The addition of high power, low aspect ratio data from the NSTX and MAST experiments has motivated a new investigation of the effect of aspect ratio on confinement scaling. Various statistical methods, including those that incorporate estimates of measurement error, have been applied to datasets constrained by the standard set of criteria in addition to the range of kappa and M-eff appropriate to ITER operation. Development of scalings using engineering parameters as predictor variables results in epsilon-scaling coefficients that range from 0.38 to 1.29; the transformation of these scalings to physics variables results in an unfavourable dependence of beta tau on beta, but a favourable dependence on epsilon. Because the low aspect ratio devices operate at low B-T and therefore high beta(T), a strong correlation exists between epsilon and beta, and this makes scalings based on physics variables imprecise.