We present a method for the global analysis of elastic electron-proton scattering when combining data sets from experiments with different overall normalization uncertainties. The method is a modification of one employed by the NNPDF collaboration in the fitting of parton distribution data. This method is an alternative to the ‘penalty trick’ method traditionally employed in global fits to proton electric and magnetic form factors, while avoiding the statistical biases inherent in that approach. We discuss issues that arise when extending the method to nonlinear models. For data with Q2>1GeV2 we find relatively minor differences to traditional model fits when the normalization uncertainties from different experiments are correctly accounted for. We discuss implications of this method for the well-known discrepancy between the form factor ratio GE/GM extracted from the Rosenbluth and polarization transfer techniques. Published by the American Physical Society 2024