Under what circumstances would you want to denormalize a logical data model? Use examples to illustrate your answer.

What will be an ideal response?

Generally, if overall performance needs to be improved, controlled redundancy can be introduced
(see also Step 7).
Examples:
If queries on staff always require the branch address, this attribute could be posted into staff.
The effect on updating would be minimal if branch data was relatively static, and it removes a
join from the query.
If the number of staff for each branch was often required with branch details, a derived
attribute could be placed in branch. This would remove the need to access and repeatedly
count the relevant records in staff. When staff join or leave a branch, this attribute would need
to be updated.