I've not attended any formal classes on the subject but the few texts and papers that I've looked thru seem to use the same assumptions that almost everyone else does, and that is to use population parameters instead of sample statistics. If using say a 30 piece sample and only having a sample mean and sample standard deviation it's tough to make statements about 3.4 ppm at any reasonable level of confidence without resorting to large intervals.

The 1.5 sigma meanshift is just an assumption, and either you buy it or you don't. It's a better assumption than no meanshift, which is what a lot (most ?) seem to use when something like statistical tolerancing. The 1.5 sigma meanshift wasn't conservative enough for a lot of our processes so we used uniform distribution assumptions.