This paper, authored by James J. Heckman, explores the formulation and estimation of simultaneous equation models that include both discrete and continuous endogenous variables. The paper aims to expand the multivariate probit structure to accommodate continuous endogenous variables and to provide a more general framework for analyzing econometric structural equation models. The author discusses the importance of discrete data in microeconomics and the need for statistical models that can handle such data. The paper introduces the concept of dummy endogenous variables, which can serve as proxies for unobserved latent variables or as direct shifters of behavioral equations. It presents five models incorporating these dummy variables, with a focus on the most novel and general hybrid model that includes both continuous and discrete endogenous variables. The paper derives conditions for the existence of meaningful statistical models, presents consistent estimators, and discusses maximum likelihood estimators. It also addresses the identification of parameters and the resolution of overidentification problems. The paper concludes with a comparison between the models developed and those proposed by Goodman and Nerlove and Press, highlighting the advantages of the hybrid model in handling both discrete and continuous random variables.This paper, authored by James J. Heckman, explores the formulation and estimation of simultaneous equation models that include both discrete and continuous endogenous variables. The paper aims to expand the multivariate probit structure to accommodate continuous endogenous variables and to provide a more general framework for analyzing econometric structural equation models. The author discusses the importance of discrete data in microeconomics and the need for statistical models that can handle such data. The paper introduces the concept of dummy endogenous variables, which can serve as proxies for unobserved latent variables or as direct shifters of behavioral equations. It presents five models incorporating these dummy variables, with a focus on the most novel and general hybrid model that includes both continuous and discrete endogenous variables. The paper derives conditions for the existence of meaningful statistical models, presents consistent estimators, and discusses maximum likelihood estimators. It also addresses the identification of parameters and the resolution of overidentification problems. The paper concludes with a comparison between the models developed and those proposed by Goodman and Nerlove and Press, highlighting the advantages of the hybrid model in handling both discrete and continuous random variables.