When geotechnical 1D ground response analyses are performed to evaluate site effects in lieu of the use of attenuation relations, it is with the expectation that the attenuation model's standard deviation, and any bias in the median, would be reduced. In this article, we evaluate the degree to which these and other benefits of equivalent-linear, 1D ground response analyses are realized and develop recommendations for implementing the results of such analyses into hazard calculations. This is accomplished by comparing response spectral accelerations (Sa) from recordings to predictions derived using ground response analysis procedures as well as attenuation relationships with and without amplification factors. The results are compiled for 134 motions from 68 sites, and prediction residuals are interpreted to assess the models' relative bias and dispersion. We find that ground response analyses are unbiased for T ≤ ∼1 sec, but underestimate longer period Sa in deep basins. For soft soils, ground response analyses reduce dispersion for T < 1 sec relative to alternative models. This dispersion reduction is not observed for other site categories nor at longer periods. These results suggest that ground response analyses are beneficial for Sa predictions at soft soil sites, but generally provide no identifiable benefit for typical stiff soil or rock sites.