Saturday, 2 January 2016

How Not to Model Human Behaviour

... humans are not random. They (we) are strange and wonderful. Their behavior may be unexpected or inconsistent (i.e., noisy), but it is not random. As an example, here is a simple demonstration. An easy question will be presented below and you may take hundreds of milliseconds to answer, but do answer. The question is: “Pick a number between one and four.” Have an answer?

The most common response is “three” and there is a secondary effect of this task: people feel a need to explain why they chose whatever answer they did. The second most common answer is “two”. Very few people decide to respond with either “one” or “four”. Sadly, there is not a serious study of this behaviour but undocumented sources suggest that the response statistics are close to 50 percent for “three”, 30 percent for “two” and about 10 percent for the other two answers.

The common explanation for the selection of “three” is that it was the most “interesting” number in the range. There is also a small number of people who are compelled to answer outside of the range, with fractions, or irrational numbers. These are rare occurrences. Similar results are obtained when the task is to pick a number between one and twenty. The similarity is that people pick their most interesting number. For this range, the most common response is 17, occurring about 40 percent of the time, well above a “rationally”, “logically” expected 5 percent. Other primes are also favored as answers because they too are interesting.

This behaviour is interesting. The decision-­‐making process should be simple, but it certainly does not appear to be a simple random selection among equally likely options. What this shows us is that people cannot even be random when they want to be. Further, if this task had been modeled as a uniform random distribution among equally likely choices, it would have been very different from actual behaviour.

Modeling human choices as uniform random distributions is making a very serious claim about human behavior. It is saying that all choices are equally likely even when we know nothing about how people actually decide. It also assumes people have no preferences, do not consider the consequences of their actions, have no memory of previous choices, and can be more consistent than the data shows. Modeling human behavior requires some data or some experience. Luckily, modelers are human and should know better.
Kennedy, William G. "Modelling human behaviour in agent-based models." Agent-based models of geographical systems (2012): 167-179.