The Meaning of Likelihood in IPCC Assessment Reports
The following table gives the probability range for the various likelihood descriptions in the report. This is no problem for the drafters of the report to add another language to a simple mathematical description, since they are professional academics with advanced language skills. They stop others like me in their tracks.
For presentations in the vernacular, the 99% probability could be called better than 100 to 1 for. The 95% better than 20 to 1 for, the 90% better than 10 to 1 for, the 66% better than 2 to 1 for.
“More likely than not” is so vague outside of the table, but what is really meant is better that 50-50. At first, I thought that it only covered the inclusive range of 50% to 66%, but then decided that that was too exacting, and “More likely than not” has to be taken literally only. That range also overlaps the 33% to 66% category “about as likely as not”, and is therefore only a little bit confusing. Also confusing is the lack of a symmetrical phrase for between 33% and 50%, although I imagine they can use “Less likely than sure” instead of less than 50-50.
Then to use the numerical range of 33% to 66%, it could be “the middle 1/3 of likelihood about equal chances”, or just plain 33% to 66% probable. We all read or hear rain predictions every day that run from 0% to 100% numbers, so just sticking with the original percentages is not going to confuse the average public.
A gap in probability stages exists between the “10 to 1” and “2 to 1” stages, since these differ by a factor of 5, while “20 to 1” differs from “10 to 1” by a factor of two, and “2 to 1” differs from 50-50 by a factor of two. The appropriate stage could be “5 to 1” following the decimal “20 to 1” and “10 to 1” stages. Similarly, at the low end, “1 to 5” odds would fill in the gap between “1 to 2” and “1 to 10”.
By the way, what happened to the standard phrases “like winning the daily double”, “its a sure thing”, “like drawing to an inside straight”, and “you bet on WHAT?”.
The caption also says that when ranges are cited they are at 90% confidence intervals. That should be cited everywhere it is used, rather than just at one place in the thousands of pages of documents. The casual reader might just be interested in one page or point of the report, and shouldn’t have to spend hours searching through thousands of pages to find what the range is. The authors of the report are so familiar with the conventions that they don’t realize how cumbersome it can be for the general scientific public to read a small section of the report, or even the summary, or even a single table.