Descriptive data on the children’s performance on the measures used in the data analyses are provided in . Physiological data are provided for children in the TLA group as well, although this group was excluded from analyses of physiological measures.
Performance on Sustained Looking and Physiology Measures
Due to the amount of missing physiological data, we considered the possibility that results might be biased by patterns of missing data. We examined where children with missing physiological data fell within their respective groups on the variables of chronological age, MSEL Early Learning Composite (i.e., total standard score), PLS Expressive Communication age equivalent scores, and mean proportion of sustained looking during the CDS condition. These variables were selected due to their significant levels of correlations with one or more physiological variables for either the ASD or the TCA group. Specifically, for the TCA group, chronological age correlated significantly with IBI during sustained looking in both the nonsocial condition, rho(11) = .667, p = .03, and the CDS condition, rho(12) = .604, p = .04. For the ASD group, the MSEL Early Learning Composite correlated significantly with mean RSA during the CDS condition, rho(18) = .504, p = .03; the PLS Expressive Communication age equivalent scores correlated significantly with RSA during the nonsocial condition, rho(18) = .482, p = .04; and the proportion of sustained looking during the CDS condition correlated significantly with mean RSA during the CDS condition, rho(18) = .498, p = .04. Based on this visual analysis of outliers, children with missing data did not present any clear patterns that might bias the study outcomes. Participants with missing data were deleted on a case-by-case basis in each analysis.
Group Comparisons of Behavioral Responses to Nonsocial and to CDS Stimuli
We first used a MANOVA to compare the three groups on sustained looking at nonsocial and CDS stimuli. The Box’s test of equality of covariance matrices was non-significant, but Levene’s test of equality of error variance indicated a significant difference in error variance for sustained looking at CDS. The multivariate result was significant for group, Pillai’s Trace = .594, F = 9.50, df = (4, 90), p < .001, partial eta squared = .297. The univariate tests of main effects indicated significant effects for both sustained looking at nonsocial stimuli, F = 11.03, p < .001, partial eta squared = .329, and sustained looking at CDS stimuli, F = 8.31, p = .001, partial eta squared = .270. In post-hoc comparisons between groups for nonsocial stimuli, the ASD and TCA groups did not differ from one another, p = .256, but both looked a larger proportion of the time than the TLA group, p < .001, 95% CI .076, .432; and p < .001, 95% CI .151, .519, respectively. For CDS stimuli, the ASD and TLA groups did not differ from one another (p = .447), but both looked a smaller proportion of the time than the TCA group, p < .001, 95% CI −.496, −.141; and p = .001, CI −.371, −.091, respectively. The comparisons of groups are displayed graphically in . Our results for this planned analysis partially supported our first hypothesis in that children with ASD performed like the TCA group in sustaining attention to nonsocial stimuli, but showed less sustained attention to CDS. They did not, however, look less overall to CDS stimuli than the TLA group.
Proportion of Sustained Attention by Group during Nonsocial and Child-Directed Speech Stimuli
Visual inspection of the data suggested a more complex pattern of group differences in responses to the three CDS vignettes, however. Thus, as an exploratory analysis, we ran second MANOVA to examine the differences among the three groups in proportion of looking at CDS when the three different CDS vignettes (i.e., video story, live puppet show, and video nonsense toy) were separated rather than aggregated. The multivariate effect for group was significant, Pillai’s Trace = .576, F = 6.06, df = (6, 90), p < .001, partial eta squared = .288. Examining the univariate main effects, we found that the main effect for group was significant for each CDS vignette: Video Story, F = 8.68, p = .001, partial eta squared = .27; Puppet Show, F = 8.4, p = .001, partial eta squared = .269; Video Toy, F = 5.94, p = .005, partial eta squared = .205. Due to a significant Levene’s test for each of the three CDS vignette, the Games-Howell test was used in the post-hoc group comparisons. The ASD and TLA groups did not differ in proportion of looking for either Video Story or Video Toy, but the ASD group looked significantly less than the TLA group during the live Puppet Show, p = .004, 95% CI −.406, −.072. The ASD group looked a smaller proportion of the time than the TCA group during Video Story, p = .001, 95% CI −.541, −.136. The TLA group also looked a smaller proportion of the time than the TCA group during Video Story (p < .001, 95% CI −.639, −.218), but did not differ significantly from the TCA group in proportion of looking during Puppet Show (p =.766) or Video Toy (p = .058). shows the proportions of looking for the three groups of children during the different CDS vignettes. As illustrated in this figure, variability in the post-hoc comparisons is accounted for by an increase in the proportion of sustained looking by the TLA group during the Puppet Show compared to the videotaped CDS vignettes, rather than by a decrease in the proportion of sustained looking during the Puppet Show for children with ASD.
Proportion of Sustained Attention by Group during Three Child-Directed Speech Vignettes
Although the use of proportions potentially can change the distributional patterns of the data, the proportional data in this study had the same general distributional properties as the raw data due to the reasonable consistency in length of conditions across children. Alternative analyses in which seconds of looking were prorated to a standard viewing time of 60 seconds for each condition yielded the same pattern of findings with comparable statistical values. Also, analyses using a censored linear regression model to examine influences of potential ceiling effects in sustained looking for children in the TCA group yielded results consistent with those presented above.
Group Comparisons of Physiological Responses to Nonsocial and CDS Stimuli
The TLA group was omitted from the analyses of physiological responses due to known age-related differences in both IBI and RSA not of interest in these analyses. For the analysis examining IBI, the Box’s test of equality of covariance matrices was significant, and the Levene’s test of equality of error variances indicated that the two groups differed in error variance for both the nonsocial and CDS stimuli. The omnibus test showed a significant multivariate effect for group, Pillai’s Trace = .263, F = 4.64, df = (2, 26), p = .019, ηp2= .263. The univariate models indicated significant main effects for group for both nonsocial stimuli (F = 4.34, p = .047, ηp2 = .138) and CDS stimuli (F = 8.56, p = .007, ηp2 = .241). For each type of stimuli, the ASD group had smaller IBIs (i.e., faster heart rates) than the TCA group (see for values).
For the repeated measures ANOVA examining RSA during nonsocial and CDS stimuli, the Box’s test for equality of covariance matrices and the Levene’s test for equality of error variances were nonsignificant, suggesting that the model assumptions were met. The analysis showed no significant effect for group (F
= 1.29, p
= .265, ηp2
= .046), stimulus type (Pillai’s Trace = .065, F
= 1.88, p
= .182, ηp2
= .065), or the group x stimulus type interaction (Pillai’s Trace = .045, F
= 1.32, p
= .261, ηp2
= .046). The direction of the nonsignificant group differences, as shown in , was for the ASD group to show lower RSA than the TCA group, and for the ASD group to show more decline in RSA from nonsocial to CDS stimuli than the TCA group. Following recommendations of Onwuegbuzie and Leech (2004)
, we examined the observed power associated with these nonsignificant results and found it to be at low levels of .195 for group, .262 for stimulus type, and .198 for the group x stimulus type interaction.