Relationship between imitation ability and social communication in autistic childrenCorrelation between imitation and social communication levels

We combined the imitation and social communication data from both participant groups and conducted correlation analyses to explore their association in autistic children. Pearson correlation analysis revealed significant positive correlations between imitation and social communication scores at both pre- and post-intervention time points (Pre-intervention: r = 0.91, p < 0.01; Post-intervention: r = 0.75, p < 0.05). Further analysis revealed a significant positive correlation between the change scores in imitation ability and those in social communication (r = 0.65, p < 0.01), calculated as the differences between pre- and post-intervention scores. Indicating a close relationship between these skills in autistic children.

We also performed partial correlation analyses on the four subcomponents of imitation and social communication, controlling for aggregate scores of imitation and social communication scores, as well as age, to examine the independent relationships between their subcomponents.

Table 2 Partial correlation analysis of imitation and social communication subdomains (Controlling for Age)

Partial correlation analyses from Table 2 revealed several significant findings: There were strong positive partial correlations between vocal imitation and expressive communication, and between gesture imitation and social skills, indicating a close relationship between vocal imitation with expressive communication, and gesture imitation with social skills in autistic children. Additionally, the relationship between object imitation and joint attention approaches significance. Notably, some negative correlations, such as between gesture imitation and expressive communication, and between vocal imitation and social skills, were also observed. These negative correlations, resulting from controlling overall scores, without this control, all subcomponents of imitation and social communication would show significant positive correlations. Thus, these negative values indicated a relative reduction in association strength rather than a true negative relationship.

Impact of imitation factors on social communication improvement

To investigate the impact of imitation on social communication improvement—a primary goal in autism interventions—we conducted a regression analysis. First, we combined data from both the experimental group and the control group into a unified dataset. The improvement in social communication, defined as the difference between post-test and pre-test scores, was used as the dependent variable. Independent variables included all potential factors that might influence the improvement in social communication: age at pre-test, imitation at pre-test, change in imitation, and social communication at pre-test.

The stepwise regression model identified baseline imitation and change in imitation as significant predictors of social communication improvement (p < 0.01). The standardized coefficients were 0.81 for changes in imitation and 0.31 for baseline imitation, with an adjusted R² of 0.48 and VIF = 1.32, indicating no multicollinearity.

Further subgroup regression analyses within the ESDM and conventional intervention groups, using the significant predictors identified in the earlier stepwise regression (baseline imitation and change in imitation), confirmed these findings (see Table 3). In the ESDM group, both change in imitation and baseline imitation were significant predictors (standardized coefficients of 0.88 and 0.26, respectively, with an adjusted R² of 0.67). In the conventional intervention group, change in imitation was a significant predictor, while baseline imitation was marginally significant (standardized coefficients of 0.84 and 0.36, respectively, with an adjusted R² of 0.46).

Table 3 Regression analysis of the impact of imitation factors (including improvement degree and baseline level) on the enhancement of social communication skills

In summary, the regression analyses indicated that imitation abilities, both baseline levels and improvements, played a significant and positive role in enhancing social communication skills in autistic children. Improvements in imitation abilities had a stronger predictive effect, while baseline imitation levels also contributed to future social communication improvements, albeit to a lesser extent.

The predictive role of imitation on social communication is moderated by age — a moderation effect model

The regression analyses indicated that imitation factors (baseline levels and improvement) have a strong positive predictive effect on the improvement of social communication skills in autistic children across the ESDM group, the conventional intervention group, and the combined dataset. However, the coefficients of the independent variables and the R² values varied among these groups. Given the significant age differences between the two groups, age may moderate the impact of imitation abilities on social communication skills.

To test the moderating effect of age, we conducted moderation analyses based on the regression analysis results. Specifically, change in imitation was used as the independent variable, age as the moderator, change in social communication as the dependent variable, and baseline imitation as a control variable. Moderation analyses were performed separately for the combined dataset, the experimental group, and the control group. Additionally, to address potential ceiling effects (where baseline imitation scores are at or near maximum, leading to changes in imitation approaching or equal to zero, which could severely affect the moderating effect), we applied a ceiling treatment to baseline imitation data (pre-test), removing the top 10% of scores.

The moderation analysis for the combined dataset revealed that baseline imitation, change in imitation, and the interaction between change in imitation and age were all significant, while the impact of age alone was not significant (p > 0.05) (see Table 4). This suggested a significant moderating effect of age. Adding the interaction term between change in imitation and age (in months) significantly improved the fit of the regression model, with the adjusted R² increasing from 0.46 to 0.59 (ΔR² = 0.13), indicating that the interaction term contributed an additional 13% to the explanation of the dependent variable. The ΔF value was 14.12, reaching significance (p < 0.001), indicating a significant improvement in the regression model due to the interaction term.

Table 4 The moderating effect of age on the relationship between imitation and social communication for the combined dataset (n = 46)

A simple slopes analysis (Fig. 1; Table 5) revealed that the effect of imitation changes on social communication varies by age. The greatest impact was observed at an older age (mean age + 1 SD, ~ 50 months; regression coefficient = 7.33), with decreasing effects at mean age (40 months; coefficient = 4.74) and a younger age (mean age − 1 SD, ~ 30 months; coefficient = 2.15). This indicates that age positively moderates the relationship between imitation improvements and social communication gains in autistic children.

Fig. 1

Simple Slope Analysis of the Impact of Imitation Improvement on Social Communication at Different Levels of Age

Table 5 Simple slope analysis results for the relationship between imitation change and social communication change at different levels of age

To verify the stability of the moderating effect across different groups, we included group membership as a random effect variable in the model and conducted a mixed-effects analysis (Table 6). The independent variables included pre-test imitation scores and imitation change scores, with age (in months) as the moderating variable and group membership as a control variable.

The results of the mixed-effects model indicated that the moderating effect of age was significant and remained consistent across both the experimental and control groups. Specifically, the interaction term for “imitation change * age” was significant (β = 0.245, p < 0.001), suggesting that the positive impact of imitation change on social communication increased with age. The main effects of pre-test imitation scores (β = 1.879, p = 0.004) and imitation change scores (β = 4.895, p < 0.001) on changes in social communication were also significant. However, age as a main effect did not directly influence the change in social communication scores (p = 0.624). The simple slopes analysis is shown in Fig. 2.

Table 6 Mixed effects model results for moderating effect of age across groupsFig. 2

Simple slope analysis of the impact of imitation change on social communication change at different levels of age in two groups

In summary, the moderation effect model revealed that age significantly moderated the predictive relationship between imitation abilities and social communication skills in autistic children. As age increased, the impact of imitation improvements on enhancing social communication became more pronounced and was consistent across both the ESDM and conventional intervention groups.

Significant improvement in imitation abilities through the Early Start Denver Model interventionOverall intervention effects on imitation and social communication abilities

Due to age differences between the two groups, we conducted an ANCOVA with age as a covariate to enhance the evaluation of the intervention effect.

Both the ESDM and control groups showed significant improvements in imitation abilities and social communication scores from pre- to post-intervention (see Table 7). These results suggest that both interventions likely supported development in autistic children, although maturation effects cannot be ruled out.

Table 7 Descriptive statistics of imitation ability and social communication scores in Pre- and Post-test for the control and ESDM group

We conducted a two-way mixed ANOVA (Groups: ESDM vs. conventional intervention; Time: pre-test vs. post-test), with age as a covariate, to examine the effects on improving imitation abilities and social communication skills.

For imitation abilities, there was a significant main effect of time (F(1,50) = 22.09, p < 0.001) and a significant time by group interaction (F(1,50) = 4.54, p < 0.05), indicating differences in imitation ability changes between the groups. Simple effects tests revealed that while the control group had a higher mean imitation ability at baseline (7.16 vs. 5.46), the ESDM group surpassed the control group post-intervention (19.54 vs. 16.05), showing a significantly greater improvement (see Fig. 3A).

For social communication skills, there was a significant main effect of time (F(1,50) = 23.964, p < 0.001), indicating improvements in both groups. However, the time by group interaction was not significant (F(1,50) = 0.003, p > 0.05), demonstrating that the ESDM intervention did not facilitate significantly greater gains in social communication skills compared to the control group (see Fig. 3B).

Fig. 3

Pre- and post-test comparison of imitation (A) and social communication (B) scores between two groups

Intervention effects on subcomponents of imitation and social communication

Further analysis of the pre- and post-intervention data for each subcomponent revealed significant improvements in all imitation and social communication subcomponents for both the ESDM and control groups (p < 0.01).

Using the difference between post-test and pre-test scores as indicators of improvement, ANCOVA was conducted with age as a covariate. The ANCOVA results for imitation subcomponents improvement are presented in Table 8.

In the subcomponents of imitation abilities, significant differences were observed between the ESDM and control groups in the improvement of object imitation, gesture imitation, and oral-facial imitation (p < 0.05). This indicated that the ESDM group exhibited more significant improvements in all types of imitation abilities except for vocal imitation. The improvement in vocal imitation, while higher in the ESDM group, did not reach a significant level (p = 0.09). Figure 4 illustrates the differences in the improvement of imitation subcomponents between the two groups.

Table 8 Results of covariance analysis on the improvement of imitation and social communication subcomponents for the control and ESDM group (Controlling for Age)Fig. 4

Comparison of improvement in imitation and social communication subscores for the control and ESDM group

In the social communication subcomponents, the differences in improvement scores between the four subcomponents were minimal, with no significant differences between the experimental and control groups (see Table 8; Fig. 4).

Overall, these results suggest that ESDM intervention has led to significantly greater improvements in imitation abilities compared to conventional intervention, with notable differences observed in all imitation subcomponents except for vocal imitation. However, this study’s non-randomized design requires cautious interpretation of its results, as it may have introduced selection bias that complicates attributing improvements in imitation abilities solely to the ESDM intervention. The potential bias may also impact the generalizability of the findings. Although the average improvement in social communication was higher in the ESDM group, this difference did not reach statistical significance.