I need to run a stochastic model to conduct a global sensitivity analysis. For some of my parameters, I only have one value from the literature. I was wondering if anyone knows a paper and/or a rule of thumb how a problem like this could be approached. I expect the parameters which are described by this lack of information to be mostly non-influential for the final distribution of outputs. However, they need to be included stochastically to proof this and I need to justify the chosen standard deviations for the normal distributions. As I only have point estimates I was wondering if there is a rule of thumb e.g. vary it +- 25 percent to generate a range, or if there is no scientifically justifiable way of running normally-distributed inputs for which only the mean is known.
Thanks for your insights!