Example Bayesian Network

Using data from a psychological study, a Bayesian network model is used to aid the discovery of connected variables within a questionnaire.

Each participant was asked a selection of questions and had to answer either ‘Agree a little’, ‘Agree strongly’, ‘Disagree a little’, ‘Disagree strongly’ or ‘Neither agree nor disagree’. The answers scored individuals on their agreeableness, conscientiousness and extraversion.

For this example, the model is aiming to discover the questions that link to how agreeable an individual is. With this in mind, the ‘Agreeableness’ field was selected as the classifier.

The default settings were used and the following network was generated:

Due to the size of the network, Tile View was then used to manipulate probabilities, changing the outcome of the classifier to 100% probability of scoring in the highest band.

By making this change, viewing the other nodes in the network gave an indication of the qualities an individual posesses if they are to be deemed ‘agreeable’.

From the above nodes, the results indicate that individuals who fall into the highest scoring category spend time reflecting on things, thinking of others while professing to ‘have a soft heart’.

If the model is accurate, changing the classifier to 100% probability of scoring in the lowest band should produce very different results.

With individuals who score in lowest band, the results suggests that they feel little concern for others, are quiet around strangers and and do not have an interest in people.

While these are only simplistic examples, they do show how individual nodes are linked and how they affect the network as a whole. Other tests could include comparing answers beteen male and female participants or between localities – the only limitation is the data that is analysed.