Running tests from coverage.py vs running coverage from test runner

During the Coverage.py with Ned Batchelder podcast, Brian and Ned briefly discussed that, if you need to run tests with coverage, it is preferred to run tests from coverage.py executing the coverage run as opposed to invoking a test runner with coverage. Why is that and what is the difference?

---

To put some context into this: currently I'm using nose test runner and execute the tests with the help of nosetests command-line tool with --with-coverage option:

$ nosetests --with-coverage --cover-html

Should I do it via the coverage run -m instead?

$ coverage run -m nose
$ coverage report