How can we take heterogeneity of both lies and lying preferences into account when analysing and predicting individual lying decisions? This paper provides a unifying framework to analyse and predict lying behaviour. At the theoretical level, the framework provides a complete categorisation of lie types and a simple, testable model. At the experimental level, the framework guides a novel experimental design which allows the observation of individual choices, can assess variation of behaviour across lie types as well as disentangle the effects of lie types from standard social preferences. I find that varying the lie type has a large and significant impact on behaviour and that this behaviour is highly heterogeneous across individuals. In order to quantify the heterogeneity of decision-makers, I employ an unsupervised machine learning algorithm to identify coherent types of decision-makers and their distribution in the data. I show that the identified types are meaningful as accounting for them in an out-of-sample forecasting exercise significantly improves the precision of the forecast. In contrast, I find that social preferences have only limited explanatory power for lying preferences. Finally, to show that my framework can be used for the purpose of model building and prediction, I present a parametric version of the framework and calibrate it at the individual and the group level.

Practical information