We investigate the difference between those relativistic models based on in
terpreting a realistic nucleon-nucleon interaction as a perturbation of the
square of a relativistic mass operator and those models that use the metho
d of Kamada and Glockle to construct an equivalent interaction to add to th
e relativistic mass operator. Although both models reproduce the phase shif
ts and binding energy of the corresponding nonrelativistic model, they are
not scattering equivalent. The example of elastic electron-deuteron scatter
ing in the one-photon-exchange approximation is used to study the sensitivi
ty of three-body observables to these choices. Our conclusion is that the d
ifferences in the predictions of the two models can be understood in terms
of the different ways in which the relativistic and nonrelativistic S matri
ces are related. We argue that the mass squared method is consistent with c
onventional procedures used to fit the Lorentz-invariant cross section as a
function of the laboratory energy.