Editorial: 鈥�Cause test could end up in court鈥�
IT鈥橲 one of the fundamental problems in science: when two events take place, can we tell whether one directly caused the other, or if the two follow a similar pattern for a different reason?
Separating correlation and causation is notoriously difficult. People might become more tanned when they eat more ice cream, but that doesn鈥檛 mean ice cream gives you a tan; it probably just means it鈥檚 summer.
Advertisement
In many cases the link between two variables seems more likely to be causal 鈥� for instance, smoking and lung cancer, or atmospheric carbon dioxide levels and climate change. But because it is so tricky to prove causality, it鈥檚 easy for those sceptical of the link to belittle it with reference to spurious correlations like ice cream and tanning.
A new mathematical test could help settle these arguments. It is the first of its kind that can strengthen the claim, even in complicated scenarios, that one event really has led to another.
鈥淚t allows you to find out which things are actually affecting one another,鈥� says of the Scripps Institution of Oceanography in La Jolla, California.
Sugihara has applied his test to a real-world challenge: preserving a major fishery. The US National Oceanic and Atmospheric Administration (NOAA) has been managing the Californian sardine population for decades, but Sugihara says his test shows that its methods need upgrading.
When scientists look for cause-and-effect relationships, and can鈥檛 do controlled experiments, they often look for correlations. If lung cancer is more common in smokers than non-smokers, it suggests that smoking causes lung cancer.
Once a correlation is found, the so-called can sometimes be used to strengthen the case that there is a causal link at work. This can help differentiate it from cases like ice cream and tanning, where the apparent correlation is misleading 鈥� for example, there are probably peaks in ice-cream consumption at other times of year too, such as during a festival.
This test simply asks if one variable is useful for predicting another. It shows, for instance, that people who smoke are more likely to develop lung cancer, which suggests the two are causally linked.
鈥淕ranger causality has been a very powerful technique,鈥� says Sugihara. But it only works under relatively simple circumstances. For example, the test can鈥檛 be used on predator-prey relationships. Predators kill their prey, but if there are fewer prey then some predators will starve. So predator populations are a good predictor of prey populations, but prey numbers are also a good predictor of predator populations. The Granger test doesn鈥檛 work when causality goes both ways like this.
鈥淧redator populations are a good predictor of prey numbers, and vice versa, but it鈥檚 hard to show that鈥�
Sugihara says his new test, convergent cross mapping (CCM), can deal with two-way causality. What鈥檚 more, the test can ferret out various causal linkages in systems with several variables.
CCM asks whether one variable predicts another, much like Granger causality. To deal with the two-way causality problem, each data set is put through mathematical transformations, creating a three-dimensional shape called a manifold. Points on one manifold may be used to predict points on the other, but not necessarily the other way round. That means causal relationships, of one direction or another, can be measured separately ().
Sugihara tested his method on a real-world case where the causal relationships have already been established. The microorganism preys on another, . The population of each species affects the other, but Didinium鈥榮 effect on Paramecium is much stronger than the reverse. 鈥淭he predator has left a gigantic footprint on the prey,鈥� Sugihara says. CCM detected both causal relationships, and their respective strengths.
Several statisticians contacted by 快猫短视频 said CCM looks promising. 鈥淚t will be worth studying in more detail,鈥� says of the University of Oxford. Others are less keen. According to , who studies causality at the University of California, Los Angeles, the paper may be theoretically flawed. The starting point of any study is a hypothesis about why a causal relationship exists, he says. Sugihara鈥檚 test instead involves scouring data for links and then retrofitting a hypothesis to suit.
Flawed or not, fisheries scientists are keen to use CCM to , says of the National Marine Fisheries Service in Woods Hole, Massachusetts.
Sugihara has applied CCM to the , which was once among the biggest fisheries in the world, but collapsed in the 1940s. By the 1980s the sardines had recovered enough to be harvested stably. The modern fishery has lasted longer than any other, says of the NOAA Fisheries Service in Santa Cruz, California.
To keep it that way, NOAA sets fishing quotas according to oceanic temperature, which is thought to influence sardine numbers. That makes the fishery the only one in the world to base quotas on environmental factors.
But Sugihara鈥檚 analysis of the fishery using CCM suggests the quota system could be improved.
NOAA chose to base sardine quotas on ocean temperatures because decades-worth of data seemed to suggest that lower temperatures stunt the sardine population.
By the 1990s, though, the effect seemed less strong (). Sugihara鈥檚 method suggests why: it鈥檚 the size of the fish populations at the time of the temperature change that determines the effect. Warm temperatures can cause the population to rise or fall depending on the population size before the temperature change, Sugihara says.
That suggests sardines are sometimes vulnerable to collapse even when seas are warmer than the danger threshold. Quotas should be based on predictions of population change, incorporating temperature and other factors.
Sugihara is now applying CCM to other fisheries to find out which factors affect which species, and which don鈥檛. Beyond fisheries, the same principles could be used to better understand all manner of systems, he says 鈥� including the world economy.
