Sorry I missed a word in my last post; second paragraph, after the parentheses the word “uncertainty” is missing.
I want to cover two more concepts. It may feel like I’m lecturing a bit; that’s not my intent and I don’t insist or expect anyone agree with me. I’m simply reflecting on my own observations about how people think probabilistically, as I’ve observed in younger people (junior hires), colleagues, clients and the general public. Others’ experiences and points of view may differ from mine.
I want two briefly discuss two more progressively more difficult / nuanced concepts: the first revolves around Taleb’s writings and the second underscores my own reluctance to get off the fence all these years.
Taleb writes about how rare large events dominate many phenomena. Most people may assume that all random phenomena can be described by a normal distribution, or bell curve, or some other distribution with mild variability. When that assumption holds true, we can often used measures of central tendency to describe the expected long term behaviour. However, many phenomena have wilder variability, or heavy tails in their probability distributions. These may decay as a power law instead of exponentially. If the slope of the power law is sufficiently steep, then the long term behaviour is dominated by rare, major events (or black swans in Taleb’s terminology).
I make this first point as a reply to Think Again’s comment about expected value; it’s important to think carefully about how you are representing the inputs, which are random variables and may not be best represented by a simplification like mean, median or some other measure, if they have relatively wild variability.
My second point, which remains, and will alimony certainly remain until I die, the conceptual barrier preventing me from taking one foot of the fence, deals with non-linearity. CR mentions the use of regression as a way to tease out important variables. The meaning of any regression depends strongly on the statistical strength of the range of input values, and also more strongly on the nature of relations between variables. If response is nonlinear, it complicates the analysis considerably, particularly if more than one relationship is nonlinear, or if the relationship between two variables depends on the value of one or more other variables.
If predicting future climate were straightforward, or by extension determining the importance of various forcing functions were easy and confident, we would have good models that give strong predictions. Which is not the case. Hence our ability to be certain (or at least very confident) of the degree of importance of human influence on climate is very weak.
Which leads me personally back to the subjective, or Bayesian, basis for estimating probalities to inform my personal opinion.
All that aside, I don’t ask or expect anyone to agree with where I’ve landed on the question of anthropogenic influence. I remain much more interested in social context and being robust against global warming independent of its cause.