This recent article in The Guardian caught my eye (thanks to Euan Ritchie for posting it on twitter).
The authors argue that human society is approaching the limits of growth that our planet can sustain and we will soon be faced with the collapse of civilization as we know it.
A group called the Club of Rome developed a very sophisticated model over 40 years ago. The model predicts global collapse in 2015-2030. Turner and Alexander argue that the the model has done a great job predicting the last 40 years, so we should be very concerned about what will happen in the next 15 years or so.
While I am certainly concerned about environmental sustainability of human endeavours (they aren’t at a global scale!), I do not agree that the historical data presented in the new study is convincing evidence that this Club of Rome model’s predictions will unfold.
Why? Well the co-trending of data since 1970 and the model’s predictions could well be spurious.
As an aside I saw this excellent post on my same twitter feed. I recommend reading it, it’s an example of how human bias can lead to spurious findings.
Scroll down through The Guardian article or the author’s original report and you will see plots of several variables over time, as collated from global datasources and predicted by the 1970 model.
The variables include global population size, birth rate, death rate, among others, plotted on a ‘common scale’. The use of a ‘common scale (normalized values)’ is the first hint that something may be wrong. Often it is very hard to model actual values, but models can do a better job of predicting relative change. So we normalize model’s output to give a picture of relative change.
Plotting ‘normalized’ model predictions and ‘normalized’ data can be misleading, because it gives the impression the model is predicting exactly what is happening. In reality it is only predicting the trend.
The second, bigger issue, relates to the reliability of predicting trends.
The key predictions of the 1970s model all revolve around a series of ‘turning points’: points in time where food production and services output suddenly drop and consequently birth rate drops, death rate increases and global population peaks then declines.
Well none of the data show a drop yet. So the validation of the model is only done on increasing or decreasing trends. It is pretty easy to build a model that predicts a continuing trend and such correlations of trends over time are often spurious.
The real test of the model will be when the drop happens, or doesn’t.
To the authors credit they do spend some time looking at whether the processes underlying the model are likely to be realised, like peak oil. It is having plausible evidence for processes that distinguishes causation from correlation.
However, the main argument for the 1970s model prediction of collapse is the historic comparison of model predictions to data.
We have learned a lot since the 1970s. There is still widespread belief among scientists that we are approaching global planetary boundaries, and some may have already been exceeded. However, it is not clear whether all of them have been reached yet and if they lead to the collapse of civilization. Not least because of the highly unpredictable role of technology in human change.
So while we should be concerned about doomsday scenarios, we should not give up hope for a better future on the basis of spurious correlations.
Designed by Chris Brown. Source on Github