidou, Maridaki-Kassotaki, & Berridge, 1996), and socioeconomic status (Pears & Moses, 2003) affect the timetable of RTM development. More dramatically, the development of RTM is greatly protracted in children with highly atypical experiences owing to being born either blind (Peterson, Peterson, & Webb, 2000) or born deaf into non-native signing families (Woolfe, Want, & Siegal, 2002). An important question, then, concerns whether and how these experiential factors might affect the maturation of the neural systems we identified as associated with RTM development. Understanding these fundamentally developmental processes will provide deep insights into how endogenous neuromaturational constraints might interact with experience to render the time course and trajectory of RTM development in the preschool years.


There are several limitations of this study. First, the sLORETA estimates are based upon volume conductance parameters (how electrical activity is conducted through the scalp, skull, and cerebrospinal fluid) that were developed from research on adults. There is some research to suggest that adult skulls are somewhat thicker than children’s skulls, though the age-related trends are modest and not as pronounced as individual differences within age groups (Simms & Neely, 1989). It is unclear as to how these minor differences in the conductance parameters might have affected our intracerebral current density estimates. Thus, some caution is warranted in interpreting the present findings.

Second, there are two factors that may have reduced the statistical power of our tests. One factor stems from the fact that we developed the transformation parameters for the sLORETA based upon average electrode locations because we did not have efficient technology for gathering electrode coordinates from participants individually. Relying on average sensor locations, though common in much EEG–ERP research, may have added noise to the group localization analyses and prevented us from detecting the involvement of additional areas. Another factor that may have reduced the statistical power of our tests detecting the involvement of additional neural regions was the relatively limited range scores possible on our RTM measure. Although neither of these factors call into question any of the findings we report, both leave open the possibility that other regions might have been detected, including regions that have been shown in the prior literature—such as the left temporal–parietal juncture and the precuneus or posterior cingulate—with more sensitive dependent measures.

Finally, it was unfortunate that our inexperience with pediatric EEG collection led to our excluding so many participants at the outset of our study. Although statistical analyses of the behavioral measures showed that there was no reason to believe that the participants we excluded were substantially different from the participants retained in the final sample, our high exclusion rate does raise some concern about the generality of these findings. Future work that replicates and extends these findings will be valuable in assuaging these concerns.


We used regional current density estimates of preschoolers’ resting EEG alpha to show that individual differences in the functional maturation of the dMPFC and rTPJ are associated with preschoolers’ performance on a battery of RTM tasks. Statistical analyses showed that these relations could not be accounted for by their relations with children’s age, or their relations with EF skills. In addition, we found that the cuneus, posterior inferior temporal lobe, and precentral gyrus were associated with RTM performance; however, statistical analyses suggested that their influence was mediated through their common associations with dMPFC and rTPJ. Taken together, these findings suggest that the functional maturation of the dMPFC and rTPJ may be a critical neural correlate of preschoolers’ RTM development. Future work is necessary to determine both the specific contributions of these regions to RTM development and the endogenous and experiential factors that affect their functional maturation.