When spin-orbit torques are measured using spin-torque ferromagnetic resonance, two alternative ways of analyzing the results to extract the torque efficiencies - lineshape analysis and analysis of the change in linewidth versus direct current - often give inconsistent results. We identify a source for these inconsistencies. We show that fits of spin-torque ferromagnetic resonance data to the standard analysis framework leave significant residuals that we identify as due to (i) current-induced excitations of a small volume of magnetic material with magnetic damping much larger than that of the bulk of the magnetic layer, which we speculate is associated with the heavy-metal-magnet interface and (ii) oscillations of the sample magnetization at the modulation frequency due to heating. The dependence of the residual signals on direct current can interfere with an accurate extraction of spin-torque efficiencies by the linewidth method. We show that the discrepancies between the two types of analysis can be largely eliminated by extrapolation of the window of magnetic fields used in the linewidth fits to small values so as to minimize the influence of the residual signals. © 2021 American Physical Society.