>I'm having a discussion with our DBA about adding an index and we seem to have different ideas about what the term 'selectivity' means.
>
>He argues that because the index will have many duplicate values, it will not be selective. (In a table with 5M rows, only 2000 rows will have a unique value for that column, the rest will be NULL.)
>
>I think it will be selective, because any query on that column will return only a single row.
>
>I am trying to speed up an update query that uses the column in the where clause.
>
>This is on a table with a lot of inserts and updates, so I understand there will be some overhead associated with index maintenance.

I'd agree with you. All the nulls will be off on their own branch of the b-tree.