Morris Screening

The Morris method screens many inputs with few runs, ranking them by their average and variability of elementary effects.

Cheap global screening

When a model has many inputs and each run is expensive, a full variance-based analysis is too costly. The Morris method provides a coarse global screening at low cost: it identifies which inputs are negligible, which are important and roughly linear, and which are important with strong nonlinearity or interactions.

Elementary effects

Kronos motion — uncertainty

An elementary effect is the change in output from a one-step move in a single input along a trajectory through input space. The method samples several trajectories, each perturbing every input once, so the number of runs scales linearly with the number of inputs rather than exponentially.

The two statistics

mu*: the mean of the absolute elementary effects, measuring overall importance
sigma: the standard deviation of the elementary effects, measuring nonlinearity and interaction

Morris interpretation (mu* vs sigma)

low mu* low sigmanegligiblehigh mu* low sigmaimportant, near-linearhigh mu* high sigmaimportant, nonlinear or interacting

Reading the plot

Plotting sigma against mu* separates inputs into these regimes at a glance. Inputs near the origin can be fixed; inputs with high mu* deserve a fuller variance-based study. Inputs with high sigma warn that a linear surrogate will be inadequate for them.

Cautions

Morris screening ranks inputs but does not quantify variance contributions precisely, so it is a filter, not a final answer. Results depend on the sampling design and the chosen step size, and a poorly spread set of trajectories can mislead. Use it to cut the input set before committing expensive runs to Sobol analysis.