135 Case study challenging behavior 6 2.4.2 Describing change trajectory The subsequent steps were quantitative analyses – all performed in RStudio-2022.02.2–458 (RStudio Team, 2022) which runs on R software version 4.2.0 (R Core Team, 2020). To evaluate concurrent validity of self-ratings, we performed χ2 tests between self-ratings and informant-reported (daily records) accounts of days with self-injury and physical aggression. Kazdin (2019) recommends evaluating single-case timelines by combining visual inspections of graphed timeseries with statistical analyses. We therefore visualized the two self-report timeseries (physical aggression and self-injury) using functionality from ggplot2 (Wickham, 2016). Next, we pinpointed transitions in the physical aggression and selfinjury timeseries on the 560-day timeline. This transition-point detection was done with the ts_levels function from package casnet (Hasselman, 2023), which uses recursive partitioning (Therneau & Atkinson, 2022) to classify segments (or phases) on a timeseries with a relatively stable mean. We did this for the physical aggression and self-injury variables. Because these two variables are binary (0 = behavior did not occur on that day; 1 = behavior occurred on that day), mean levels effectively reflected the proportion of days with incidents within a phase. In the ts_levels function the minimum duration of one phase was set to seven days, comprising a whole weekly routine, and controlling for day-of-the-week effects. The absolute change criterion was set to 25%, meaning that each identified transition reflected at least a 25% increase or decrease compared to the mean of the preceding phase (cf. Lutz et al., 2012; Tang & DeRubeis, 1999). Based on suggestions by Kazdin (2019), we searched for transitions by visually inspecting a graph of the raw binary timeseries and a plot of the levels identified using the ts_levels function (Hasselman, 2023). After pinpointing transitions, we characterized the different attractor states in terms of what makes them (dis)similar from one another on the 560day timeline. We calculated – per phase and across the whole 560-day timeline – the mean frequency of self-rated challenging behaviors (i.e., mean days with challenging behaviors) and the mean frequencies of (sub)themes that the participant’s clinician hypothesized to be explanatory. Furthermore, we examined – per phase and across the whole 560-day timeline – whether these clinically relevant (sub)themes were associated with challenging behaviors. That is, Fisher’s exact tests evaluated whether a reported challenging behavior occurred (beyond chance) on the same days as reports of staff-hypothesized risk- or protective factors. Additionally, we performed Fisher’s exact tests to evaluate the relation between staff-hypothesized risk- or protective factors
RkJQdWJsaXNoZXIy MTk4NDMw