| 其他摘要 | Decision-making strategies are cognitive processes that resolve choice problems. Based on the hypothesis of limited rationality, researchers have proposed that decision makers select appropriate strategies from a "toolbox" of strategies depending on the environment and task requirements. In previous research, strategy recognition was mainly based on the classification of strategies using decision outcomes by comparison of structured models. However, the simplistic information underlying the decision outcome of a single trial may not be sufficient enough to reveal the real underlying decision strategies. The classification based on decision outcomes demands the behavioral data be at the aggregated-trials level, consequently it will be hard to identify strategies at the granularity of trial level. This structured model comparison approach also may suffer a risk of misclassifing strategies given that the strategy "toolbox" are lack of definite content. This methodological gaps may have hinder the theoretical development to describe the decision making process and strategy selection of decision makers. Therefore, there is an urgent need for current research to develop methods that can characterize and observe the underlying cognitive processes of decision making.
Take the risky decision-making as an example, this thesis aimed to explore methods that can effectively quantify the underlying decision strategy selection process based on eye-movement process in two studies. The key assumption of this thesis is that the eye-movement search process in decision-making process can be characterized and classified by the sequential eye-tracking information. The thesis compared methods, used the best-performing method to characterize the decision-making process, and tested the cost-benefit trade-off hypothesis to explore how strategies are selected.
The eye-movement data sets from two risky choice studies (Schoemann, et al., 2019; Schulte‐Mecklenbeck, et al., 2017) were re-analyzed in Study 1. Participants in both studies were instructed to make choices following the rules of several strategies (EV, PH, or MM) or make free risky choices. In study 1, two types of eye-movement distance measures were compared: distance mesure based on non-sequential information (eg. gaze features) and distance mesure based on sequential information (scanmatch and SR). Using the supervised classification (KNN) and the unsupervised clustering (PAM), Study 1 found that: 1) sequential information featured measures recognized decision strategies more effectively than gaze attributes featured measures: the former's prediction accuracy in KNN and clustering mutual information in PAM were both higher than the later; 2) in the free choice condition, individual differences in decision strategies selection were observed among participants, and participants tended to switch their strategies in successive choices. Meanwhile, the higher the computational difficulty of the risky item, the fewer participants used the EV strategy while making choices in the corresponding item. This result supports the cost-benefit trade-off hypothesis of strategy selection.
Study 2 aimed to verify the distance measure established in study 1, and focused on the decision-strategy selection in free and successive choice without guidance strategy. The sample for study 2 was 42 college students. Participants were asked to make two-outcome risky choice following four guidance strategies (EV, PH, MM or EW) (study 2a) or without guidance strategy (study 2b). Meanwhile, behavioral and eye movement data were collected. Study 2a validated the results of the method comparison in Study 1 that sequential information featured measure Scanmatch recognized the cognitive process of decision strategy more effectively. Besides, the efficiency of Scanmatch extends to the identification of new strategies. Study 2b examined decision makers' strategy selection in sequential risky decisions in uninstructed free choice and found that decision makers' decision processes under free decision-making conditions were more similar to EV and PH strategies. However, increasing the time cost of decision makers using strategies on given items didn’t lead to adaptive strategy choice.
In conlusion, this thesis found that the distance measure based on eye-movement sequence identifies decision strategies more effectively. This conclusion was based on method comparisons on publicly available datasets and validation on experimentally collected data, compared the accuracy of supervised learning, the mutual information of cluster analysis, and visualization of data after dimensionality reduction. After the application of the Scanmatch method to eye-movement data under free decision- making conditions, this thesis also found that individuals' choices can be generally classified into multiple strategies in successive risky choices, and their decision making process were more similar to EV and PH strategy. However, no consistent evidence was found to support the adaptive strategy selection hypothesis of cost-benefit trade-off.
The thesis proposed and valiated a set of distance measures and visualization tools based on the relative positions of various types of data from the perspective of the cognitive process. The tool can be servered to describe the decision making process of decision strategies as well as free choices. These results may provide a new methodological approach to decision making research, thus may promote theoretical development in the field of decision making strategy. |
修改评论