流行病學研究所;Graduate Institute of EpidemiologySHAU, WEN-YIWEN-YISHAUHU, FU-CHANGFU-CHANGHU2008-09-162018-06-292008-09-162018-06-292006http://ntur.lib.ntu.edu.tw//handle/246246/81832This article generalizes linear path analysis (PA) and simultaneous equations models (SiEM) to deal with mixed responses of different types in a recursive or triangular system. An efficient instrumental variable (IV) method for estimating the structural coefficients of a 2-equation partially recursive generalized path analysis (GPA) model and equivalently generalized simultaneous equations model ( GSiEM) is developed. The 1st equation is of a generalized linear model form for any responses from the exponential family of distributions, and the 2nd one is a linear equation for continuous responses. The consistency and asymptotic normality of the proposed IV estimator is proved, and its finite sample properties are examined in simulations. Finally, how the IV estimator can be applied to more complicated GSiEM/GPA models is discussed.en-USMAXIMUM-LIKELIHOOD METHODSLEAST-SQUARESLINEAR-MODELSVARIABLESjournal article