In the case of global warming, the consensus view, as represented by the IPCC, is that we ought to have a very high confidence (ie 90% probability of being correct) that the warming currently experienced will progress over the next two decades at 0.2ºC per decade and by more in subsequent decades unless global emissions are stabilised at 25% below current levels by 2050 and reduced by 80% in the long run. Increases in emissions from developing countries must be balanced by reduction in more developed economies. The likely range (ie probability of being correct of at least 80%) of temperature increases by year 2100 for various scenarios involving increases of emissions have upper limits from 2.9ºC to 6.4ºC and lower limits from 1.1ºC to 2.4ºC. The IPCC regards these estimates as a considered basis for advice to policy makers.

Practical reasoning based on such advice could proceed on the basis of expected benefit: here the benefits minus the costs of various scenarios would be multiplied by the probability of their occurrence (based on the consensus view) to identify the best policy, assumed to be that with the best expected outcome. In precautionary reasoning, policy makers ought to adopt policies that maximise avoidance of the worst possible outcomes, even at the cost of forgoing some of the benefits of adopting policies with better expected outcomes.

Prudential reasoning has the desirable feature of dynamic consistency, that is, if you act for the best on its basis, it will be best to continue with that course of action, whatever happens. Prudential reasoning also accommodates inclusive measures of costs and benefits. These integrated measures of costs and benefits may be unrealistic if welfare is a multidimensional quantity. However, they permit rational comparisons between alternatives on the basis of the same simplification, provided we have no reason to suppose that the simplification distorts the welfare outcomes of one policy alternative any more than that of any other.

Precautionary reasoning, on the other hand, can be modelled by weighting the expected benefit of the worst outcomes (Cf. Kahneman and Tversky (2000), Quiggin (2001)). Some of these models may not be dynamically consistent. However, while this makes some views of choice with uncertainty less desirable, it is not a decisive objection: people often come to think in hindsight that it might have been better not to take the precautions they did, although they continue to consider those precautions rational at the time they took them.