<span>Dansgaard–Oeschger (DO) warming events oc</span><span>curred throughout the last glacial period. Greenland ice cores </span><span>show a rapid warming during each stadial to interstadial tran</span><span>sition, alongside abrupt loss of sea ice and major reorgani</span><span>sation of the atmospheric circulation. Other records also in</span><span>dicate simultaneous abrupt changes to the oceanic circula</span><span>tion. Recently, an advanced Bayesian ramp fitting method </span><span>has been developed and used to investigate time lags between </span><span>transitions in these different climate elements, with a view </span><span>to determining the relative order of these changes. Here, we </span><span>subject this method to a critical review. Using ice core data, </span><span>climate model output, and carefully synthesised data repre</span><span>senting DO warming events, we demonstrate that the method </span><span>suffers from noise-induced bias of up to 15 years. This bias </span><span>means that the method will tend to yield transition onsets that </span><span>are too early, and we find that the estimated timings of nois</span><span>ier transitions are more strongly biased. Further investigation </span><span>of DO warming event records in climate models and ice core </span><span>data reveals that the bias is on the same order of magnitude as </span><span>potential timing differences between the abrupt transitions of </span><span>different climate elements. Additionally, we find that higher-</span><span>resolution records would not reduce this bias. We conclude </span><span>that time lags of less than 20 years cannot be reliably detected, as </span><span>we cannot exclude the possibility that they result </span><span>solely from the bias. This prevents the unambiguous deter</span><span>mination of the temporal phasing of DO warming events.</span>