Morris method: can the grid jump be adjusted?

[mjeanner]

Hi again !

I saw while reading the past issues that the Morris sample method used to have a grid_jump argument, allowing for the manual selection of the step applied for elementary effects calculation.
This argument was removed, and now the grid step value is automatically calculated using a formula recommended by Morris, i.e. n_levels/(2n_levels-2).
Whatever the number of levels, this jump value is always larger than half of the admissible range of the parameters. I would like to take a smaller step value for elementary effects calculations. Is it possible using the SALib package, and would you recommend doing such a thing?

Many thanks for your help !

Martin

[jds485]

I think choices of delta would be problem-specific, but could be justified by the resulting sampling distribution. Some of my thoughts on the sampling distribution, as of this current implementation:
delta = 0.5 is the only choice that results in a uniform distribution in expectation.
delta < 0.5 chops off the tails of the sampling distribution. In the limit to 0, a dirac delta function would result.
delta > 0.5 cuts out the middle of the distribution. In the limit to 1, a binomial distribution results.