A decision support system for detecting and handling biased decision-makers in multi criteria group decision-making problems

摘要

Detecting and handling biased decision-makers in the group decision-making process is overlooked in the literature. This paper aims to develop an anti-biased statistical approach, including extreme, moderate, and soft versions, as a decision support system for group decision-making (GDM) to detect and handle the bias. The extreme version starts with eliminating the biased decision-makers (DMs). For this purpose, the DMs with a lower Biasedness Index value than a predefined threshold are removed from the process. Next, it continues with a procedure to mitigate the effect of partially biased DMs by assigning different weights to DMs with respect to their biasedness level. To do so, two ratios for the remaining DMs are calculated: (i) Overlap Ratio, which shows the relative value of overlap between the confidence interval (CI) of each DM and the maximum possible overlap value. (ii) Relative confidence interval CI which reflects the relative value of CI for each DM compared to the confidence interval CI of all DMs. The final step is assigning weight to each DM, considering the two values Overlap Ratio and Relative confidence interval. DMs with closer opinions to the aggregated opinion of all DMs, or those with an adequate level of discrimination in their judgments gain more weight. The framework addresses and prescribes possible actions for all possible cases in GDM including without outliers, cases with partial outliers, and extreme cases with complete disagreement among DMs, or when none of the DMs show an adequate level of discrimination power. The moderate version preassigns a minimum weight to all unbiased DMs and then follows the weighting step for the remaining total weight. However, the soft version follows the preassignmnet of weights to all DMs in the initial pool, meaning there is no elimination in this setting. The proposed approach is tested for several scenarios with different sizes. Four performance measures are introduced to evaluate the effectiveness of the proposed method. The resulted performance measures show the reliability of the outcomes.