Scale modeling may look old fashioned, but it always features a kind of magic to onlookers. Actually, one may discuss scattering objects for hours in front of a dubious client using some computer simulation results and images without much success, but having the client look at a scale model while being explained the problem is usually quite effective, not to say awesome.
Among early attempts at scale modeling, work was carried out in France in the beginning of the 1940s by Canac  in the study of antique theatres using water as the medium of propagation.
Scale modeling proves efficient when dealing with singularities such as balconies, coupled volumes, and funny-looking surfaces. What scale should be used? Clearly, the smaller the scale, the higher the frequency range will be, and the more complicated the measurements will turn out, as air absorption becomes quite a problem. Scales like 1/10 to 1/16 are rather nice and popular, as air absorption is not yet too much of a problem, but while the model is easy to work with, it also happens to be quite cumbersome. Experiments have been performed with cardboard 1/50 models .
Scale modeling and computer modeling actually are quite complementary: While the former will usually not manage to come up with a good estimate of the RT value, it will nevertheless nicely display any focusing effect. On a high-stakes project, both modeling methods will be used.
As a Short Summary
A statistical model like the Sabine model will enable the user to easily evaluate the rough dimensions and characteristics of a room or hall. This is especially true for small volumes in which their three dimensions are of the same order of magnitude.
A ray tracing model can provide a good estimate of the reverberation time; it may point at unwanted focusing, but will be in difficulty when coupled volumes (e.g., narrow balconies) and complicated surfaces are featured.
Scale modeling enables one to take into account coupled volumes and diffusion, but it can be quite costly and cumbersome.