Statistical Screening of Input Variables in a Complex Computer Code

A method is presented for “statistical screening” of input variables in a complex computer code. The object is to determine the “effective” or important input variables by estimating the relative magnitudes of their associated sensitivity coefficients. This is accomplished by performing a numerical experiment consisting of a relatively small number of computer runs with the code followed by a statistical analysis of the results. A formula for estimating the sensitivity coefficients is derived. Reference is made to an earlier work in which the method was applied to a complex reactor code with good results.