# Minimum no. of observations required for statistical distribution fitting!

I have one conceptual (but sort of vague) question regarding distribution fitting.

How many observations one would need to best fit any statistical distribution to given data. Like I am dealing with loss data and in one situation, I had only three observations still I could fit Normal distribution, which in reality I think is absurd. One more data and the scenario may change.

Hence, is there any thumb rule for minimum number of observations in order to have reasonably stable distribution fitting?

• Such rules will only really be feasible if you fairly precisely define 'reasonably stable' - and even so it will depend on the distributions. – Glen_b Apr 10 '13 at 0:16