Mac Planck Institute for Research on Collective Goods
We provide an example for an errors in variables problem which might be ofen neglected but which is quite common in lab experimental practice: In one task, attitude towards risk is measured, in another task participants behave in a way that can possibly be explained by their risk atitude. How should we deal with inconsistent behaviour in the risk task? Ignoring these observations entails two biases: An errors in variables bias and a selection bias. We argue that inconsistent observations should be exploited to address the errors in variables problem, which can easily be done within a Bayesian framework.