Data harmonization and synthesis for mediation and moderation analysis

点击翻译按钮获取中文摘要

My research focuses on the development and evaluation of Bayesian methods for the analysis of mediation models and structural equations models (SEMs). This work falls under NSERC topic MS28, and is applied in the social, behavioral, and medical sciences. Scientific studies often examine intermediate variables that transmit the effect from one variable to another (mediators), and variables (moderators) that provide additional information about conditions in which effects occur. Advancing the field and using scientific findings to inform policy first requires a synthesis of findings from all relevant previous studies. There are three main obstacles to obtaining accurate estimates of effects in data synthesis: 1) the same parameter is not directly comparable between studies if there are differences in measurement instruments, target populations, covariates used in the analysis, and/or study procedures, 2) most data synthesis methods were developed for simple models that yield standardized mean differences for two groups, correlations, or regression coefficients, and there is a lack of empirically tested methods for synthesizing findings from more complex models (e.g., mediation models), and 3) the developments of methods for data synthesis have historically proceeded independently of developments of methods for data harmonization (equating scores on the same construct obtained using different measurement instruments and/or different response categories) and vice-versa. As of now there is no unified framework for data harmonization and data synthesis for mediation and moderation analyses. Therefore, the long-term objective of the proposed research program is to integrate data harmonization and data synthesis for SEMs in a single comprehensive statistical framework. The short-term objectives of the proposed research program are to: 1) Develop and test data harmonization methods that can accommodate various sources of between-study differences in measurement instruments 2) Create an integrated statistical framework for data harmonization and synthesis in mediation and moderation analyses Without methods to effectively synthesize findings from studies that examine mediators and moderators, existing research cannot be used for policy and decision making. This proposal is potentially ground-breaking because it describes the first integrated statistical framework for data harmonization and data synthesis for mediator and moderator effects. Mediators and moderators are examined in thousands of social science studies every year and the proposed methods have the potential to improve policy and decision-making based on findings in fields ranging from psychology to epidemiology and education research.