The "roulette" wheel below depicts the MIT Joint Program's estimation of the range of probability of potential global warming over the next hundred years, assuming a scenario in which "no policy" action is taken to try to curb the global emissions of greenhouse gases.
The face of the wheel is divided into six slices, with the size of each slice representing the estimated probability of the temperature change in the year 2100 falling within that range.
The size of the slice for greater than 7 degrees Celsius warming (shown in red) has a probability of 9%. Or, if stated in another way, that probability has the same likelihood as the "odds" of (about) 1 chance in 11. The slice representing the smallest predicted change, less than 3 °C (shown in blue), has a probability of less than 1% (1 in 100 odds).
The median value, that level where there is a 50% chance of falling above or below (even odds) is 5.2 °C. The other areas of the wheel represent the following likelihoods of occurring: 3 to 4 °C, 12% (about 1 in 8); 4 to 5 °C, 30% (almost 1 in 3 odds); 5 to 6 °C, 33% (1 in 3 odds); 6 to 7 °C, 15% (about 1 in 7 odds).
The numbers given above are obtained in simulations with input climate parameters based on the Levitus et al. (2005) estimates of 20th century changes in the heat content of the ocean. If the distribution for climate parameters based on a recently published analysis by Domingues et al. (2008) is used instead, the median projected warming is only 4.1 °C and the 90% probability range of the warming, which is 3.5 °C to 7.4° if the Levitus et al. data is used, is reduced to 2.7 °C to 5.7°. While more oceanic heat uptake in the Domingues et al. case leads to lower estimated warming in year 2100 than the Levitus et al. case, it leads to higher ultimate warming under stabilization because the median effective climate sensitivity of 3.6 °C implied by the Domingues et al. analysis is significantly greater than the 2.8 °C implied by the Levitus et al. analysis.
For more technical detail see Sokolov et al., 2009.