If you want a statistics you need to do statistical analysis on your docking. For example, if you run a docking for 100 times under totally randomized conditions and get a particular binding mode for 70 times, you can say with 70% probability that the drug binds to that site (if it binds at all). And to get the probability of binding you need to standardize the docking with a library of known positive and negative controls; you can also obtain the rate of false positives and false negatives.
For the first part you may see this article:
Data Novel anti-inflammatory activity of epoxyazadiradione agains...
There is huge research going in CADD. Actually the reliability of docking results is very tricky. As already Uttam Pal has explained the statistical basis for selecting best binding. There are others factors too which should be kept in mind.
1) Validation
2) Manual Interpretation
3) Relevance to the real model
Validation: This is the main criteria while going for docking if your process has been correctly validated then the chances of success are high.
Manual Interpretation: Once you get the different poses you should be able to properly evaluate them. For example the ligand binding to a site will give you the docking result. What if that ligand removes from the binding site after some fractions of second. So there you see the MD etc.
Relevance to the real model: This step is really important see binding scores are just the binding of a ligand into selected site. One can change these values by introducing 'N' as they will increase the chances of hydrogen bonding. Similarly introduction of some other groups will also lead to such increase in binding free energy.
So before going for any kind of studies you need to perform such analysis. That will only improve the chances of success.