Conclusion

The FACTORFINDER-Procedure assigns a relevance value to each factor. Moreover, the variants are classified into showing ideal or non-ideal values.

As data base, a record of all variants that is as complete as possible has to be compiled. Pairwise all successful variants are compared. Variants with inacceptable success do not enter the evaluation process directly. Instead, they are used for supporting plausibility checks.

The procedure evaluates how tolerable non-ideal values appear from the perspective of the successful variants. This evaluation would be simple if it were certain what values are ideal and what values act inhibiting on successful variants. This is not known to the analyst. The only sure fact is that all possibly non-ideal values of a successful variant are not important enough to let that alternative fail. From the assumption that all acceptable variants are included in the example base is derived, that at least one other example differs only in subcritical factors from a successful variant. By that, the other successful alternatives serve as reference instead of the unknown ideal examples. These comparisons are the basis for inferences on the relevance.

The evaluations of the tolerance are cyclic repeated several times: The procedure assigns values to the factors that are generated evolutionary. These values represent the tolerability of non-ideal values in each factor. After many re-evaluations, there is a hypothesis on the tolerability of non-ideal values assigned to each acceptable variant. This evaluation is generated from the perspective of a specific example, the candidate.

For each tolerance option exactly one counter value exists. This values expresses how much a tolerance hypothesis is confirmed by former evaluation cycles. At the same time, the counter values save information on the validity of competing tolerance hypothesis. The FACTORFINDER-Procedure develops a characteristic distribution of counter values assigned to each candidate example. Beginning from tolerance option 0 up to a certain maximum value, the tolerance options are represented by high counter values. Tolerance options larger than that specific value are represented only by small counter values. This distribution of counter values can be characterized by marking the border between the high and the low counter values. The counter values of all tolerance options of all factors form a counter matrix. It is a memory and a buffer for competing hypothesis. The counter matrix maps the tolerability of non-ideal values in each factor from the perspective of the currently analysed example.

As soon as multiple tests of various tolerance hypothesis have generated a stable distribution within the counter matrix, a clear border can be seen between those tolerance options with high and those with low counter values. A largest tolerance option (with high counter value) marks the border in each dimension. Processing just this border condenses the information of the counter matrix. Each largest tolerance option can be complemented by a complementary value. The procedure assigns to factors that tolerate non-ideal values to a large extend (with high tolerance values) low relevance values. On the other hand, factors count as being important when non-ideal values are tolerated only little. Tolerance values close to 0 are complemented by relevance values close to 1. These complementary values are an intermediate result and the starting point for the search for variants that are ideal. These ideal examples can be compared best with all other successful examples. Non-ideal values seem to be most tolerable at those examples.

From the intermediate result the relevance is calculated. The relevance is a yardstick for evaluation of alternatives. It serves to measure the similarity of examples and to built hypothesis on the likely success of unrealized alternatives. The success of non-ideal alternatives becomes measurable and based on that the validity of the relevance hypothesis can be verified.