In my previous post about modelling buildings for energy efficiency, I effectively made the claim that back in the old days of mainframe computers, people were much more concerned about whether modelling work was valid. I can give some evidence for that claim from my own personal experience.
In the first two years of my working life, at the end of the 1970s, I worked on noise and vibration reduction for nuclear submarines using various computer models. The models we used were taken from methods given in mechanical engineering textbooks, and we used some finite element analysis programs, STRUDL and later on PAFEC, and also a program devised by Imperial College called COUPLE which combined analysis with test data. Whenever we compared our computer models with test data, the comparison was not very impressive. Basically we plodded on doing this work, which was funded by various MoD research establishments to get some sort of onsite analysis capability into a shipyard, and it had no influence on the design of any submarine. The computer models were not regarded as being "ready for primetime". About ten years later I happened to meet somebody who was currently working in the section that did this noise and vibration reduction analysis work, and I discovered that all our computer models from the end of the 1970s had just been thrown away.
But that strikes me as the correct attitude to computer modelling work. If it isn't valid, you can still potentially fund it to some extent, but you shouldn't really make any use of it. There seemed to be a more sceptical attitude towards computer modelling work in the days of mainframe computers. When desktop computers took over, they were portrayed as representing the future of office work, and the attitude to things like computer modelling became less sceptical.