It is better to ready books or research papers about "how to use the error in approximation". I think numerical mathematics will answer this question. Most of the time it is recommendable to use random error of 0.5, 0.05, or 0.005, and it depends on the solving method. Please read about numerical mathematics and you will get the answer rather than just using it by try and error.