OBJECTIVES: Ren (2018) proposed the difference method (DM) for generating samples for PSA for pairs of parameters where the true values of the parameters have uncertainty, but their order is known with certainty. They addressed some limitations of earlier typical sampling approaches, eliminating the problem of extreme high correlation between the two parameters in the generated sample, or fully independent sampling. However, DM is still not capable of targeting a pre-specified correlation, their resulting samples are relatively highly correlated, and the resulting marginal distributions are unknown. Our objective is to propose alternative methods that are simpler to implement and address these remaining limitations.

METHODS: We generate ordered samples in the [0,1] interval. We introduce a modification of the DM method in which the parameters have symmetric roles (Symmetric method, SM). We also consider the joint normal sampling with omitting the samples that violate the required constraints (Naïve normal method, NNM). For all three approaches we introduce a modification where we calibrate key parameters of the sample generation process to match targeted sample statistics of mean, standard deviation, and correlation. We assess the applicability, quality of fit, and characteristics of the resulting distributions of each method in different scenarios.

RESULTS: The modified DM and SM methods are applicable in most scenarios and produce similar fits to the targeted correlation. The NNM is applicable in many realistic real-world scenarios when the distribution restrictions are less binding.

CONCLUSIONS: The new methods’ key benefit over the original DM is their simplicity and their ability to generate samples with pre-specified, potentially low correlation. The SM may be preferred over the DM for the symmetry of the resulting distributions. When NNM is applicable, its implementation is simpler and more transparent, the resulting distributions are more natural and familiar to the HEOR modeler than those of the DM and SM methods.