Statistical power for longitudinal developmental trajectories: The (non-)impact of age matching within measurement occasions.
Statistical power for longitudinal developmental trajectories: The (non-)impact of age matching within measurement occasions.
复制标题
纵向发展轨迹的统计功效：测量场合中年龄匹配的（非）影响。
DOI：
10.1037/dev0001459
复制
发表时间：
2023
影响因子：
4
通讯作者：
Kelleher,BridgetteL
中科院分区：
文献类型：
--
作者：
Lane,SeanP;Kelleher,BridgetteL
招募参与者进行早期纵向发展研究具有挑战性，通常会导致样本量的实际上限和由于自然减员而缺失的数据。这些因素对此类研究的统计功效构成风险，具体取决于预期的分析模型。一种缓解策略是通过对尽可能接近固定实际年龄的儿童进行评估来提高测量精度。我们提出的分析，说明这种做法是如何只有有时有用的，专注于使用多层次建模方法分析时间轨迹的情况下。使用两项纵向发展研究的结果进行了模拟。根据连续和离散发育过程生成数据，并在间隔、顺序或分类量表上对治疗时间进行因子分析。检测持续产生的发育过程的能力对目标年龄周围的变异性增加是稳健的，甚至从中受益。对于离散过程，以普通/分类方式建模时，功效不受影响，但如果使用间隔尺度上的精确实足年龄建模，则功效稳步下降。我们的研究结果表明，在许多情况下，研究人员可能不必要地投入资源，以尽量减少年龄抽样的变化时，研究功能模式的时间。事实上，当理论的发展过程是连续的，增加年龄抽样的评估变异性和利用多层次模型，有利于潜在的增长曲线的替代品，可以与显着的收益，而不是降低功率。这种考虑还延伸到其他常见发展模型的有限等效公式，如面板分析。（PsycInfo数据库记录（c）2023阿帕，保留所有权利）
Recruiting participants for studies of early-life longitudinal development is challenging, often resulting in practical upper bounds in sample size and missing data due to attrition. These factors pose risks for the statistical power of such studies depending on the intended analytic model. One mitigation strategy is to increase measurement precision by conducting assessments of children as close to a fixed chronological age as possible. We present analyses that illustrate how such practices are only sometimes useful, focusing on cases where temporal trajectories are analyzed using multilevel modeling approaches. Simulations were conducted using results from two studies of longitudinal development. Data were generated according to both continuous and discrete developmental processes and factorially analyzed treating time on either interval, ordinal, or categorical scales. The power to detect continuously generated developmental processes was robust to, and even benefited from, increased variability around target ages. For discrete processes, power was unaffected when modeled ordinally/categorically, but declined steadily if modeled using exact chronological age on an interval scale. Our results suggest that in many circumstances, researchers may be unnecessarily devoting resources toward minimizing age sampling variability when studying functional patterns across time. In fact, when the theoretical developmental process is continuous, increasing the age sampling variability of assessments and utilizing multilevel models in favor of latent growth curve alternatives can be associated with substantial gains rather than reductions in power. Such considerations also extend to limited equivalent formulations of other common developmental models, such as panel analysis.(PsycInfo Database Record (c) 2023 APA, all rights reserved)