個案總覽

We interpret the past--what we've experienced and what we've been told--to chart a course for the future. It seems like a reasonable approach, but it could be a mistake. The problem is that we view the past through filters that distort reality. One filter is the business environment, which focuses on outcomes rather than the processes that lead to them and celebrates successes while ignoring failures, thus making it hard for us to learn from mistakes. Another is our circle of advisers, who may censor the information they share with us. A third filter is our own limited reasoning abilities. We tend to focus on evidence that confirms our beliefs and gloss over information that contradicts them, and we read too much into our personal experience, which inevitably involves a small sample of incidents. We can base our decisions on a clearer view of the world if we study failures and near misses--especially the processes behind them; encourage all employees to pursue preventive measures instead of just solving problems; surround ourselves with people who will speak frankly; search for evidence that our hunches are wrong, and encourage employees, data scientists, and consultants to do the same; and broaden our perspective in order to give new meaning to our varied experiences.

In an increasingly complex economic and social environment, access to vast amounts of data and information can help organizations and governments make better policies, predictions and decisions. Indeed, more and more decision makers rely on statistical findings and data-based decision models when tackling problems and forming strategies. So far, discussions of data-based decision making have centered mainly on analysis: data collection, technological infrastructures and statistical methods. Yet another vital issue receives far less scrutiny: how analytical results are communicated to decision makers. Data science, like medical diagnostics or scientific research, lies in the hands of expert analysts who must explain their findings to executive decision makers who are often less knowledgeable about formal, statistical reasoning. Yet many behavioral experiments have shown that when the same statistical information is conveyed in different ways, people make drastically different decisions. Description, the authors note, is the default mode of presenting statistical information. This typically involves a verbal statement or a written report, which might feature one or more tables summarizing the findings. But the authors'own research suggests that descriptions can mislead even the most knowledgeable decision makers. In a recent experiment, they asked 257 economics scholars to make judgments and predictions based on a simple regression analysis. To the authors'surprise, most of these experts had a hard time accurately deciphering and acting on the results of the kind of analysis they themselves frequently conduct. In particular, the authors found that their description of the findings, which mimicked the industry standard, led to an illusion of predictability -- an erroneous belief that the analyzed outcomes were more predictable than they actually were. The authors argue that simulated experience enables intuitive interpretation of statistical information,

People have trouble understanding probability, yet it's crucial to all sorts of decision making, from choosing whether to buy new equipment to putting together an investment portfolio. Graphic representations and computer simulations can present information in ways decision makers can easily grasp--of critical importance in the era of Big Data.