Part 1, part 2, part 3, part 4, and part 5 give the context.
Engineers create models. So do architects, economists, sociologists, physicists, biologists, and sculptors, among others. What makes our models special?
Not every single structural engineering problem is life and death, but a large percentage of them carry the possibility of catastrophic failure if dealt with improperly. So our models carry some moral weight for us. They don’t have to be perfectly accurate – they are not and will never be perfectly accurate – but they have to be wrong in a predictable manner. When engineers says that an assumption, or a calculation , or a model is “conservative,” they mean that any inaccuracies are predictably on the side of overestimating loads, underestimating strength, or both. Load and strength are both bell curves* and if safety consists of keeping the load mean lower than the strength mean** then making assumptions that push those bell curves in the desired direction*** is a useful trick. In short, we assume that don’t know exactly what either the loading or the strength is and then we jimmy the math to make sure that the possible error doesn’t accidentally reverse our desired outcome.
The other professions I mentioned try to make their models neutral. They don’t succeed – part of training in the social sciences is learning that you will never be truely neutral – but that is the goal. There is, generally speaking, no penalty for a model failing**** in one direction compared to failing in the other. So structural engineering modeling is out of the ordinary in that regard. The closest comparison is, in some ways, medical diagnosis, where failures again need to be kept conservative. When I wrote yesterday about decades of buildings designed with inaccurate models still being safe, this is what I meant: the models were inaccurate but predictably so.
* Not normal curves, but rather asymmetrical bell curves.
** This is a very rough idea of qualitative safety.
*** Up for load, down for strength and stiffness.
**** Failure in this context means the model being unacceptably bad in terms of predicting empirical reality.
