The Effectiveness of the Game of Dice Task in Predicting At-Risk and Problem Gambling Among Adolescents: The Contribution of the Neural Networks.

Abstract

The Game of Dice Task (GDT; Brand et al. in Neuropsychology 19:267-277, 2005a; Psychiatry Res 133:91-99, 2005b) measures decision-making under objective risk conditions. Although disadvantageous decision-making has been shown in individuals with substance dependency, such as pathological dependency, any studies have been conducted with adolescents by using the GDT to investigate the relationship between the performance on the task and gambling behavior. Moreover, all the previous studies have considered only the GDT net score and not the single choices. In the current study, focusing on adolescents, we wanted to investigate the relationship between the sequence of the choices at the GDT and gambling behavior, measured with the SOGS-RA. To analyze the predictive power of the sequence of choices made in the GDT and problem gambling and gambling frequency, we used the Neural Networks (NNs), which are often used to find relationships between a series of input actions and the correspondent empirical outputs in order to discover behavioral patterns that may be predictive of at-risk behaviors. Results showed that neither a linear or a non-linear relationship could be detected between the GDT performance and the SOGS-RA classification both in terms of gambling problem severity and gambling frequency. Indeed, different training algorithms produced different performances of the NN on the training sets, but all of them showed a very low prediction capability on new samples. Thus, the performance at the GDT did not discriminate between adolescent gamblers with different and progressive levels of problematic gambling behavior and gambling frequency. Limitations and future studies are discussed.