This project develops methods for modeling incomplete data that arise in social and behavioral sciences (SBS). The main focus is analysis of data with nonignorable nonresponse, using structural equation models and methods that take into account the missing data mechanism. The investigators study and develop (1) selection and pattern mixture model approaches to jointly model the missing data mechanism and the variation in the observed data, (2) methods to segment the data into groups having the same missing data mechanism via development of statistical tests of homogeneity of mean and covariances, utilization of clustering methods as well as latent variable regression models, (3) multiple imputation methods that use predictive models for nonignorable nonresponse data to impute missing data, and (4) application of various types of bootstrap methods that take into account missing- ness. The investigators develop theoretically sound statistical methods, theories are assessed by extensive simulation studies, and methods are examined by application to real data, specifically the data from the University of Notre-Dame Adolescent Parenting Project, an on-going longitudinal study of teen parenting.
The investigators develop statistical methodology for analysis of data that are not complete. In social and behavioral sciences, data are often collected in longitudinal studies and through questionnaires. Lack of compliance of subjects (e.g., dropping out of studies and/or incomplete responses) that leads to incomplete data is commonplace. This project focuses on analysis of data that are missing not at random (MNAR). MNAR occurs when a case of a variable is not observed due to the value of that variable being atypical; for example, a subject does not submit to a measure of the level of her depression because she is unusually depressed. To-date, adequate statistical methodology to analyze MNAR data has not been explored in SBS. The investigators formulate new models, develop inferential and computational methods for MNAR data, and illustrate the methods with social and behavioral science data sets. In the latter respect, the investigators concentrate in applying the methodology to analyze a set of data collected by University of Notre Dame which studies teen parenting. The analyses are carried out in the context of structural equation modeling which has been widely used in a variety of disciplines including education, medicine, psychology, sociology, and other areas related to human behavior.
