The ACTIVE conceptual framework as a structural equation model.

Background/Study Context: Conceptual frameworks are analytic models at a high level of abstraction. Their operationalization can inform randomized trial design and sample size considerations. METHODS: The Advanced Cognitive Training for Independent and Vital Elderly (ACTIVE) conceptual framework was empirically tested using structural equation modeling (N=2,802). ACTIVE was guided by a conceptual framework for cognitive training in which proximal cognitive abilities (memory, inductive reasoning, speed of processing) mediate treatment-related improvement in primary outcomes (everyday problem-solving, difficulty with activities of daily living, everyday speed, driving difficulty), which in turn lead to improved secondary outcomes (health-related quality of life, health service utilization, mobility). Measurement models for each proximal, primary, and secondary outcome were developed and tested using baseline data. Each construct was then combined in one model to evaluate fit (RMSEA, CFI, normalized residuals of each indicator). To expand the conceptual model and potentially inform future trials, evidence of modification of structural model parameters was evaluated by age, years of education, sex, race, and self-rated health status. RESULTS: Preconceived measurement models for memory, reasoning, speed of processing, everyday problem-solving, instrumental activities of daily living (IADL) difficulty, everyday speed, driving difficulty, and health-related quality of life each fit well to the data (all RMSEA < .05; all CFI > .95). Fit of the full model was excellent (RMSEA = .038; CFI = .924). In contrast with previous findings from ACTIVE regarding who benefits from training, interaction testing revealed associations between proximal abilities and primary outcomes are stronger on average by nonwhite race, worse health, older age, and less education (p < .005). CONCLUSIONS: Empirical data confirm the hypothesized ACTIVE conceptual model. Findings suggest that the types of people who show intervention effects on cognitive performance potentially may be different from those with the greatest chance of transfer to real-world activities.

IV. DEVELOPMENTS IN THE ANALYSIS OF LONGITUDINAL DATA.

Longitudinal data analytic techniques include a complex array of statistical techniques from repeated-measures analysis of variance, mixed-effects models, and time-series analysis, to longitudinal latent variable models (e.g., growth models, dynamic factor models) and mixture models (longitudinal latent profile analysis, growth mixture models). In this article, we focus our attention on the rationales of longitudinal research laid out by Baltes and Nesselroade (1979) and discuss the advancements in the analysis of longitudinal data since their landmark paper. We highlight the developments in growth and change analysis and its derivatives because these models best capture the rationales for conducting longitudinal research. We conclude with additional rationales of longitudinal research brought about by the development of new analytic techniques.

Estimation of Latent Variable Scores with Multiple Group Item Response Models: Implications for Integrative Data Analysis.

Integrative data analysis (IDA) involves obtaining multiple datasets, scaling the data to a common metric, and jointly analyzing the data. The first step in IDA is to scale the multisample item-level data to a common metric, which is often done with multiple group item response models (MGM). With invariance constraints tested and imposed, the estimated latent variable scores from the MGM serve as an observed variable in subsequent analyses. This approach was used with empirical multiple group data and different latent variable estimates were obtained for individuals with the same response pattern from different studies. A Monte Carlo simulation study was then conducted to compare the accuracy of latent variable estimates from the MGM, a single-group item response model, and an MGM where group differences are ignored. Results suggest that these alternative approaches led to consistent and equally accurate latent variable estimates. Implications for IDA are discussed.