Re: Normalizing the ER model

David Cressey wrote:
>I think whether one wishes to apply principles of normalization to an ER
>model or not depends on what one is using the ER model for.

Just for the record, I never argued that one always *should* apply normalization to ER models, just that one sometimes *could*. Greg seems to think that even the very idea is already quite absurd. :-)

>But there's a different way to use ER modeling. If you use ER for analysis
>and relational for design, then normalization becomes pretty much beside
>the point for the ER model. Why do it? And I think ER is more usefully
>used this way.

Hmmm, this smells a bit like the old discussion about whether normalization is a logical issue (getting your terminology straight) or an implementation issue (minimizing redundancy, avoiding update anomalies, optimizing constraint maintenance, et cetera). I would say it can be both.

>There is, however, a question one can ask about each attribute in an ER
>model: does the attribute truly describe the entity or relation that it has
>been attached to, or should it be attached to some different entity?

Indeed. There should in that case be a multi-valued dependency (or a functional one if it is not a multi-valued attribute) from the entity to the attribute. But why limit it to this? Why not conclude that a relationship that is not in BCNF should be split into smaller relationships?

• Jan Hidders