We make our tests as independent as we can. Each one will create it's own data or read it from Excel, but it should not be relying on a previous test. This works for data entry and calculations, so most unit test type things and some functional tests.
This can change when we get into workflow tests. In this case, the dependency is sometimes part of the test. The test might enter a bunch of things in one screen and then expect certain results in another screen. In that case, if the first one is inaccurate, I want the second one to fail.
For your example, if 3 is dependent on 2, but 2 fails, then would you still have the correct data for 3 to go on?
p.s. Can I suggest another title for your post? Something like "test dependencies" might be more helpful when others are searching the forum. :)