Fillmore K M, Johnstone B M, Leino E V, Ager C R
Department of Social and Behavioral Sciences, University of California, San Francisco 94143.
J Stud Alcohol. 1993 Jan;54(1):37-47. doi: 10.15288/jsa.1993.54.37.
In contextual (cross-level) analysis within multiple longitudinal general population studies, individual-level drinking behaviors (quantity per typical occasion, frequency of drinking per month and total volume of drinks per month) at final measurement are assessed by three models that simultaneously enter individual- and group-level measures. Two age groups (15-20 and 21-30) are independently assessed. In each model, the Time 1 individual-level drinking behavior and one of three group-level factors are entered. The group-level factors are (1) the percentage of abstainers at Time 1 for each age/sex cohort, (2) the Time 1 group mean for the drinking measure for the age/sex cohort and (3) the mean difference of the age/sex cohort's change in the drinking measure over time. All variables in the model are controlled by variations to exposure in per capita consumption of alcohol during the age/sex cohort's formative years and at Time 2. Meta-analysis assesses the homogeneity of the findings across studies. Models were proposed with the rationale that (1) understanding of individual drinking behavior can be advanced if individual-level data and group-level data are considered in the same models, and (2) integration of these two levels of analyses are, to date, rare. The rationale for using meta-analysis is that findings from the models can be assessed across social contexts with respect to their generalizability. The mean difference model, controlling for individual drinking at Time 1, is the most influential of the group-level models for the younger age group: the degree to which the group changes its drinking pattern is positively related to individual-level drinking behavior at final measurement, over and above the individual's drinking behavior at Time 1, for individual-level frequency of drinking among males (homogeneous among drinkers only). Younger females show more significant relationships for the mean difference females show more significant relationships for the mean difference model. Findings are significant for all relationships examined for the mean difference of the drinking of the group and the individual drinking among the older males and females. Measures of individual-level drinking for all measures at Time 1, controlling for the group-level effects, are significantly related to individual-level drinking at final measurement. The results are homogeneous for quantity (drinkers only) and volume among the young. Findings indicate that characterizations of the drinking for both the individual and the group to which the individual belongs predict measures of drinking practices on the individual level over time.
在多项纵向普通人群研究的情境（跨层次）分析中，通过同时纳入个体层面和群体层面测量指标的三个模型，对最终测量时的个体层面饮酒行为（每次典型场合的饮酒量、每月饮酒频率和每月饮酒总量）进行评估。独立评估了两个年龄组（15 - 20岁和21 - 30岁）。在每个模型中，纳入了时间1的个体层面饮酒行为以及三个群体层面因素之一。群体层面因素包括：（1）每个年龄/性别队列在时间1的戒酒者百分比；（2）年龄/性别队列饮酒测量指标在时间1的群体均值；（3）年龄/性别队列饮酒测量指标随时间变化的均值差异。模型中的所有变量都通过年龄/性别队列成长阶段和时间2的人均酒精消费量的暴露差异进行控制。荟萃分析评估了各研究结果的同质性。提出这些模型的理由是：（1）如果在同一模型中考虑个体层面数据和群体层面数据，对个体饮酒行为的理解可以得到推进；（2）迄今为止，这两个分析层面的整合很少见。使用荟萃分析的理由是，可以根据模型结果在不同社会背景下的可推广性对其进行评估。在控制时间1个体饮酒情况的均值差异模型中，对于较年轻年龄组而言，是最具影响力的群体层面模型：在排除个体在时间1的饮酒行为后，群体饮酒模式变化的程度与最终测量时个体层面饮酒行为（仅针对男性饮酒频率，且仅在饮酒者中具有同质性）呈正相关。较年轻女性在均值差异模型中显示出更显著的关系。对于年龄较大的男性和女性，群体饮酒均值差异与个体饮酒之间所有检验关系的结果均具有显著性。在控制群体层面效应的情况下，时间1所有测量指标的个体层面饮酒量与最终测量时的个体层面饮酒量显著相关。对于年轻人，饮酒量（仅针对饮酒者）和饮酒总量的结果具有同质性。研究结果表明，个体及其所属群体的饮酒特征可以预测个体层面随时间变化的饮酒行为测量指标。
