Retracted analysis

Since version 8.0

Retracted analysis evaluates the difference between the actual value of a variable (the evidence) and the predicted value of the same variable, when calculated without the evidence set on it. This is known as retracted evidence. The retracted analysis tool calculates the retracted values over a data set.

Retracted analysis can be used to:

  • See which individual pieces of evidence when retracted vary significantly
  • See which individual pieces of evidence significantly impact the overall log-likelihood
  • Aid anomaly detection diagnostics

Retracted analysis

When a retracted value is significantly different to the actual value, this alone does not indicate that the variable is at fault, as it could be a different variable(s) that are causing the difference. However when both the difference and the change in log-likelihood are significant this could indicate a problem with the particular variable.

By setting the Case Column in the Data Map window you can change the order in which cases are output.