Warren Meyer explains why computer models can be incredibly useful tools, but they are not the same thing as an actual proof:
Among the objections, including one from Green Party politician Chit Chong, were that Lawson’s views were not supported by evidence from computer modeling.
I see this all the time. A lot of things astound me in the climate debate, but perhaps the most astounding has been to be accused of being “anti-science” by people who have such a poor grasp of the scientific process.
Computer models and their output are not evidence of anything. Computer models are extremely useful when we have hypotheses about complex, multi-variable systems. It may not be immediately obvious how to test these hypotheses, so computer models can take these hypothesized formulas and generate predicted values of measurable variables that can then be used to compare to actual physical observations.
[…]
The other problem with computer models, besides the fact that they are not and cannot constitute evidence in and of themselves, is that their results are often sensitive to small changes in tuning or setting of variables, and that these decisions about tuning are often totally opaque to outsiders.
I did computer modelling for years, though of markets and economics rather than climate. But the techniques are substantially the same. And the pitfalls.
Confession time. In my very early days as a consultant, I did something I am not proud of. I was responsible for a complex market model based on a lot of market research and customer service data. Less than a day before the big presentation, and with all the charts and conclusions made, I found a mistake that skewed the results. In later years I would have the moral courage and confidence to cry foul and halt the process, but at the time I ended up tweaking a few key variables to make the model continue to spit out results consistent with our conclusion. It is embarrassing enough I have trouble writing this for public consumption 25 years later.
But it was so easy. A few tweaks to assumptions and I could get the answer I wanted. And no one would ever know. Someone could stare at the model for an hour and not recognize the tuning.
