关键词: 风险决策, 跨期决策, 眼动追踪, 分层贝叶斯模型, 确定效应, 即刻效应

Abstract:

Risky choice (RC) and intertemporal choice (IC) are two types of common decisions that are vital to human’s everyday life. RC and IC share similarities regarding theoretical development, behavioral effects, and neural basis. One critical challenge is that, although previous studies have revealed that RC and IC involve similar cognitive processes, results are mixed regarding what the exact mechanism might be. The mainstream discounting model hypothesizes that both RC and IC follow a compensatory and alternative-based rule. However, other models suggest that RC and IC commonly involve non-compensatory and attribute-based processing. Moreover, prior studies primarily based their findings on outcome data and few have attempted to determine whether RC and IC shared a common decision process at the cognitive computational level.
To fill this gap, the present study adopts a systematic approach to disentangle the exact mechanism of RC and IC. We considered two well-studied behavioral effects, namely, certainty effect of RC and immediacy effect of IC, respectively, and compared their underlying local and holistic process characteristics by using eye-tracking technique. Besides, we employed hierarchical Bayesian modeling to assess whether alternative- or attribute-based models better fit both RC and IC. We designed a 2×2 within-subject paradigm, with the choice task (RC vs. IC) and the construct of decision options (with vs. without certain/immediate option) as factors. Thirty-three postgraduate students participated in our study. As we were particularly interested in two pairs of decision rules, i.e., compensatory/non-compensatory rules and alternative-based/attribute-based rules, we included a series of decision attributes that reflected them, based on the local and holistic process characteristics derived from eye-movement data to test our hypotheses.
Our entire set of analyses aimed to (1) determine whether the decision processes of RC and IC are similar and (2) identify the best computational model that is more suitable for both decisions. For the first aim, results show that RC and IC indeed share comparable decision processes, albeit having a few differences in other aspects. Specifically, RC and IC differ in process characteristics, such as complexity and holistic eye-movement dynamics, and IC is processed in a relatively more deliberate, deeper fashion than RC. However, they are similar in other characteristics, such as search direction, which is more relevant to making decisions. For the second aim, computational modeling of process characteristics suggests that both types of decisions are consistent with non-discounting models. In particular, results of search direction, in light of Bayesian model comparison, reveals that participants are more likely to follow the non-compensatory, attribute-based rule rather than the alternative-based/attribute-based rule when deciding for both RC and IC. Furthermore, different task constructs of decision options, i.e., with or without certain/immediate option, show distinct process characteristics, such as direction, complexity, and depth in both RC and IC.
To conclude, the present study shows that although differences exist between RC and IC, they indeed have shared cognitive mechanisms at the core of the decision processes. In both types of decisions, contrary to classic discounting models, individuals seem not to follow compensatory, attribute-based rules, which undergoes a “weighting and summing” or “delay discounting” process. Instead, they are more likely to use simple heuristic rules hypothesized by non-discounting models. Moreover, when including certain or immediate options, individuals tend to follow less compensatory and non-dominant (neither attribute-based nor alternative-based) rules. In sum, our findings not only provide a theoretical and empirical basis for the establishment of a common framework for RC and IC, but also provide a novel direction for thorough theoretical and methodological comparisons between variant decision tasks.

Key words: risky choice, intertemporal choice, eye-tracking, hierarchical Bayesian modeling, certainty effect, immediacy effect