To address the issue of large deviations in the traditional dynamic shear stress reduction coefficient (r_d) model for predicting earthquake-induced settlement in loess sites, the three-factor r_d calculation method is introduced to loess sites for the first time. The study indicates that depth (D), magnitude (M), and the 12-m equivalent shear wave velocity (v_s,12^*) are the three most significant factors affecting r_d. Therefore, based on measured borehole wave velocity data from 48 typical sites on the Loess Plateau, the SOILQUAKE program was used to conduct equivalent linear site response analysis, and a three-factor r_d model incorporating the three significant factors was introduced and verified through comparison with traditional methods. This model has a continuous curve form within the depth range of 0-30 m in thick loess sites, overcoming the shortcoming of traditional methods that rely on depth interpolation and splicing. The model reasonably reflects the physical laws that the r_d curve shifts leftward and rises with increasing magnitude, and tends to become straighter with increasing shear wave velocity. In validation at a typical site in Lanzhou, the predicted seismic subsidence of this model remained within 1.3 to 1.8 times the SOILQUAKE benchmark value, whereas the traditional method sharply increased from 2.0 times to 8.3 times, exhibiting significant non-physical growth. In the hazard assessment of the 2023 Jishishan M6.2 earthquake, the model's predictions were generally consistent with the field investigation scope, preliminarily verifying the model's applicability. This method only requires conventional investigation to obtain v_s,12^* and can quickly generate an r_d profile, greatly reducing data requirements and computational costs, thus providing an economically feasible technical solution for rapid earthquake disaster assessment in loess regions.