Statistical Methods Course Description
Overview: This course teaches skills necessary to conduct analysis of complex multilevel data-structures from an nLevel Structural Equation Modeling perspective. Conventional multilevel modeling and multilevel-structural equation modeling work well with ‘standard’ multilevel data. The n-Level structural equation modeling (NL-SEM) framework is intended for both conventional and non-standard data-structures. Non-standard data are indeed very common across multiple domains but rarely analyzed in a satisfactory manner.
Some aspects of such non-standard data can be handled within most MLM or ML-SEM packages. Currently available approaches, however, do not scale well with increasingly complex data-structures. All of these models, on the other hand, are readily specified within the general NL-SEM perspective. NL-SEM provides a broader and simpler conceptual framework and matching statistical software. Within the NL-SEM framework all types of dependencies can be described using a consistent jargon-free language. Using the NL-SEM framework, all manner of models for nested data structures are easily specified and estimated using xxM, a free software platform. This course is designed to introduce participants to the modeling mindset of the NL-SEM framework and gain ample experience with the xxM software package.
Instructor: Paras Mehta
Paras is an Associate Professor of Clinical Psychology and Industrial Organizational Psychology at the University of Houston. His research interests include multilevel structural equations modeling, growth curve modeling, and applications of ML-SEM in educational and organizational research. He is a creator of the xxM R package for N-level Structural Equation Modeling.
Software and Computer Support
Participants will receive an electronic copy of all course materials, including slides, datasets, fully annotated xxM scripts and output with step-by-step instructions for fitting each model. Participants should bring a laptop computer and are encouraged to work-through each example on their own. Extra time is allocated for participants to seek individual assistance with the material or to work on their own analysis projects.
This course may be thought of an introductory-advanced course on Multilevel Structural Equation Modeling. The course assumes basic knowledge of MLM and SEM. A graduate course or a five-day workshop in SEM and MLM is probably necessary. That said, all essential concepts of SEM and MLM will be reviewed and re-introduced from a unified perspective during the first two days.
Novice participants with limited experience in MLM or SEM can expect to learn key ideas behind the modeling framework and learn how to fit fairly complex models using xxM. Expert participants will acquire a deeper understanding of how SEM and MLM may be readily and simply extended to
Participants are encouraged to contact the instructor before the workshop with their own data and research questions.
Participants from a variety of fields, including psychology, education, human development, sociology, marketing, business, biology, medicine, political science, and communication, will benefit from the course.
View Course Materials Page (Password Protected)
Example Data Structures
- Longitudinal data with switching classification in which nesting of a lower- level unit within the higher-level unit changes over time. For example, students switch teachers and classrooms across grades.
- Multiple ratings over time (longitudinal data with cross-classification and multiple membership). For example, each student may be rated by multiple different teachers and each teacher may rate multiple different students on multiple occasions. Such data are common in business settings as well.
- Non-hierarchical teams. Each person may belong to multiple teams over time. Examples of such data are common in business settings. Other examples include patients treated by teams of doctors and nurses.
- Dependent variables at “multiple lower levels.” For example, in a health-setting context we may have outcome variables for patients (satisfaction), nurses (job-satisfaction) and doctors (stress) over time, and nesting of patients within nurses and doctors may not be hierarchical.
- Round-robin data where each person may rate multiple other individuals within a small group. Variants of such data are common in business & I/O psychology (360 evaluations, team-performance), social-psychology (person perceptions) and education (peer-evaluations).
- Partially nested data. Only a subset of subjects are nested within a higher level unit. Such data occur commonly in “intervention studies” (medical, educational etc.) where individual subjects are randomly assigned to intervention and control conditions. Intervention itself is delivered in small groups leading to clustering; however, control subjects are not nested.
- Data with multiple different types of dependencies and many levels. For example, large longitudinal state datasets with multiple cohorts of students nested within teachers, classrooms, schools and districts. Classification may change across grades.