Testing rules of thumb

Yesterday our paper on damselfish movements came out online. Its available online and open access, so download it as many times as you want.

This research started when James 'Teddy' Herbert-Read, Alex Jordan and Ashley Ward went up to the University of Sydney's One Tree Island field station on the Great Barrier Reef. Damselfish live on the coral reefs there and frequently have to decide when to move between the relatively safety of one bit of coral to another. Teddy set up an experiment to test how they make these moves, placing two bits of coral at either end of a long tank. Then he filmed and tracked the fishes movements back and forward.

Based on our previous work, we expected the fish to make these risky moves between coral together in small groups. This time we wanted to know more about the 'rule of thumb' they use to decide when to go. It turned out to be deceptively simple. Roughly speaking, it is "follow the last fish": if you are a fish on one of the corals then you are more likely to leave if another fish has just left. It appears that the total number of fish on the coral is less important than where the last move was made.

Recently, in our research, we have tried to become more precise in how we test which rules of thumb animals are using to make decisions. We use a method developed by Richard Mann for comparing how well different behavioural rules explain the data. In this case, the "follow the last fish" rule gave the most convincing model of the moves made by the fish. Other rules of thumb might give a better fit when we tune the parameters, but when we test the model robustness using Bayes factor, the "follow the last fish" rule gave the best predictions.

In the earlier study (I mentioned above) we had found that a different rule---"that the probability of leaving increases with number which have left"---fitted the data best. This worried me a bit at first, since we wouldn't necessarily expect the fish to use different rules in different contexts. Furthermore, Richard was pretty confident that if he used his method on the data from the earlier study then he would get a better fit with the new, simpler "follow the last fish" rule. It was with some trepidation I emailed him the data from the earlier study. It isn't fun to have to admit you might have got it wrong, especially after something is published!

But it turned out my rule still came out top when he calculated Bayes factor on the older experiments. The differences in the experimental setup, and probably differences in the risks perceived by the fish, mean that the fish that they use different rules in different contexts.  You can find out more by reading the paper!