Multilevel Multiprocess Modeling with Gsem

Abstract

Multilevel multiprocess models are simultaneous equation systems that include multilevel hazard equations with correlated random effects. Demographers routinely use these models to adjust estimates for endogeneity and sample selection. In this article, I demonstrate how multilevel multiprocess models can be fit with the gsem command. I distinguish between two classes of multilevel multiprocess models: nonrecursive systems of hazard equations without observed endogenous variables and recursive systems that include a hazard equation with observed endogenous qualitative variables. I illustrate the estimation of both classes of models using sample datasets shipped with the statistical software aML. I pay special attention to identifying structural coefficients in nonrecursive simultaneous systems.

Interaction and online courses for satisfactory university learning during the COVID-19 pandemic

“…As independent variables concurrently affect the two dependent variables of overall satisfaction and recommendation, the analytical method of simultaneous equations or structural equations that estimate several related equations ( Kennedy, 2008 ) with ordinal logistic regression could be more appropriate than those with linear regression. In particular, when one or more equations in n equations are nonlinear relational expressions, such as ordinal logistic or probit regression caused by survey data (i.e., n -point Likert scale) in the multilevel estimation model, it could be reasonable for researchers to utilize the generalized structural equation model (GSEM) rather than the structural equation model (SEM) ( Bartus, 2017 ; Dil & Karasoy, 2020 ; Palmer & Sterne, 2015 ; StataCorp, 2021 ).…”

merlin—A unified modeling framework for data analysis and methods development in Stata

“…More details on the formulation of the likelihood can be found in Crowther (2017) or the Stata manual entry for gsem because the formulation and default integration options (adaptive Gauss-Hermite quadrature) are the same.…”

“…Given this starting point, the natural extension is to allow for more than one biomarker, recurrent events, competing risks, etc. I have also released stmixed (Crowther, Look, and Riley 2014) for two-level parametric survival models, in particular for the Royston-Parmar spline-based model, and stgenreg for general hazard-based regression (Crowther andLambert 2013, 2014), where users could specify their own functional form for the hazard function and stgenreg would use numerical quadrature to calculate the likelihood.…”

“…There is also the widely used community-contributed package by Roodman (2011), which implements joint modeling of multivariate processes, assuming joint normality. Bartus (2017) further demonstrated ‘s capabilities in modeling simultaneous equation systems, including multilevel hazard equations with correlated random effects, with a focus on piecewise exponential models. builds directly on these works but focuses on more flexible parametric distributions, extending linking between outcome models, and nonlinear modeling capabilities.…”

“…In this article, I will introduce the core features and syntax of merlin and illustrate a variety of models that can be fit with it. More details on the methodological framework can be found in Crowther (2017). The article is structured as follows: In section 2, I describe the top-level syntax of the merlin command and describe many of the options.…”