Missing data is a pervasive problem in longitudinal treatment research studies. Missing data due to study non-completion complicate the task of drawing conclusions about the effect of a treatment or policy on a measure of interest (e.g., a process measure or outcome). Biased estimates of change over time in a measure could result if attrition is related to the constructs that are being measured. Identifying potential biases in estimates is critical for research involving longitudinal assessments. The pattern-mixture model (PMM) provides a way to understand and account for attrition when analyzing data and communicating results to research stakeholders. This paper demonstrates the use of PMMs in a study of the quality of care in therapeutic communities (TCs) using the Dimensions of Change Instrument (DCI) to measure longitudinal client-level change and TC treatment process. The effect of choice of missing data pattern and its effects on conclusions drawn from analyses is highlighted along with the role of clinical expertise in formulating PMMs.
Paddock, Susan M., Maria Orlando Edelen, Suzanne L. Wenzel, Patricia A. Ebener, and Wallace Mandell, Pattern-Mixture Models for Addressing Nonignorable Nonresponse in Longitudinal Substance Abuse Treatment Studies. Santa Monica, CA: RAND Corporation, 2007. https://www.rand.org/pubs/working_papers/WR441.html.
Paddock, Susan M., Maria Orlando Edelen, Suzanne L. Wenzel, Patricia A. Ebener, and Wallace Mandell, Pattern-Mixture Models for Addressing Nonignorable Nonresponse in Longitudinal Substance Abuse Treatment Studies, Santa Monica, Calif.: RAND Corporation, WR-441-HLTH, 2007. As of January 12, 2022: https://www.rand.org/pubs/working_papers/WR441.html