Spontaneous belief attribution in younger siblings of children on the autism spectrum.

Stimuli

The stimulus was a video recording, which depicted five main events: two familiarization trials, two true belief trials (TB) and one final false belief (FB) trial (see Figure 1 ). We familiarized children with two events in which an actress reached through two doors for a toy strawberry placed on the left (first trial) or the right of two boxes (second trial). An audiovisual cue (the windows were illuminated and a chime sounded) was given and 2.5 s later the actor reached through the window and grasped the strawberry. The actor wore a visor so that her gaze direction could not betray the direction of her reach through the windows. The purpose of the familiarization trials was (a) to show the children that the actor’s goal was to reach for the object and (b) to teach the children that when the audiovisual cue was presented one of the windows was about to open. At the beginning of the two TB trials, a puppet monkey appeared and placed a banana in the left box (first TB trial) or the right box (second TB trial). After leaving the scene and 2.5 s after the cue appeared the actress reached through the door behind the box that contained the banana. The FB trial is depicted in Figure 1 . Crucially, in this trial, the actor turned away from the scene and the puppet monkey returned to remove the banana from the right side box, which induced a false belief in the actor. After the cue was given in this trial the scene froze for another 5 s. Because we could not counterbalance the locations of the banana in the FB trial within each outcome group (the outcome was not known at the time when the study was carried out), the same video clip was used for all participants.

Key events during the false belief trial. The last frame depicts the three areas of interest used for data extraction.

Procedure

An integrated Tobii (Stockholm, Sweden) T120 17” Eye Tracker was used to collect data on direction of gaze. Data were collected at 60 Hz. Tobii Studio was used to present the stimuli and for data analysis. Children sat on their own on a chair, at approximately 60 cm from the Tobii monitor. At this distance the diagonal of the screen subtended approximately 40°. A 5-point calibration was run before stimulus presentations. Children were told that they would see a movie about a cheeky monkey. An experimenter stood behind the child and encouraged her to look if she got distracted.

Data Reduction and Analysis

The 2 min 45 s long video was segmented into scenes of various lengths corresponding to the various important events. To measure anticipatory eye-movements in the second true belief trial we defined a 2.5-s interval after the visual cue appeared, which corresponded roughly to the time taken for the person to reach through the doors in both the familiarization trials and the TB trials. Because the actor never reached through the door in the false belief trial, a 5-s interval (until the end of the movie) was used for analysis. For clarity, details about the length of other intervals analyzed are given together with the results of those particular analyses, in the Results section. Three areas of interest (AOI) were defined manually for all scenes analyzed (see Figure 1 ), two covering the left and right doors and boxes and another one corresponding to the face. Cumulative looking time within areas of interest was calculated automatically using Tobii Studio software. Only fixations longer than 100 ms were included in the analyses. Data loss could occur during the video presentation at different time points (either due to looking away or to the eye-tracker not detecting the eyes despite the fact that the child was looking). We decided to only exclude children if they accumulated less than 20% data overall and not if only certain intervals had valid data, the consequence of which was that slightly different numbers of participants were entered in the analysis of different events (e.g., in the TB and the FB trial analysis).

Outcome Characterization of the At-Risk and Control Groups

Standard measures of cognitive development (Mullen Scales for Early Learning [MSEL]; Mullen, 1995 ) and adaptive development (Vineland Adaptive Behavior Scale [VABS]; Sparrow, Cicchetti, & Balla, 2005 ) were collected. The MSEL is a standardized direct developmental assessment that yields a standardized score ( M = 100, SD = 15) of overall intellectual ability (Early Learning Composite, and subscale T-scores ( M = 50, SD = 10) for receptive language (RL) and expressive language (EL), as well as nonverbal fine motor (FM) and visual reasoning (VR) abilities. The VABS is a standardized parent-reported interview of everyday adaptive functioning that measures social, communication, daily living and motor skills. In addition (and for both groups) a semistructured play-based assessment, the Autism Diagnostic Observation Schedule—Generic (ADOS-G; Lord et al., 2000 ) was used to assess autism-related social and communication behavioral characteristics (44 children were administered Module 2 and the other three children Module 1 of the ADOS-G). This was augmented (At-risk group only) with the parent-report Autism Diagnostic Interview—Revised (ADI-R; Lord et al., 1994 ). In common with other research groups studying familial at-risk siblings ( Zwaigenbaum et al., 2007 ) a “best estimate clinical consensus” approach to diagnosis was taken following review by experienced clinical researchers (TC, KH, SC, GP), taking account of all information about the child (i.e., MSEL, VABS, informal observation) in addition to information from the ADI-R and ADOS-G. Children were included in the At-risk ASD group if they met ICD-10 ( World Health Organization, 1993 ) criteria for ASD. Given the young age of the children, and in line with the proposed changes to the Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM-5 ; http://www.dsm5.org ), no attempt was made to assign specific subcategories of pervasive developmental disorder/ASD diagnosis. Children from the At-risk group were considered typically developing (At-risk Typical) if they (a) did not meet ICD-10 criteria for an ASD, (b) did not score above the ASD cutoff on the ADOS or ADI, (c) scored within 1.5 SD of the population mean on the Mullen Early Learning Composite (ELC) score (>77.5) and Receptive Language (RL) and Expressive Language (EL) subscale T scores (>35). Children from the At-risk group were considered to have atypical development if they did not fall into either of the above groups. That is, they either scored above the ADOS or ADI cutoff for ASD or scored < 1.5SD on the Mullen ELC or RL and EL but did not meet ICD-10 criteria for an ASD. From the 47 At-risk participants taking part in this task, 17 met criteria for an ASD diagnosis, 18 were At-risk Typical, and 12 were in the At-risk Atypical group (nine scoring above ADOS ASD cutoff, one scoring above ADOS ASD cutoff and <1 SD Mullen ELC cutoff, one scoring above ADI ASD cutoff, and one scoring <1.5 SD Mullen ELC cutoff).

Results

We analyzed separately the true belief (TB) and the false belief (FB) trials. As in previous studies ( Southgate et al., 2007 ), only the second TB trial was analyzed. By not including the first TB trial, we thus gave children more opportunities to understand the actor’s goal—to reach for the objects—as well as the role of the audiovisual cue. For each trial we assess performance by analyzing the difference between the looking time (LT) to the correct and incorrect doors’ AOIs, scaled by the amount of looking to those AOIs: (LT Correct − LT Incorrect )/(LT Correct + LT Incorrect ). Values go from −1 (exclusive looking toward the Incorrect location) to 1 (exclusive looking to the Correct location), with chance level at zero. In the TB trial, the correct location was that which contained the banana. In the FB trial, the correct location was that in which the actor thought the banana was. We start the analysis by comparing the Control and At-risk groups to chance levels and to each other and then compare the three outcome groups within the At-risk participants (At-risk ASD, At-risk Atypical, and At-risk TD) to chance levels and to each other (using Bonferroni correction for multiple comparisons). Where a significant difference between Controls and At-risk is found, we also test whether all at-risk groups are significantly different than Controls (using Dunnett correction for multiple comparisons). We also test whether any individual variables that showed groups differences, like total IQ, verbal IQ, or age (see Table 1 ) explain group differences in TB or FB performance. Finally, we examine whether visual attention distribution during the false belief trial may account for children’s performance. Three AOIs were entered in this analysis, the door AOIs and another AOI corresponding to the actor ( Figure 1 ).

Looking time differential scores were significantly above chance (zero) for both Control and At-risk participants: Control t (37) = 3.01, p = .005; At-risk t (40) = 2.99, p = .005. A univariate analysis of variance (ANOVA) with Group (At-risk vs. Control) as between-subjects factor yielded a nonsignificant effect of Group, F (1, 78) = 0.13, p = .716, η 2 = .002. Verbal and General IQ significantly predicted performance—Verbal IQ F (1, 76) = 1.78, p = .03, η 2 = .06; General IQ F (1, 76) = 1.78, p = .02, η 2 = .06—however, entering these factors in the above ANOVA did not change the significance level for the main Group factor. When analyzing the behavior of the three at-risk groups separately, they only performed marginally better than chance—At-risk Typical t (15) = 1.24, p = .23; At-risk Atypical t (10) = 2.01, p = .07; At-risk ASD t (13) = 2.03, p = .06)—possibly also because of the reduced power of this analysis. A univariate ANOVA comparing the three at-risk subgroups (At-risk ASD, At-risk Atypical and At-risk Typical) yielded no significant effect of Group, F (2, 40) = .19, p = .82, η 2 = .01.

False Belief

At the point at which anticipatory looking is measured in the False Belief trial the two boxes were empty, thus preventing a reality bias. Correct anticipation is reflected in longer looking toward the box that contained the banana just before the person looked away. As seen in Figure 2 , looking time differential scores were higher for the Control group than for the High-risk groups. Preliminary analyses confirmed that Total IQ, Verbal IQ, or Age did not have a main effect on looking time distribution nor did they interact with the factor Group. We therefore removed these factors from further analyses. There was also no group difference in the overall amount of time spent looking at the three target AOIs (correct, incorrect, and face) after the light prompt, in the FB trial ( M Control = 3.35, SD Control = 1.4 s; M At-risk = 3.56, SD At-risk = 1.15 s), t (82) = −0.71, p = .47. Mean looking time difference scores were significantly above chance only for the Control participants: Control t (35) = 5.13, p < .001; At-risk t (42) = 0.86, p = .39. A univariate ANOVA with Group (Control and At-risk) as between-subjects variable resulted in a significant main effect of Group, F (1, 78) = 9.35, p = .003, η 2 = .10. The significance level of the Group factor did not change when the TB looking time performance was entered as a covariate and TB performance did not have a significant impact on FB performance (see Table 2 ). When the three At-risk subgroups (At-risk ASD, At-risk Atypical, and At-risk Typical) performance was analyzed separately, none of the groups performed different than chance: At-risk Typical t (16) = 1.21, p = .23; At-risk Atypical t (9) = 0.36, p = .72; At-risk ASD t (15) = –0.36, p = .71. A univariate ANOVA comparing the looking time difference scores for the three At-risk subgroups yielded a nonsignificant effect of Group, F (2, 42) = 0.79, p = .46, η 2 = 0.03. Post hoc t tests, were used to compare each at-risk group to the Control participants. Only At-risk ASD significantly differed from Control participants ( p = .009), At-risk Atypical and At-risk Typical were not significantly different from Control ( p = .11 and p = .33).

Looking time differential scores in the false belief trial. The chance level is at zero. ASD = autism spectrum disorder. Error bars represent standard error.

Relationship With Social and Communication Abilities (ADOS)

The lack of a difference in performance between the three at-risk groups suggests that difficulties with mental state understanding may be unrelated to ASD symptom severity. To confirm that performance in this task is only related to the risk status and not to children’s social and communication abilities as measured by the ADOS, we split the Control and At-risk groups depending on their ADOS scores into a Low ADOS (ADOS < 8; 25 out of 35 Controls and 21 out of 42 At-risk participants) and High ADOS group (ADOS ≥ 8). Looking time performance was entered in a univariate ANOVA with Group (Control, At-risk) and ADOS (Low, High ADOS). This analysis yielded a main effect of risk Group, F (1, 76) = 9.41, p = .003, η 2 = 0.11. The ADOS scores did not significantly predict performance, F (1, 76) = .19, p = .66, η 2 = 0.003, and there was no significant interaction between risk Group and ADOS levels, F (1, 76) = 1.52, p = .22, η 2 = 0.02.

Differences in Attention to the Placement/Displacement Events

What can explain the poorer performance of the At-risk participants in the FB trial? We were interested in determining whether children’s looking behavior during the task differed in any way that would explain their performance. One possible source of error could arise from not paying attention to the hiding and displacement events during the FB interval, especially the last hiding event before the actress looks away. Visual inspection of looking time distribution along the FB trial suggests that all groups followed closely this event ( Figure 3a : “Banana placed in the right box”; “Monkey steals banana from right box”) and that major differences between groups only emerge at the very end, when FB is tested ( Figure 3a : “Person turns back”). We looked more specifically at attention distribution during key events. Groups spent equal amounts of time looking at the box during the 8 s that the monkey took to place the banana ( M Control = 4.9, SD Control = 2.0; M At-risk Typical = 5.3, SD At-risk Typical = 1.6; M At-risk Atypical = 4.9, SD At-risk Atypical = 1.7; M At-risk ASD = 5.3, SD At-risk ASD ), F (3, 84) = 0.51, p = .67, η 2 = .01. Groups also spent equal amounts of time looking at the box from which the monkey surreptitiously removed the banana ( M Control = 7.4, SD Control = 2.9; M At-risk Typical = 6.5, SD At-risk Typical = 3.0; M At-risk Atypical = 7.9, SD At-risk Atypical = 2.3; M At-risk ASD = 6.7, SD At-risk ASD = 2.7), F (3, 84) = 0.91, p = .44, η 2 = .03. It is also important to have noticed that, when the banana was removed from the box, the person was looking away. Visual inspection of looking time spent on the face during the FB trial does not highlight consistent group differences ( Figure 3b ), and, indeed, when we compared the amount of time spent looking at the person’s face while the monkey removed the banana from the box no group difference was found ( M Control = 3.8, SD Control = 1.9; M At-risk Typical = 3.1, SD At-risk Typical = 2.4; M At-risk Atypical = 4.0, SD At-risk Atypical = 2.7; M At-risk ASD = 4.5, SD At-risk ASD = 2.6), F (3, 84) = 1, p = .41, η 2 = .03. None of these measures correlate with the FB looking time difference score, for either the whole group or the low and at-risk groups separately.

a. Proportion of children looking at the face area of interest (AOI) during the false belief (FB) trial. b. Proportion of children looking at the correct AOI (the right box) during the FB trial to illustrate all groups keeping track of where the banana was placed. At-risk autism spectrum disorder (ASD) and At-risk Atypical data were pooled together for clarity (continuous gray line). At-risk Typical (dashed gray line) and Controls (black line). The onset of important events is indicated on the time line.

Closer exploration of the data revealed that at the point in the video where the monkey had placed the banana in the right box and left the screen and before the person turned away, children looked up at the person ( M Control = 2.0, SD Control = 1.1; M At-risk Typical = 2.2, SD At-risk Typical = 1.0; M At-risk Atypical = 2.7, SD At-risk Atypical = 1.3; M At-risk ASD = 2.1, SD At-risk ASD = 0,9), F (3, 80) = 1.42, p = .24, η 2 = .05, and then looked toward the right door and box. Encoding where the person last saw the object or her goal at this point where a TB is still held may be crucial for predicting their behavior later. We analyzed looking time distribution to correct (here where the banana had been placed) and incorrect locations at this time point. Both Low-risk and High-risk participants looked longer at the Correct side—average and SD for Correct versus Incorrect for Low-risk: 620 ms (105) versus 370 ms (81) and High-risk: 552 ms (90) versus 369 ms (69). A 2 × 2 ANOVA with Side and Group confirmed that there was a main effect of Side, F (1, 79) = 5.01, p = .02, η 2 = .06, but no main effect of Group, F (1, 79) = 0.21, p = .64, and no interaction between Side and Group, F (1, 79) = 0.11, p = .74, which means that both groups looked longer at the box containing the banana. To investigate whether looking time distribution at this point was related to performance later in the FB trial we calculated difference scores in both cases (Looking time Correct − Looking time Incorrect). These measures were correlated in the whole sample, r (70) = .27, p = .01, as well as in the Low-risk group, r (31) = .40, p = .02, but not in the High-risk group, r (39) = .13, p = .41. A Chow test demonstrated that the slope and intercept of the regression analysis predicting test performance from looking time distribution when the person last saw the object was not significantly different for the high-risk and low-risk participants, F (1, 69) = 2.02, p = .15.