Empirical scaling laws (Goldston, Kaye-Goldston, ITER89P and ITER93H) for the confinement time τE in tokamaks depend on dimensionless global plasma parameters in a way which is difficult to associate with some general plasma physics model. It is shown that such dependences can arise from artefacts in the database used to derive the scaling expression. The artefacts involve inner relationships, for example, collinearities between variables. For the ITER L mode and H mode databases, it is shown that subsets of the scaling expressions are approximately constant. An alternative description of L mode, ELM-free, small ELM and giant ELM H mode data is given. The alternative description does not include the dependences upon plasma β, elongation and safety factor which are present in the above scaling laws; extrapolations to ITER parameters by this alternative data description are virtually the same as those by ITER89P and ITER93H. The alternative description is based on a simple electrostatic physics model (plateau scaling) in which confinement is degraded by non-collisional processes. It can describe global confinement data for all four confinement regimes; the common scaling features a multiplier C which is assigned a different value for each regime. The complex physics associated with MHD instabilities (ELMs, sawteeth, outer modes) is hidden in the multiplier C, and attempts are made to uncover such physics. In these attempts global confinement is made to scale with physics models rather than physics variables.