Planning Fallacy (Adaptive Optimism)

When Optimistic Predictions Are Adaptive

We suggest that people overweight their specific plans about a given future project and underweight more general or distributional information, leading to a strong tendency toward overly optimistic predictions. We find that planning processes are often more directly related to predictions than to actual behavior. Plans can guide behavior even when people’s timing estimates remain inaccurate. Completion times can be influenced through any of three mechanisms: planning (determining the series of steps necessary to achieve a goal), mental simulation (vividly imagining a coherent scenario or story about how the steps will unfold), and the very act of making the prediction (privately or publicly estimating a completion time). Plans are useful, even necessary, in guiding behavior but tend to be overweighted relative to other sources of information in prediction. Mental simulations may increase the overweighting of future plans, but they may also work to bring behavior more in line with people’s goals (Gollwitzer, 1999; Gollwitzer & Brandstatter, 1997; Roese, 1994; Taylor et al., 1998). Optimistic predictions themselves may also affect behavior, either through self-fulfilling prophecy effects (e.g., Johnson & Sherman, 1990; Sherman, 1980), or because the predictions serve as guiding goals and standards (Locke & Latham, 1990). Even if projects are typically completed later than predicted, they may be completed earlier than if no detailed plans or predictions were made.

We suggest that whether (or when) the planning fallacy is adaptive depends on the reason for the past failure of similar projects. If a class of projects typically runs late because of relatively controllable factors (e.g., lack of organization, attention, or motivation), then the current effort may well benefit from an optimistic prediction. The presence of an optimistic standard or goal (Heath, Larrick, & Wu, 1999; Locke & Latham, 1990) may increase motivation to at least come close to the goal, leading to an earlier outcome. However, if a class of projects typically runs late because of relatively uncontrollable factors (e.g., the mechanical failure of a needed computer or photocopier, the sudden arrival of a visiting relative or influenza epidemic, an unexpected delay because of a different project running late), then the current effort will be unlikely to benefit from an optimistic prediction.

In a series of experiments we manipulated participants’ predictions by anchoring them on more or less optimistic values and thus explored whether – or more precisely, when – optimistic predictions in themselves are functional (Buehler & Griffin, 1996). The specific anchoring manipulations varied across studies, from drawing a card to sliding a pointer along a numbered time line, but each one provided participants with an ostensibly random starting point that was either optimistic or pessimistic. Participants were then asked to adjust from this starting point to communicate their predictions for a particular project. In each study, the anchoring manipulation yielded a strong effect on predictions, allowing us to investigate whether the completion times would follow predictions. Studies also varied in the complexity of the target task. For example, in our initial study we asked participants to predict when they would complete a 1-hour computer tutorial required for their Introductory Psychology classes. This defined the simple and controllable end of the task dimension: Once participants showed up for the tutorial, it was entirely under their control whether they stayed to complete it within the hour. On the other end of the task dimension were more complex tasks such as completing income tax returns or school assignments that were vulnerable to outside influences and hence not completely under the actor’s control.

We expected that participants induced to make particularly optimistic predictions would take these predictions to heart. Using these predictions as guiding standards, they would tend to start their assignments earlier. However, according to our analysis, the link between predictions and behavior should decrease as the task becomes longer and more complex. In fact, for the simplest task (the computer tutorial), participants exposed to the optimistic anchor predicted that they would finish their assignments 5 days earlier than those exposed to the pessimistic anchor, and did finish about 5 days earlier (although both groups showed a substantial optimistic bias). For the tax return task, participants exposed to the optimistic anchor predicted that they would send in their returns almost 20 days earlier than those exposed to the pessimistic anchor, started the task some 18 days earlier, but completed the task only 5 days earlier (a non-significant difference between conditions). In other complex tasks we examined (e.g., a library research project, a class assignment), the effects of prediction anchoring on completion times averaged about zero. It seems that the beneficial effects of optimistic predictions on completion times may be limited to tasks that are self-contained, temporally compact, and isolated from outside interruptions. Although optimistic predictions may provide the impetus to get action started, their impact appears to diminish over the course of extensive, long-term projects that are vulnerable to interruptions and delays caused by external influences.

The problem of identifying conditions under which optimistic expectations to complete a project determine completion times is reminiscent of the classic social psychological problem of identifying when attitudes and associated behavioral intentions affect behavior (Ajzen, 1991; Eagley & Chaiken, 1993; Fishbein & Ajzen, 1975). Two points of contact between research on attitudes and our analysis of the planning fallacy are especially noteworthy. First, an active but still controversial question in attitude research is when and how often past behavior predicts future behavior independent of attitudes and intentions (Oulette & Wood, 1998; Sheeran, Orbell, & Trafimow, 1999). Second, both the actual and perceived controllability of the task have been identified as determinants of intention–behavior consistency (Ajzen, 1991; Ajzen & Madden, 1986; Schifter & Ajzen, 1985). Thus, for some kinds of personal projects, lessons from attitude research may help to distinguish situations in which optimistic biases are “self-correcting” – that is, when strong intentions to finish early actually lead to early completions – from situations in which past experiences are the best predictors of completion times. Large-scale organizational projects also depend on resources and decisions that a forecaster may not control. Similarly, many personal projects are subject to factors that are beyond the control of a person’s best intentions and efforts. For such projects, bias reduction techniques (“debiasing”) must target the optimistic predictions themselves.