From Max Bazerman’s “Judgement In Managerial Decision Making”
STRATEGY 6: UNDERSTAND BIASES IN OTHERS
“The nature of managerial life requires you to work closely with the decisions of others, reviewing recommendations, transforming recommendations into decisions, and adjusting decisions made by others in the past. The task of evaluating the decisions of others is fundamentally different from the task of auditing your own decisions. Nonetheless, from reading this book, you have learned that everyone’s decisions are influenced to some degree by a shared set of biases. How can you systematically detect bias in the decisions of those around you? Consider the following managerial situation:
You are the director of marketing for a retail chain that has 40 stores in 14 cities. Annual sales in these stores average between $2 million and $4 million with mean sales of $3 million. Twenty-five of the stores have opened in the last three years, and the company plans to open 30 new stores in the next four years. Because of this growth, you have hired a site location analyst to predict the sales in each potential site. Unfortunately, predicting sales in new markets is difficult, and even the best analyst faces a great deal of uncertainty. As the marketing director, you are evaluated in part by the accuracy of the forecasts coming out of your department. The site location analyst has just given you her latest forecast, $3.8 million in annual sales for a potential site. Demographic data backs up the analyst’s claim that this area should make the store one of the top producers in the chain. What is your reaction to the forecast?
At a naive level, there is reason to have confidence in the analyst’s forecast. After all, she knows more than you about the details of the data that underlie the prediction. In addition, your overview of the area also predicts that the store will do well in comparison to existing stores; this evaluation is based on matching the representativeness of this site to other existing sites. The prediction begins to lose force, however, when we consider the prediction in light of a basic but counterintuitive statistical concept: regression to the mean. In Chapter 3, we saw that the extremeness of our predictions should be moderated toward the mean by the degree of uncertainty in the prediction (Kahneman & Tversky, 1982). With this rule in mind, let’s imagine that the site location analyst is known for her extreme accuracy. In fact, her predictions are almost perfectly accurate and have a correlation of actual sales equal to 1.0. If this is true, it would be appropriate to rely on the $3.8 million prediction. Now let’s consider the case in which there is a correlation of zero between the analyst’s predictions (based on demographic data) and actual sales. If this is true, her forecast is meaningless, and the only pertinent information is that the average store has sales of $3 million. Therefore, this figure becomes your best estimate. It is most likely, in fact, that the analyst has achieved neither total success nor total failure, but an intermediate level of predictability over the course of her career. The forecast should then fall between sales of the mean store and the analyst’s estimate, becoming progressively closer to the analyst’s estimate as her ability to predict sales increases (Kahneman & Tversky, 1982).
This analysis suggests that, as the director, you will want to reduce the forecast to somewhere between $3 million and $3.8 million, depending on your assessment of the correlation between the analyst’s forecasts and actual sales. In essence, the understanding of human judgment taught by this book should help you to systematically adjust the analyst’s initial decision.
The preceding analysis offers a rough guide to adjusting the decisions of others. Kahneman and Tversky (1982) have formalized this process into a five-step procedure whose steps are outlined here, using the site location problem as an example. In reviewing each step, you should think about how you might convert this systematic training into an intuitive, natural response. This will allow you, as a manager, to recognize the existence and direction of a wide range of biases across a wide range of decisions and make adjustments accordingly.
- Select a comparison group. This first step consists of selecting the set of past observations to which the current decision or forecast is to be compared. In the site location problem, comparing the new store to the population of all company stores is an obvious group. Other comparison groups often exist. For example, you might decide that only stores that have opened in the last three years are appropriate for comparison, particularly if recent stores are closer in description to the future store than established stores. A more inclusive group allows for a larger basis of comparison, but its heterogeneity may reduce its comparability to the targeted forecast.
- Assess the distribution of the comparison group. The next step involves assessing the characteristics of the past observations to which the current decision is being compared. If the comparison group consists of all stores, we know the range and mean from the data presented. If we limit the group to recent stores, these data would need to be recalculated. In addition, we might want to get additional data about the shape of the distribution around the mean.
- Incorporate intuitive estimation. This step calls for identification of the decision or forecast of the expert. In this case, the site location analyst’s assessment, $3.8 million, is the intuitive estimate that needs to be adjusted. The next two steps attempt to improve this forecast.
- Assess the predicted results of the decision. This is the most difficult step in the corrective procedure, as it requires us to determine the correlation between the decision or forecast and the comparison group data. It may be possible to assess this correlation by comparing past estimates to actual sales. In the absence of these data, you must determine some subjective procedure for this assessment. Kahneman and Tversky (1982) discuss this process in more detail. For our purposes, the key point is that the analyst’s estimate assumes a correlation of 1.0 between her prediction and actual sales. In virtually all cases, we must adjust away from this biased estimate.
- Adjust the intuitive estimate. In this step we must calculate the adjustment that reduces the bias error of the initial decision or forecast. For example, this procedure should produce an estimate of $3.8 million when the correlation in Step 4 is 1.0, an estimate of $3 million when the correlation is zero, and estimates proportionally in between when the correlation falls between zero and one. This adjustment can be formalized as follows:
adjusted estimate = group mean + correlation (initial estimate – group mean)
In our example, it is easy to see that this leads to a prediction of $3.4 million when the correlation is 0.5, $3.6 million when the correlation is 0.75, and so on. The person making the adjustment should fully understand the logic of the procedure and evaluate its relevance to the decision at hand. When arguing for this adjustment, you must recognize that you are likely to face resistance to change.
These five steps provide a clearly delineated process for debiasing an individual’s intuition by adjusting for the regression-to-the-mean bias. The formal procedure will typically improve the forecast. More important, a manager who understands the process will become capable of intuitively assessing the degree to which an initial estimate should be regressed to the mean.
This section shows that we can use an understanding of biases to identify systematic error in the decisions of others. Adjusting for regression to the mean is simply one example of how such a technique can be systematized. When we consult with organizations, our knowledge of the various biases documented in this book allows us to identify biases across a variety of problem types. We now have a model for adjusting a wide range of biased decisions in both individual and multiparty contexts. Broadly, it involves three phases. First, we need to accurately perceive and analyze the context within which the decision is being made. Next, we need to distinguish the potential bias(es) surrounding the decision and the decision makers. Finally, we need to identify and make the appropriate logical adjustments for that decision. This judgment-improvement technique can be used to evaluate and adjust our own, as well as others’, intuitive judgments in a variety of situations.”