Abstract
When test sets are generated according to a coverage criterion, it is often sufficient to fix values only for some of the inputs to achieve 100% coverage. Other inputs are immaterial and the coverage is achieved with any of their values ("don't care" values). The research question is: How do these "don't care" values (which can reach up to 20% of all input values) influence the effectiveness and other characteristics of test sets? The paper empirically investigated this question for pair-wise test sets applied for logical expressions with different sizes and complexities. Variations of the effectiveness and the Modified Condition/Decision Coverage (MC/DC) levels of pair-wise test sets were analyzed. Our results show that these variations are low and so pair-wise test sets with different "don't care" values are very stable. Any test set with randomly selected "don't care" values can be similarly used for practical testing.