The spread of model estimates of equilibrium climate sensitivity (ECS) and transient climate response (TCR) is often interpreted as a measure of the uncertainty in these quantities.  The spread in ECS and TCR, however, is impacted by multiple factors, including the spread in individual climate feedbacks, the spread in forcing, and, for TCR, the spread in ocean heat uptake efficiency.  Furthermore, the individual factors producing the spread are not independent, and in some cases negative correlations between factors tend to mask the true extent to which there is model disagreement. 

We analyzed the CMIP experiments with abrupt quadrupling of CO2 and with an imposed 1% per year increase.  Despite the nonlinear nature of the feedback formulation and the lack of factor independence, we have been able to separate the variance of ECS and TCR into contributing terms that have reasonably straight-forward interpretation.  This enables us to analyze the CMIP5 and CMIP6 models and say why the spread in normalized ECS increased and normalized TCR decreased slightly from CMIP5 to CMIP6.  For ECS the increase in spread of the climate sensitivity parameter, due primarily to an increase in the spread of cloud feedbacks, nearly doubled the spread in normalized ECS.  In contrast, for TCR, the increase in spread of climate sensitivity parameter was more than offset by decreases in the spread of forcing and the spread of ocean heat uptake and an increase in the negative correlation between the CO2 forcing and the climate sensitivity in models.  This resulted in a decrease in the spread of normalized TCR.