In this talk, I will show some tried and true ways to get reasonable-looking results that are actually incorrect. Issues include choice of stochastic variables, their distributions, and the numerical implementation. Effects can include increased prediction errors, mistaken attribution of these errors, and erroneous estimation of ensemble variance. We will discuss the progress that has been made in resolving some of these issues.