Stochastic simulation results

Hi list,
I find the illuminance results obtained from the 5 phase method is different every time with the same model. Do we need to perform the same simulation for several times and then acquire the average results? If the difference between those results is small, is it reliable to select the results with the smallest error?Besides, the errors of horizontal sensor points that are exposed under the direct sun is larger than other points like the vertical points. Is this phenomenon reasonable? I am trying to adjust the reflectance of materials of the room in a range cause the coarse measured value is obtained with the illuminance meter.
The following is the optical characters of materials used in this model:

Any advice is appreciated!