Over the weekend, I came across an 1980 article by Joseph Berkson in The Annals of Statistics on "Minimum Chi-square, not Maximum Likelihood".
This paper was a discussion paper. Brad Efron began his discussion with
"Before tearing into the paper, let me first applaud Professor Berkson's skeptical attitude toward asymptotics and fancy theory in general. Throughout his productive career he has always been primarily concerned with the practical, the computable and the verifiable---the right attitude for a good scientist doing good science. His mistake is not crediting Fisher (and Rao, Savage, Ghosh, Subramanyam, and me) with some of the same good sense. "
Efron mentioned three components of good science in his comments: the applicability, the computability, and the "verifibility". This reminded me of the research of Andrew's student Jouni on Fully Bayesian Computing. In their research, they were building around general models (applicability), facilitizing computing (computability) and promoting model checking (verifibility).