In calculating unknown equilibrium constants by the least-squares tech
nique from pH (or potentiometric) titration data, errors in the chemic
al model (initial volume, reactant concentrations, carbonate or other
impurities, pK(w) and other known equilibrium constants) and measureme
nt errors (electrode calibration, drifts in ionic strength, non-ideal
titrant mixing or temperature control) can strongly influence the resu
lts. Most of these errors will produce non-randomly distributed pH (or
e.m.f.) residuals over some regions of data and thus violate the assu
mptions of the least-squares method. Simulations and deliberate errors
in real data show that fitting the point-to-point changes in pH is le
ss sensitive to such errors than is fitting the raw pH data, producing
more trustworthy and reliable refinements of unknown equilibrium cons
tants and more randomly distributed residuals. The refinement results
are also insensitive to errors in calibration and more robust with reg
ard to the range of pH spanned by the data. When applied to titrations
of glycine-proton and Ni2+-glycine-proton mixtures from one laborator
y, the results more closely matched the averages from several laborato
ries. Further, this approach will actually signal the presence of unsu
spected errors.