test of independence

[nadjet]

Hello
I have a simple question concerining the statistical test used for (conditional) independence testing:
If I remember correctly, Hugin uses the chi square test of independence. is that right?
and if so, when learning a model from a dataset, does the data have to be normally distributed for independence testing?

Many thanks

Nadjet

[Frank Jensen]

No, the data is assumed to be qualitative (i.e. discrete and not numeric).

[nadjet]

Dear Frank,
thank you for your reply, BUT I still do not understand your answer. I beielve the data has to be normally distributed for chi square test of independence. how does this relates to any data used when learning a model? do we have to check that the data is normally distributed or ...?

Pls if you know any review/book i should refer to understand this let me know.
thank you

[Frank Jensen]

We use the likelihood test statistic G^2 described in the following book:

P. Spirtes, C. Glymour, and R. Scheines.  Causation, Prediction, and Search.  Adaptive Computation and Machine Learning.  MIT Press, Cambridge, Massachusetts, second edition, 2000.

[nadjet]

thank you, I will c consult it

Regards