The statistical methodology at the core of alarmist climate claims is flawed. Climate attribution method overstates “fingerprints” of external forcing Posted on December 18, 2023 by curryja by Ross McKitrick I have a new paper in the peer-reviewed journal Environmetrics discussing biases in the “optimal fingerprinting” method which climate scientists use to attribute climatic changes to greenhouse gas emissions. This is the third in my series of papers on flaws in standard fingerprinting methods: blog posts on the first two are here and here. Climatologists use a statistical technique called Total Least Squares (TLS), also called orthogonal regression, in their fingerprinting models to fix a problem in ordinary regression methods that can lead to the influence of external forcings being understated. My new paper argues that in typical fingerprinting settings TLS overcorrects and imparts large upward biases, thus overstating the impact of GHG forcing. While the topic touches on climatology, for the most part the details involve regression methods which is what empirical economists like me are trained to do. I teach regression in my econometrics courses and I have studied and used it all my career. I mention this because if anyone objects that I’m not a “climate scientist” my response is: you’re right, I’m an economist which is why I’m qualified to talk about this. I have previously shown that when the optimal fingerprinting regression is misspecified by leaving out explanatory variables that should be in it, TLS is biased upwards (other authors have also proven this theoretically). In that study I noted that when anthropogenic and natural forcings (ANTH and NAT) are negatively correlated the positive TLS bias increases. My new paper focuses just on this issue since, in practice, climate model-generated ANTH and NAT forcing series are negatively correlated. I show that in this case, even if no explanatory variables have been omitted from the regression, TLS estimates of forcing coefficients are usually too large. Among other things, since TLS-estimated coefficients are plugged into carbon budget models, this will result in a carbon budget being biased too small. Continue reading →
