گرایش مطلوب در پیش بینی : پیامدهایی برای کیفیت تصمیم گیری بر اساس نتایج دلفی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|1030||2011||17 صفحه PDF||سفارش دهید||11940 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Technological Forecasting and Social Change, Volume 78, Issue 9, November 2011, Pages 1654–1670
In foresight activities uncertainty is high and decision makers frequently have to rely on human judgment. Human judgment, however, is subject to numerous cognitive biases. In this paper, we study the effects of the desirability bias in foresight. We analyze data from six Delphi studies and observe that participants systematically estimate the probability of occurrence for desirable (undesirable) future projections higher (lower) than the probability for projections with neutral desirability. We also demonstrate that in the course of a multi-round Delphi process, this bias decreases but is not necessarily eliminated. Arguably, the quality of decisions based on Delphi results may be adversely affected if experts share a pronounced and common desirability for a future projection. Researchers and decision makers have to be aware of the existence and potential consequences of such a desirability bias in Delphi studies when interpreting their results and taking decisions. We propose a post-hoc procedure to identify and quantify the extent to which the desirability bias affects Delphi results. The results of this post-hoc procedure complement traditional Delphi results; they provide researchers and decision makers with information on when and to which extent results of Delphi-based foresight may be biased.
Foresight activities are essential elements of effective and efficient long-term planning. By addressing possible future developments systematically, decisions makers are able to allocate resources and make more effective (investment) decisions ,  and . When decisions about far reaching strategies and future developments have to be made, uncertainty is high and decision makers frequently have to rely on human judgment in addition to historical data. In such cases, human judgment can support in the estimation of the probability of events possibly occurring in the future . Thereby, methods which elicit experts' opinions are important instruments to improve decision quality  and . Human beings' forecasts about the likelihood of future events are likely to be biased due to cognitive limitations in the complex process of estimating probabilities, especially when uncertainty is high  and . In order to avoid such biases and to improve the quality of subjective probabilities, the application of the Delphi method has been suggested to be an appropriate procedure ,  and . Through (1) exchange of expert knowledge, (2) iteration in the survey process, (3) provision of controlled feedback, and (4) convergence of probability assessments, the adverse effects of cognitive limitations on probability assessments, such as over-confidence, can be reduced , , ,  and . However, as Rowe and Wright  indicate partially, despite Delphi's ability to improve the quality of probability estimates for future events, the method is not generally capable of eliminating the effects of a so-called ‘desirability bias’. Desirability bias occurs when the desirability of an event positively influences the judgment of whether the event is likely to occur and a perceived undesirability of the event would negatively affect the judgment . Thus, prevalence of the desirability bias in Delphi studies would lead experts to assess desirable events as more likely and undesirable events as less likely to occur. The effect of desirability bias in Delphi studies can be severe. The desirability bias can distort resulting probability assessments and thus limit the explanatory power of the Delphi forecast. In addition, an obvious consensus among experts, as well as an apparent dissent, could be induced by the desirability bias, and not just by the exchange of information within the survey process. Decision makers using Delphi in foresight have to be aware of and understand the effects of the desirability bias in order to derive appropriate conclusions and recommendations from Delphi results and to ensure reliability of the data that is used to support decision making. Our paper contributes to existing research by providing empirical evidence for the significant effect that the desirability bias has on experts' probability estimates for far-future, high-impact projections in Delphi studies. Furthermore, we show that the impact of desirability on experts' probability estimates decreases but is not necessarily eliminated during the Delphi process. Finally, we derive complementary information to traditional Delphi results by quantifying desirability's impact and thus stress the consequences of desirability bias for interpretability and quality of decisions based on Delphi results. We analyze data obtained from six individual Delphi studies in which a total of 200 experts assessed the probability of occurrence and desirability of a total of 134 far-future, high impact projections. We develop and test two important hypotheses on how the desirability of projections of the future impact estimated probabilities of occurrence and how this impact decreases in the course of a multi-round Delphi-process. To test our hypotheses, we deploy several regression models that treat estimated probability as the dependent variable and desirability as the independent variable. We use generalized methods and control for cluster effects, correlations, and variable issues that result from Delphi survey data  and . Based on the insights from these analyses we develop a post-hoc procedure that identifies and quantifies the extent to which the desirability bias affects Delphi results. The results of this post-hoc procedure complement traditional Delphi results; they provide researchers and decision makers with information on when and to which extent results of Delphi-based foresight may be biased. The remainder of this paper is organized as follows: Following the introduction, we provide a brief review of the theoretical background of major cognitive biases that may apply to Delphi studies. In this context, we also discuss the desirability bias in more detail. Based on this discussion, we derive our research hypotheses and discuss our modeling approach. Our analyses will first focus on the research hypotheses and demonstrate the significance of the desirability bias in experts' estimated probabilities. Thereupon, we highlight the practical implications of our findings and propose a post-hoc procedure to assess the consequences for Delphi-based decision making.
نتیجه گیری انگلیسی
When experts estimate the probability of occurrence for far-future projections in Delphi studies, their estimates are biased from their perceived desirability of that projections. This situation causes experts to assess the probability of desirable (undesirable) events too high (low). This effect decreases but is not necessarily eliminated throughout the Delphi process with its quantitative, qualitative, and iterative feedback. When the uncertainty of a future projection is high, desirability is likely to influence experts' probability judgments. Individuals' desirability bias in Delphi cannot be reduced if experts ignore feedback and could even be intensified through biased feedback. In such situations, if experts in a Delphi study share a pronounced common desirability for an event, final Delphi results can be distorted. This increases the complexity to interpret results appropriately and reduces quality of decisions based on Delphi results. In order to control for this, we suggest that researchers ask participants to evaluate the desirability of projections in Delphi studies. Adjacently, researchers should use a post-hoc procedure like ours to identify future projections that are likely to carry the effect of desirability in the final result. Thereupon, quality of decisions based on Delphi can be improved by quantifying the consequences of desirability bias on final results. The value to the decision-maker is generated by complementing traditional Delphi results, such as averages or measures of consensus, by information on when and to which extent results may be biased from desirability.