On the Scarcity of Observations when Modelling Random Inputs and the Quality of Solutions to Stochastic Optimisation Problems