It was great to see Rod at the Society for Multivariate Experimental Psychology meeting this year. He was so full of life and energy and friendship. The passion he held for the field, for getting it right without unnecessary embellishment, for steering a course for future psychological and methodological research, and for doing so in a direct and civilized fashion was exemplary. Mindful of my last post to this area of the forum, I don't want this area to be one where I gripe into my beer about the passing of great contributors to our lives and technology. In Rod's case, there is a direct bearing on OpenMx. His early program COSAN, which I used as a graduate student, permitted specification of structural equation models through a formula, C = F1P1F1' + ... + FmPmFm' , where Fk = Fk1 ... Fkn(k), and Pk = Pk' , k = 1, ... ,m, and Fk might be a matrix, the inverse of a matrix, or inverse(I-Fk). This flexible matrix specification of models was almost enough for me never to invent Mx. However, it lacked multiple groups which are useful for behavior genetic models, and partitioned matrices and Kronecker products are helpful for automatically generating constraints such that the parameters for relative 1 are equal to those of relative 2 without having to be explicit about every single equated parameter. These features are at the heart of OpenMx today, and they owe a lot to many people, especially Rod. So with sadness, respect and gratitude, I raise my glass to him tonight.
Copyright © 2007-2017 The OpenMx Project