A Tour of the Structural Equation Models Platform in JMP Pro 18

[#1]

The release of JMP® Pro 18 is finally here! The Structural Equation Models (SEM) platform has a variety of new features that are sure to please those who have non-normal data (robust standard error corrections and platform-wide bootstrap!) and who want to speed up their analyses with new model shortcuts (who said measurement invariance?). 

Since introducing the SEM platform in JMP 15, we've amassed numerous features that enable users to do all sorts of sophisticated work. This post's goal is to provide a complete tour of all that JMP Pro offers to SEM researchers. Watch the short video below for the visual tour or keep on reading to learn if your needs are fully or partially met with the most up-to-date SEM functionality.

SEM features

• The platform can be launched with raw data or pre-summarized data (correlation or covariance matrices). Frequency variables are also supported for analyses, as are grouping variables for multiple-group analysis.
• The mean structure is always included in models. This enables users to conduct all types of SEM analyses including confirmatory factor analysis, path models (with or without latent variables), latent growth curves, multiple-group analysis of latent variable mean differences, and more.
• Maximum likelihood is the estimator used in the SEM platform. When missing data are detected, the platform automatically uses full information maximum likelihood.
• Robust standard error corrections and overall test corrections are available for non-normal data, as is bootstrap of any statistic available in the report (including Bollen-Stine bootstrap).
• Interactive model specification is most convenient for efficiency, but scripts can be written to execute or replicate analyses if desired. JMP Workflow Builder records all your steps, from data manipulation to visualization and analyses for full replicability and sharing.
• Dynamic user feedback is provided upon any modification of a model, enabling users to identify potential issues with model specification, estimation, or their data before the model is fitted.
• A menu with numerous model shortcuts is available for fast and easy fitting of common models.
• A model comparison table is presented containing fit statistics for all models fitted in the same report. The table is interactive enabling Chi-square difference tests on the spot. Upon requesting a model comparison, an algorithm for identifying nested models is executed (Bentler & Satorra, 2010) and then alerts users if the models aren’t nested.
• All standard output is available from each fitted model, including fit indices, unstandardized and standardized estimates, total effects, indirect effects, model-implied matrices, normalized residuals, observational residuals, predicted values, covariance of estimates, R-square for endogenous variables, modification indices, Thurstone and Bartlett factor scores, and more.
• Measurement models include a dashboard to assist in the assessment of reliability and evidence for validity of constructs. Statistics in this dashboard include indicator reliability, construct and composite reliability (coefficient Omega and H), and a construct validity matrix with average variance extracted, and overlap in variance of latent variables and indicators.
• JMP aims to have a visualization for every statistic. In line with this design principle, the SEM platform has heat maps, bar plots, box plots, spaghetti plots, and bivariate plots where appropriate.
• Cutting-edge algorithms rely on the statistical computing expertise of developers with decades of experience, making the fitting of models fast and efficient.
• High-quality vectorized images can be copied and pasted to use in presentations and publications, including a fully interactive and customizable path diagram.
• The path diagram encodes information in its nodes and edges; the thickness, style, and transparency of edges can be mapped to effect sizes and statistical tests, and R-square values are shaded areas inside nodes.
• Local data filters immediately turn SEM reports into interactive dashboards that display results for selected characteristics of the sample. Here, we can see that selecting either Female or Male updates the report seamlessly to display the results of the model for those specific observations.
• Exploring outliers and missing value patterns is easily done from the main menu, which launches additional JMP platforms with their own functionality for exploring the data.
• Other statistical techniques that are commonly embedded in SEM software are latent class analysis and latent profile analysis. These, and numerous other multivariate and univariate techniques, have their own platforms in JMP Pro and are available under the Analyze menu.