The International Global H-mode Threshold Database has been revised and cleaned from inadequate data for some tokamaks. This significantly reduces the scatter for the corresponding datasets. The resulting new analyses including all ten tokamaks yields results comparable to previous work. In this case, the uncertainties are essentially determined by the inability of the linear models to take the different dependencies into account with sufficient accuracy. However, restricting the database to a more homogeneous subset of medium and large size tokamaks with single-null elongated plasma shapes yields expressions with much lower scatter. In this case the contribution from the experimental scatter in the individual machines plays a role. These regressions are stable against excluding one tokamak in turn.