The sensitivity to horizontal resolution, the sensitivity to time-sampling and the spread among models are quantified. Regarding the sensitivity to resolution, it is found that in regions where a quantity is non-robust (i.e., highly dependent on resolution) in the control simulation, the response to climate change also tends to be non-robust. For example, in the NH high latitude stratosphere in one AGCM, the boreal winter zonal wind is a non-monotonic function of resolution in the control simulation. In this region, the zonal wind response to climate change is also non-robust: the response switches sign as the resolution is increased. This analysis does not exclude other cases of non-robustness, for example, where a feature is insensitive to resolution in the control simulation but is sensitive to resolution in the warm simulation. But the rule of thumb appears to be that non-robustness in the control simulation is a good a priori predictor for non-robustness in the response to climate change.

