From the student perspective, a numerical question looks just like a short-answer question.
The difference is that numerical answers are allowed to have an accepted error. This allows a continuous range of answers to be set.
For example, if the answer is 30 with an accepted error of 5, then any number between 25 and 35 will be accepted as correct.
Numerical questions can also have case-insensitive non-numerical answers. This is useful whenever the answer for a numerical question is something like N/A, +inf, -inf, NaN etc