Level of Significance

The structure learning algorithms are based on making dependence tests that calculate a test statistic which is asymptotically chi-squared distributed assuming (conditional) independence. If the test statistic is large for a given independence hypothesis, the hypothesis is rejected; otherwise, it is accepted. The probability of rejecting a true independence hypothesis is given by the level of significance.


Figure 1: Text box in which the user specifies the level of significance to be used by the structure learning algorithms.

The level of significance can be entered into this text field. The default value is set to 0.05, which should be appropriate for most learning sessions, but it can take on any value in the open interval (0;1). In general, the higher the significance level the more links will be included in the learned structure. Reducing the level of significance will usually also reduce the running time of the structure learning algorithms.

See also the PC algorithm and the NPC algorithm.