Supplementary Material for: Effects of Skip-Logic on the Validity of Dimensional Clinical Scores: A Simulation Study

Structured assessment of clinical phenotypes is a burdensome procedure, largely due to the time required. One method to alleviate this is “skip-logic,” which allows for portions of an interview to be skipped if initial (“screen”) items are not endorsed. The bias that skip-logic introduces to resultant continuous scores is unknown and can be explored using Item Response Theory. Interview response data were simulated while varying 5 characteristics of the measures: number of screen items, difficulty (clinical severity) of the screens, difficulty of non-screen items, shape of the trait distribution, and range of discrimination parameters. The number of simulations and examinees were held constant at 2,000 and 10,000, respectively. A criterion variable correlating 0.80 with the measured trait was also simulated, and the outcome of interest was the difference between the correlations of the criterion variable and the two estimated scores (with and without skip-logic). Effects of the simulation conditions on this outcome were explored using ANOVA. All main effects and interactions were significant. The largest 2-way interaction was between number of screen items and average item discrimination, such that the number of screen items had a large effect on bias only when discrimination parameters were low. This, among other interactions explored here, suggests that skip-logic can bias results using continuous scores; however, the effects of this bias are usually inconsequential. Skip-logic in clinical assessments can introduce bias in continuous sum scores, but this bias can usually be ignored.

临床表型的结构化评估是一项耗时费力的工作，其主要负担源自所需的时间成本。缓解这一负担的一种方法是跳跃逻辑（skip-logic），即若初始「筛查」条目未被应答，则可跳过访谈的对应部分内容。跳跃逻辑对最终生成的连续得分所引入的偏倚尚未明确，可借助项目反应理论（Item Response Theory）开展相关探索。本研究模拟了访谈应答数据，并对评估工具的5项特征进行变量操纵：筛查条目数量、筛查条目难度（对应临床严重程度）、非筛查条目难度、特质分布形态，以及区分度参数的取值范围。模拟次数与受测者人数分别固定为2000次与10000名。同时模拟了与所测特质相关系数为0.80的效标变量，本研究关注的结局为效标变量与两种估计得分（分别采用与未采用跳跃逻辑）之间的相关系数差值。采用方差分析（ANOVA）探究各模拟条件对该结局的影响。所有主效应与交互效应均具有统计学显著性。其中最显著的两因素交互效应出现在筛查条目数量与平均项目区分度之间，即仅当区分度参数较低时，筛查条目数量才会对偏倚产生显著影响。结合本次研究探索的其他交互效应结果来看，跳跃逻辑确实会对基于连续得分的分析结果引入偏倚；不过此类偏倚的实际影响通常可忽略不计。临床评估中使用的跳跃逻辑可能会对连续总分引入偏倚，但该偏倚通常无需考虑。