In his presentation on "Data Modeling and XML: Two Ways to Structure Data," David Hay proposed that the ER model represents the semantics of an organization whereas XML is an implementation language for moving data. Unfortunately some who are promoting XML don't recognize the difference and skip the semantic modeling step altogether. Just because you have tags doesn't mean that you agree on the meaning of the data.
Showing a graphic representation of an XML schema side by side illustrated the value of the data model for visually analyzing the correctness of relationships. David pointed out that in modeling it is more important to be clear than correct.
Another consideration is that XML is fundamentally hierarchical, and non-hierarchical relationships are not easily represented in XML. As a result, a data model can be derived from the XML schema, but the hierarchical structure limits its use for storing, retrieving and validating data. Although there is some data type and format structure imposed for validation, with XML Schema, there is an assumption that the data is already semantically valid.
David's recommendation is to use data modeling first to understand the problem and meaning of the data, then determine how it will be converted to XML for data transport. He emphasized that XML is a technological design tool, not a data structure design tool.