The given attribute reduction approach decides the decision performance of a reduced decision table, which can give a guidance for selecting one rule-extraction method in practical applications. The objective of this study is to compare the decision performance of positive-region reduction, Shannon entropy reduction and Liang entropy reduction. In this paper, the relationships between positive-region reduction, Shannon entropy reduction and Liang entropy reduction are first investigated. Then, by means of three evaluation indices (certainty measure, consistency measure and support measure), we systemically analyse these change mechanisms for decision performance of a decision table induced by each of these three types of reduction approaches. Finally, by numerical experiments, these change mechanisms of a decision table's decision performance are verified for the above-mentioned three attribute reductions.