2.4.3. Three-Tiered Method of Analysis

To face the challenge of conducting research in this small sample and heterogeneous population and the possible occurrence of a type II error, this study utilized a three-tiered method of analysis [8,29], to firstly examine the group results of communicative interaction patterns between dyads at the general molar level, and secondly to strengthen the validity by analyzing how the communicative patterns of each dyad were congruent with or varied from the group results at the intermediate level, and lastly a further case analysis and clinical relevance at the more detailed molecular level.

Firstly, at the molar level of the group patterns, functional relationships between independent (conditions using EGAT or not) and dependent variables (communicative interaction) were examined using quantitative analysis. Rate per minute, which is the frequency divided by duration, and the proportional distribution of each communicative interaction behavior were presented using descriptive statistics (mean, standard deviation). We examined the mean differences of communicative turns in two conditions (EGAT and NEGAT) between two groups (children/youths and communication partners) using a two-way ANOVA and conducted Bonferroni post hoc analysis when the results were statistically significant. To compare the differences in frequencies of moves and communicative functions in the two conditions, parametric paired t-tests of variance was used as the data did not offend normality testing (the Shapiro–Wilk test) [37]. The significance level was set to 0.05 with two-tailed testing, and marginal significance was defined as a *p* value between 0.05 and 0.1 [38]. Data analyses were conducted using the Statistical Package for the Social Sciences (SPSS) version 25.0 (IBM Corporation, Armonk, NY, USA) and confirmed by a qualified statistician in the research team.

Secondly, at the intermediate level, the number of participants whose results were either congruent with or differed from the group result was counted to validate the group pattern. It was also possible to evaluate the degree to which individual variations influenced the mean value of the group in two conditions. In addition, a further analysis of the interrelationship between moves, communicative functions and modes of communication was conducted to determine how the dyads used their communicative functions in their initiations or response moves, and how they used EGAT and other modes, given that they were expressing specific communicative functions in the two conditions.

Lastly, at the molecular level, two cases who were typical or atypical of the group results were analyzed and compared to present the similarities and differences of the characteristics of communicative interaction and participant characteristics. It offered a basis for judgment of the potential factors from the environments and participant characteristics that could influence communicative interaction in dyads.
