Goal-Induced Risk Taking 1 Running Head: GOALS AND RISK TAKING Goal-Induced Risk Taking in Strategy Choice

We test whether specific, challenging goals increase risk taking. We propose that goals serve as reference points, creating a region of perceived losses for outcomes below a goal (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992). According to the Prospect Theory value function, decision makers become more risk seeking in the domain of losses. In three experiments we compared a “do your best” condition with a “specific, challenging goal” condition. The goal condition increased risky behavior in a skill task, financial decisions, and bargaining. The discussion considers additional implications of the reference point perspective as well as the relationship between goal-induced risk taking and innovation. Goal-Induced Risk Taking 3 Goal-Induced Risk Taking in Strategy Choice One of the most widely-documented findings in psychology is that goals improve task performance (Locke & Latham, 1990). In general, having a specific, challenging goal increases effort and persistence compared to a vague intention, such as “doing one’s best.” Although the effectiveness of goals is undisputed, some researchers have been concerned with identifying boundary conditions and unintended consequences of goal setting. For example, researchers have found evidence that difficult goals can harm performance on complex tasks (see Wood, Mento, & Locke, 1987, for a review) and can detract from performance on other tasks or task dimensions for which goals are not set (Bavelas & Lee, 1978; Polzer & Neale, 1995; Rothkopf & Billington, 1979). In the current research, we draw from research on individual decision making to propose that specific, challenging goals have another critical consequence: They make people more willing to take risks. Consider a computer programmer who is developing modules for a large software program and who sets an extremely challenging goal for completing modules this month. Assume that no tangible incentives are tied to the goal (such as promotion or bonuses)—it is simply a personal goal that she accepts. The traditional goal-setting analysis predicts that she will perform better as a result of increased effort and persistence (Locke & Latham, 1990, 1991). If she is highly committed to the goal, she will work harder while she is programming, work longer hours, and in the end, complete more modules. Our claim is that her goal also affects her willingness to take risks: A challenging goal may force her to shift away from relatively safe strategies to riskier ones where the dispersion of possible outcomes is greater. Suppose she ordinarily completes 20 modules in a month but sets a new challenging monthly target of 40. If she has “only” 30 completed modules near the end of the month, she might be willing to alter her approach. For example, she might invest time to explore a new design that would collapse many standard modules into a few restructured modules rather than pursuing the incremental strategy of completing one or two standard modules at a time. Doing this may entail two types of costs. First, it increases the variance of her outcomes—if the new strategy works it will allow her to Goal-Induced Risk Taking 4 complete a number of modules quickly, but if it fails it may produce nothing. Second, it may lower her expected value if the new strategy has a much lower probability of success than her standard approach. In the current paper, we posit that goals increase risk-seeking. In our analysis, we are concerned with the effects of “mere” goals on risk preference. By a “mere” goal, we mean a specific level of performance that has no additional payoff attached to it (Heath, Larrick, & Wu, 1999; Locke & Latham, 1990). In contrast to mere goals, many workplace goals are tied to external rewards like bonuses and promotions. If people take risks in these cases, their behavior can be easily explained by basic economic calculations. By contrast, if people respond to mere goals by becoming risk seeking, this is a fundamentally psychological phenomenon. When we use the phrase “risk-seeking,” we refer to the standard technical definition in the literature on decision making: People are risk-seeking if they prefer a gamble to a sure outcome of equal or greater expected value (e.g., if they prefer a coin flip that pays either $20 or $0 over a sure outcome of $10 or more). The idea that goals make people more risk-seeking has both theoretical and practical consequences. Risk-seeking is one of the key predictions that distinguishes a new theoretical approach to goal setting. In a recent article (Heath, Larrick, & Wu, 1999), we proposed that goals change the value of outcomes according to the principles identified in the Prospect Theory value function (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992). One of the main implications of the value function is that risk preferences change depending on whether decisions involve gains or losses: In choices between a sure gain and a risky gain, most people take the sure gain, but in choices between a sure loss and a risky loss, most people prefer to gamble. We propose that goals serve as reference points that make people feel as if they are in the domain of losses (e.g., “10 units behind a goal”) rather than the domain of gains (e.g., “20 units ahead of where I started”). By changing the frame of reference, goals change risk preference. Although the prediction of risk-seeking is a central prediction of the value function approach, other theories of goals have not systematically considered risk. Goal-Induced Risk Taking 5 The idea that goals increase risk-seeking may also shed light on some puzzling results in the traditional goal-setting literature. A small body of evidence has shown that goals lead people to make more sizeable and frequent changes in their strategies. Below, we argue that this is a form of risk-seeking and that this kind of search for new strategies can be predicted by the same theoretical mechanism we use to predict risk. From the standpoint of organizational practice, risk-taking may be associated with positive outcomes like innovation and creativity (Longswirth, 1991), as well as negative outcomes like reckless behavior (Maremont, 1995). If goals lead people to become more willing to take risks, organizations must manage the process so that they realize the advantages of innovation and creativity while avoiding the potential damage of reckless actions. In the next section, we provide a brief review of our claim that goals serve as reference points and the specific prediction that goals increase risk taking. We then consider findings in the goal-setting literature on strategy selection that are consistent with goal-induced risk taking. We finish by describing three specific experiments that provide a direct test of goal-induced risk taking. Goals as Reference Points A common theme runs through literatures concerned with goals: Goals motivate because they provide a comparison for evaluating performance (Lewin, Dembo, Festinger, & Sears, 1944; Locke & Latham, 1990). Specifically, goals transform a somewhat ambiguous stimulus— a performance level such as “27 sales so far this month”—into an outcome that has a clear valence—“better” or “worse” than the goal. Fundamentally, the effect of goals on behavior depends on a cognitive process of comparative evaluation (Locke & Latham, 1990, p. 78). Because comparative judgment is central to goal-setting effects, research on the cognitive psychology of comparative evaluation can enrich our understanding of goal-related behavior. Recently, we have proposed that a well-known theory of comparative evaluation, Prospect Theory, provides an important link between goals and motivation (Heath et al., 1999). Specifically, we suggest that goals systematically transform the valuation of outcomes consistent Goal-Induced Risk Taking 6 with Prospect Theory’s S-shaped value function shown in Figure 1. Here, we briefly review the value function before returning to its implications for goal-related behavior. The value function, v x ( ) , shows how tangible outcomes x are translated into psychological experience. Three principles govern this translation. First, the value function assumes that people judge outcomes relative to some neutral point of comparison, or reference point, and thus encode them as gains or losses. Second, the value function posits that people find losses to be more painful than comparable gains are attractive. The value function incorporates this loss aversion by depicting a curve that is steeper below the reference point in the domain of losses than above it in the domain of gains: v x v x x ( ) ( ), − > > 0 . And, third, the value function assumes that people experience diminishing sensitivity to outcomes--they are less and less sensitive to changes as they move away from the reference point (i.e., the marginal value is less and less). The value function exhibits diminishing sensitivity because the gain and loss curve both flatten as they move away from the reference point, which implies a concave function in the domain of gains and a convex function in the domain of losses. The principle of diminishing sensitivity is the most important of the three for our analysis because it is the critical determinant of risk preference. The principle of diminishing sensitivity implies that decision makers will be risk averse for choices involving gains (where the value function is concave) but risk seeking for choices involving losses (where the value function is convex). To illustrate the intuition underlying this change in preference, first consider a choice between a certain $3,000 gain and an 80% chance of a $4,000 gain (with a 20% chance of receiving nothing). In this case, most people are risk averse, favoring the sure thing over a gamble that has a higher expected value. The value function explains this preference in terms of diminishing sensitivity: Although $4,000 is more valuable than $3,000, the marginal value of the additional $1,000 is quite small and does not offset the 20% chance of $0. When choices involve losses, however, diminishing sensitivity implies a preference for risk. When people choose between a certain $3,000 loss and an 80% chance of a $4,000 loss (with a 20% chance of losing nothing), most people are risk seeking, favoring the gamble over a sure thing that has a Goal-Induced Risk Taking 7 higher expected value. Although losing $4,000 is more painful than losing $3,000, it is not much more painful; in contrast, the difference between losing $3,000 and avoiding all losses is quite attractive even if there is only a 20% chance of avoiding all losses. This shift in preference between risk aversion in gains and risk seeking in losses has been called the “reflection” effect (Kahneman & Tversky, 1979) and it has been widely documented (Davidson, Suppes, & Siegel, 1957; Fishburn & Kochenberger, 1979; Lattimore, Baker, & Witte, 1992; Payne, Laughhunn, & Crum, 1980, 1981). For moderate probabilities and values, people are risk-averse when they are in the domain of gains, and risk-seeking in the domain of losses. Our argument is that goals serve as reference points, so when people evaluate performance relative to a goal, they perceive their performance in a way predicted by the value function. Specifically, goals shift decision makers from the domain of gains to the domain of losses and change risk preference accordingly. Consider again the programmer who ordinarily completes 20 modules a month. Suppose she has finished 30 modules near the end of the month. Ordinarily, a final performance of 30 modules would be experienced as a gain. However, if she has a goal of completing 40 modules, she may perceive 30 modules as 10 “below” her goal and experience her performance as a loss relative to her goal. Furthermore, because of diminishing sensitivity, improvements far below the goal will bring little satisfaction compared to improvements near the goal. Consequently, she will be more inclined to take risks: She will see little benefit in strategies that yield small improvements, and will be more willing to risk no improvement at all to try for the large increases in value that occur near the goal. Strategy Selection and the Value Function If our argument about the value function is true, then it may help to explain some interesting results in the traditional goal-setting literature. Research has suggested that goals may affect the way that people choose and develop strategies. However, aspects of these results on strategy development are hard to explain using traditional mediating mechanisms. In their reviews of the literature, Locke and Latham (1990, 1991) listed a number of mediating mechanisms to explain why goals increase performance; the two most central are that Goal-Induced Risk Taking 8 goals increase effort and persistence. Another potential mediating mechanism listed by Locke and Latham is that goals may alter the way people develop their strategies. The evidence that goals change strategy development is somewhat difficult to interpret. Some of the evidence can be interpreted as increased effort or persistence. For example, when people are given a goal, they tend to pursue their strategies more carefully and consistently (Earley & Perry, 1987), particularly when they are provided with a specific strategy that will reliably attain the goal (Earley, Connolly, & Lee, 1989). If strategy development depends primarily on effort and persistence, then it can be subsumed under the first two mediators of Locke and Latham. However, other results on strategy development are difficult to explain in terms of effort and persistence. For example, people who have high goals make changes to their strategies that are more drastic and more frequent. Evidence that goals produce larger changes in strategies is provided by studies of a complex computer simulation (Bandura & Wood, 1989; Wood & Bandura, 1989; Wood, Bandura, & Bailey, 1990). In these studies, participants with high goals were more likely to change multiple factors of the simulation at once (i.e., to make larger changes), which made it more difficult for them to interpret feedback and led them to perform worse. Evidence that goals produce more frequent changes is provided by studies of a challenging multi-cue probability learning task; when participants had high goals, they changed their strategies more often and their strategies were less consistent as measured by the fit of a linear regression to their choices (Earley, Connolly, & Ekegren, 1989; see also Earley, Connolly, & Lee, 1989). The results of this study are particularly difficult to explain with the standard mediating mechanisms of effort or persistence. Indeed, if anything, the goal made participants less persistent. When people with high goals make larger and more frequent changes, they are accepting an element of uncertainty that has not been emphasized by previous researchers: Large, frequent changes increase both upside and downside opportunities, thereby increasing the variance in final performance. Why then, do people with high goals accept this unpredictability? Goal-Induced Risk Taking 9 We suggest that the value function provides a plausible answer. If goals serve as reference points, then people who are below their goal will see themselves as in the domain of losses. The value function in this area is convex, so people experience a lower opportunity cost if they remain far from their goal and they receive a higher upside return if they make significant progress toward it. This analysis, which depends on the property of diminishing sensitivity, is conceptually identical to the analysis of risk-seeking that we offered above. Thus, if goals lead people to become more risk-seeking, then the value function can explain some heretofore puzzling results in the goal setting literature on strategy development. In order to show that goals affect strategy development, it is important that we separate the effects that goals have on strategy choice from the effects of goals on strategy performance (e.g., effort and persistence). In the studies in this paper, we do this by (1) separating strategy choice from strategy performance (Study 1) or (2) by selecting tasks that make performance depend primarily on strategy choice, not effort and persistence (Studies 2 and 3). Goals and Risk-seeking If we can document that goals produce greater risk-seeking, then we will provide support for the idea that the Prospect Theory value function underlies goal behavior. In addition, the empirical relationship between goals and risk-seeking would be useful because many theoretical arguments have assumed such a relationship without any direct empirical evidence. A number of researchers have assumed that goals or aspirations may lead people to become more risk-taking (Schneider & Lopes, 1986; Lopes, 1987; March & Shapira, 1992) and some researchers have shown that aspirations increase risk-seeking in hypothetical settings (Payne, Laughhunn, & Crum, 1981). However, no one, to our knowledge, has manipulated goals and looked at real risk behavior. The following studies were designed to test whether goals would induce greater risk taking when actual behavior entailed actual consequences. All of the studies used the standard manipulation in the goal setting literature: Half the participants were told to “do your best” at the task; the other half were encouraged to set a “specific challenging goal.” Following Locke Goal-Induced Risk Taking 10 and Latham’s recommendation, suggested goal levels were extremely difficult, such that less than 10% of the population could achieve them (Locke & Latham, 1990, p. 349). Our hypothesis was that the standard goal-setting manipulation would lead people to take greater risks in the goal conditions than in the do your best conditions. Our assumption was that do your best is often interpreted as “gain as much as possible over your starting point.” If people in the do your best condition do treat any improvement over their starting point as a “gain,” then their behavior should be predominantly risk-averse. Even if some do your best participants do set a goal, it is likely to be vague (thereby not distinguishing sharply between gain and loss) and of moderate difficulty (thereby creating a smaller region of losses), resulting in only modest risk taking. We explored the effect of goals on risk taking in three settings: a work task (solving easy and hard anagrams for pay), a financial decision (choosing among risky gambles for actual money), and bargaining (playing an ultimatum game). In all studies, participants chose among strategies that varied in degree of risk and payoff, such that riskier strategies paid larger amounts if successful but had a lower expected value. For researchers who are interested in how mere goals affect value, the standard experimental methodology in the goal setting literature presents two problems which we attempted to overcome in our studies: external demand and spurious expectancies. External Demand In the standard goal setting study, participants are explicitly assigned a goal by the experimenter. This may create an external demand that conflicts with our focus on mere goals. Participants who are assigned a goal by the experimenter may behave differently from participants with no goal because they are trying to do what the experimenter has asked them to do. It is probably impossible to eliminate this problem while still achieving the experimental requirement of manipulating goals, but we tried to reduce the problem by not explicitly assigning goals. Instead, we merely “suggested” a goal, and we also told participants they may select any goal that they wanted as long as it was specific and challenging. Thus, we removed some of the Goal-Induced Risk Taking 11 external demand and we made it more likely that participants became committed to the goal that they personally choose (Cialdini, 1993). We also attempted to remove external demand by giving participants a reason to perform the task other than the external desires of the experimenter. Each participant in all three studies could earn money based on their actual choices and performance during the experiment. Because participants faced real monetary consequences, they should think carefully about their strategy choices, and they should be less likely to take a particular action merely because it achieves their chosen goal. Spurious Expectancies We are interested in how mere goals might alter how participants perceive the valence of an outcome (i.e., utility or value), but the standard experimental methodology may also alter how participants perceive expectancies (i.e., the probability of achieving various outcomes). According to recommendations on goal-setting methodology, experimenters should assign goals that can only be reached by about 10% of participants (Locke & Latham, p. 349). This practice may cause participants to develop spurious expectancies about the difficulty of the task. In general, people expect others to make reasonable requests during social interactions (Grice, 1975), so participants may assume that the very challenging goal set by the experimenter is actually a very reasonable goal. Indeed, although this problem has not been widely discussed in the goal-setting literature, some studies contain evidence of exactly this process—participants who are assigned high goals expect to perform better than participants who are assigned more reasonable goals (Meyer & Gellatly, 1988). In our experimental methodology, we attempted to avoid spurious expectancy effects by giving participants specific information about the difficulty of the task they were undertaking. In Study 1, we did this by giving them specific, representative examples of the task. In Study 2, we gave them explicit, specific probabilities of success for various choices. Goal-Induced Risk Taking 12 Study 1: Goals and Risk Preference in a Work Task Study 1 was designed to test whether goals would increase risk-seeking when participants chose strategies in one of the most frequently-studied tasks in the goal setting literature: solving anagrams (e.g, Shapira, 1989; Locke & Latham, 1990, p. 42). Participants were presented with the opportunity to earn money by solving anagrams that varied in difficulty and monetary value. Hard anagrams were riskier than easy anagrams because they were worth more but were also harder to solve. Because we are interested in strategy development, our main dependent variable was the riskiness of the strategy adopted by the participants. Before starting the exercise, participants had to select a mix of hard and easy anagrams that would constitute their pool of tasks for their work period. Our hypothesis was that participants with a specific, challenging goal would opt for a riskier pool of anagrams (with more hard anagrams and fewer easy ones) than participants who were told to do your best. By forcing participants to commit to a strategy up front, we could separate the effects of initial strategy selection from the subsequent performance effects of effort and persistence (Locke & Latham, 1990). We suspected that if goal participants adopted a risky strategy, they might perform worse overall even if they exerted greater effort in pursuit of their high goal during the solution period. Consistent with the methodological issues we discussed above, in this study we strove to reduce external demands by: (1) suggesting rather than assigning a goal and (2) paying people as a linear function of their performance. We also attempted to reduce spurious expectancies by giving participants specific examples of the problems they were going to be performing so they could develop accurate expectancies for the difficulty of the task. Method Participants. Participants were 171 MBA students who completed the task as part of a class exercise. They were given five minutes to read the instructions and six minutes to complete the exercise. All participants were paid based on their performance. Goal-Induced Risk Taking 13 Materials. Participants were given a two-page booklet entitled “Anagram Exercise.” The first page contained task instructions and the second page contained the actual anagram task. The first page stated: “In this exercise, you will receive money for each anagram you solve. There are two kinds of anagrams, with examples listed below. Each anagram contains the scrambled letters of an English noun. Type A anagrams have more letters than Type B anagrams and the words are more unusual. Therefore they are more difficult to solve. If you solve a Type A anagram, you will receive 40 cents. If you solve a Type B anagram, you will receive 20 cents. You will be able to solve up to 15 anagrams and you will have 6 minutes to solve them. You will choose how many Type A and B anagrams you would like to receive.” To provide participants with accurate expectancies about the difficulties of the two types of anagram, we gave them two examples of each type of anagram. To ensure that the examples adequately represented the difficulty of the actual anagrams, we made up three separate cover sheets, each of which had different examples that were randomly selected from the two pools of anagrams. For example, one cover sheet had the examples ATSDMUR and RLERAQU for the Type A anagrams, and ETOUR and LKILS for the Type B anagrams. (The complete list of anagrams is provided in the appendix.) Below, we report tests that suggest that, across the three cover pages, the examples adequately conveyed the greater difficulty of Type A than B anagrams. In the goal condition, participants were told, “We encourage you to set a specific, challenging goal for the amount of money you would like to make in this exercise. We suggest a goal between $5 and $6. You can set any goal you would like, as long as it is specific and challenging.” They were then asked “What specific, challenging goal are you setting for yourself?”, which was followed by a blank. In the do your best condition, participants were told, “We encourage you to do your best at making money in this exercise.” Next, participants were asked two questions about their abilities, “How skilled are you at doing anagrams like this?” and “How quickly can you do anagrams like these?” They marked Goal-Induced Risk Taking 14 their answers on 7-point scales with endpoints not skilled/very skilled and not quickly/very quickly, respectively. In the strategy selection phase of the experiment, participants were told they would be given 15 anagrams to solve, and they were asked to specify the number of Type A and B anagrams they desired. (For participants, we consistently used the “Type A/B” terminology to convey the choice in as neutral a manner as possible, but below, we will refer to these as hard and easy anagrams.) Note that by choosing their pool of hard and easy anagrams up front, participants were choosing a strategy. After all participants selected their strategy, they were each given a page containing the anagrams they were to solve. This page was arranged in two columns, one marked “A. $0.40 anagrams” and the other marked “B. $0.20 anagrams.” Participants were given the number of hard and easy anagrams they requested, and they had six minutes to solve them. Afterward, the results were graded and participants were paid based on their performance. Results Representativeness of examples. Recall that we used three different sets of examples to convey the difficulty of hard and easy anagrams. These examples were randomly selected from the lists, so it is not surprising that they represented the rest of the anagrams quite well. The six hard examples were solved at a rate of .17 (54/317), which was nearly identical to the other hard anagrams, .18 (70/388), χ (1, n = 705) = .06, p > .8. The six easy examples were solved at a rate of .86 (255/296), which was significantly higher than the rate at which the other easy anagrams were solved, .73 (346/475), χ (1, n = 771) = 18.02, p < .001 after correction for continuity. To the extent that the examples misrepresented the test items, it was by overstating how easy it was to solve the easy anagrams. If the examples encouraged participants to select strategies that involved more easy anagrams, then they would hurt our ability to find that goal participants would select riskier strategies (i.e., more hard anagrams). Dependent variables. Not surprisingly, participants in the do your best and goal conditions reported approximately equal skill and quickness at solving anagrams, ps > .30. Goal-Induced Risk Taking 15 Goal participants set an average goal of $4.88. (Note that this average goal is outside the range suggested by the experimenters, so this provides one piece of evidence that external demand is not an overwhelming factor.) Our major question is whether participants in the goal condition selected riskier strategies than participants in the do your best condition who did not set a goal. Because hard anagrams were worth twice as much as easy anagrams ($.40 versus $.20), participants would have earned more money by selecting hard anagrams if the hard anagrams were less than twice as difficult as the easy ones. In fact, the hard anagrams were four times harder than the easy ones, so the expected payoff of a hard anagram was much less: Choosing an easy anagram, on average, earned people about 15 cents while a hard anagram earned them only 7 cents. Thus, participants who were not risk-seeking, could, on average, maximize their payoffs by selecting a pool of only easy anagrams. As we predicted, goal participants selected strategies that were significantly more riskseeking. They selected significantly more hard anagrams (M = 8.35) than participants in the do your best condition (M = 5.45), F(1, 165) = 29.82, p < .001, and as a result, they earned less (M = $1.63 for goal versus M = $1.93 for do your best), F(1, 165) = 5.96, p < .02. Note that this difference in earnings was purely due to the fact that participants in the goal condition selected riskier strategies than participants in the do your best condition. Participants in the two conditions did not differ in the rate at which they solved hard versus easy anagrams: Both solved roughly 20% of hard anagrams (goal M = .17 vs. do your best M = .21, p > .20) and 80% of easy ones (goal M = .80 vs. do your best M = .79, p > .50). This result indicates that the effects in this study were driven purely by strategy selection. During the performance period, both conditions performed at the same level. Discussion Study 1 used a typical performance task that has been widely used in the goal literature, and it demonstrated that goals increase risk-seeking in the strategies that participants choose for this task. When faced with choosing between hard and easy anagrams that differed in Goal-Induced Risk Taking 16 profitability, participants with a specific, challenging goal chose a riskier strategy than do your best participants, and by doing so, they reduced the expected value of their final payment by a substantial amount. By adopting riskier strategies, goal subjects on average earned 18% less. We took steps in this experiment to reduce the demand characteristics that have been present in previous goal setting experiments. In our design, external demand seems not to have been a major problem. We allowed participants to set their own specific, challenging goals, and the average goal of $4.88 fell outside our suggested range of $5 to $6. Our experimental design also attempted to reduce spurious expectancies by providing participants with specific examples of the anagrams they would be solving. Although the examples represented the overall pool of anagrams, this design cannot eliminate spurious expectancies completely. If participants remained somewhat uncertain about the true probability of success for various strategies, then participants in the goal condition may have treated our “suggested” goal as information about the true difficulty of achieving high payoffs by choosing hard anagrams. If spurious expectancies are driving the effect of goals on risk in this study, then we should be able to eliminate this effect by giving people complete information about probabilities and outcomes. In the next study, we do this by using a gamble task. Study 2: Goals and Risk Preference in Gambles In the introduction of the paper, we argued that the value function may explain why people with high goals develop different strategies than people with do your best goals. The tasks that have been used in experiments on strategy development, however, do not clearly distinguish whether goal participants are performing differently or whether they are selecting strategies with different levels of risk. Study 1, by separating strategy selection from performance, indicated that goals do affect strategy selection directly. In this study, we provide even more direct evidence that goals affect strategy selection by exploring whether people are willing to choose a risk-seeking strategy in a gambling task. Because the gambling task does not involve effort or persistence, finding evidence of risk-seeking in this task would provide evidence that goals can affect strategy development directly. Goal-Induced Risk Taking 17 In addition, the probabilities of achieving different levels of performance were stated explicitly in this study. This makes it very unlikely that high goals could increase risk taking by spuriously inflating expectations. Finally, we made our test of risk seeking especially stringent by designing stimuli in which gambles with higher potential outcomes had lower expected values. Thus, participants selecting a high payoff gamble not only accepted higher variance in their outcomes, they also paid a premium in expectation to do so. Method Participants. One hundred thirty-six MBA students completed the exercise as part of a class demonstration. All students were paid based on their decision. Materials. Participants were told, "In this exercise, you will have the chance to earn money by choosing one option from a set of options, each of which offers you a specific probability of achieving a cash payoff. Options with higher potential payoffs will have somewhat lower expected value." Participants were randomly assigned to the goal or do your best condition. The participants assigned to the goal condition were told, "We encourage you to set a specific, challenging goal for the amount of money you would like to make in this exercise. You can set any goal you would like, as long as it is specific and challenging. We suggest a goal of $9." They then answered the question, "What specific, challenging goal are you setting for yourself?" The participants assigned to the do your best condition were told, "We encourage you to do your best in making money in this exercise." They were not asked to set a goal for themselves. On a second page, participants were presented with 15 gambles that increased in payoff by 50 cents, ranging from a 100% chance of winning $3 to a 21% chance of winning $10. The expected value was shown next to each gamble. Expected value decreased by approximately 5 cents with each 50 cent increase in the absolute payoff. All gambles are displayed in Figure 2. After all participants made their choice, they each drew a random number and were paid based on their choice. Goal-Induced Risk Taking 18 Results On average, goal participants set a goal of $8.26. (As in Study 1, the average goal level was below the level suggested by the experimenter.) The distribution of choices is shown in Figure 3. A test comparing the overall distributions confirmed that they were significantly different, Mann-Whitney's U = 1554, z = 2.63, p < .01 (corrected for ties). Participants selected significantly riskier gambles in the goal condition (Median = 9) than in the do your best condition (Median = 5). The expected payoff for the median participants in the do your best and goal conditions were $2.75 and 2.52 , respectively. The most notable pattern between conditions was an apparent shift from the safest option to risky options. Consider, for example, the pattern of choices for the only option that had a sure outcome (option 1, which offered $3.00 with certainty). Thirty-seven percent of do your best participants favored this option, but only 11% of goal participants (χ(1, n=130) = 10.84, p < .001). Discussion As predicted, participants who set a specific, challenging goal chose riskier gambles than participants who were trying to do their best. The median goal participant sacrificed more than 20 cents to play a gamble with greater variance in payoffs (i.e., around 10% of the expected value). It is interesting that the conditions did not differ in the frequency with which participants chose the riskiest option (option 15). This may be because some participants perceived that they were playing with “house money,” a situation that previous research has shown leads to unusual risks (Thaler & Johnson, 1990). The pattern for the other options is much clearer. When participants had no goal, more than a third of them chose the certain option (option 1); when participants had a specific, challenging goal, only 10% chose the certain option and most chose higher risk options that were more likely to approach their goal. Previous research on the relationship between goals, strategy selection, and performance has been open to alternative interpretations: Goals may affect strategy selection indirectly, by Goal-Induced Risk Taking 19 increasing effort and persistence, or directly. Because effort and persistence were irrelevant in our gambling task, Study 2 provides clear evidence that mere goals can affect strategy selection directly in the absence of other “mediating mechanisms.” Thus, change in strategy selection is conceptually and empirically distinguishable from other mediating mechanisms. Finally, the design of Study 2 effectively eliminated the possibility that risk taking was driven by spurious expectancies. Participants were presented with complete information about the probabilities and payoffs of various strategies. Moreover, they were presented with the actual expected values for each strategy. Thus, there was little potential for participants to develop spurious expectancies about the implications of choosing a riskier strategy, indicating that goals increased risk taking by changing the value of outcomes. Study 3: Goals and Risk Preference in Distributive Bargaining In the first two studies, we explored whether people would choose riskier strategies on individual tasks. In this study, we extend our tests by exploring how goals affected the strategies people adopt in a competitive social task. Specifically, we used a simple distributive bargaining exercise to test whether bargaining strategies would become riskier. Although previous researchers have not collected specific process data on strategies, several studies using repeated dyadic negotiations suggest that specific, challenging goals lead people to engage in riskier negotiation strategies. Research has shown that difficult goals sometimes improve both integrative and distributive negotiation outcomes (Huber & Neale, 1986, 1987), but they also can decrease the chances of reaching agreement and thereby lower individual profit (Neale & Bazerman, 1985). Based on these results, we can infer that negotiators may be adopting riskier tactics, such as making high demands that increase the chance of impasse. Making a high demand is a risky strategy because, although it increases one’s payoff if it succeeds, it also increases the chances of receiving nothing because of impasse (White & Neale, 1994). In previous studies that have explored the impact of goals on negotiation, researchers have used face-to-face bargaining tasks. Face-to-face interaction is desirable because it enhances Goal-Induced Risk Taking 20 the realism of the task, but it also makes it difficult to detect whether participants are using specific risky strategies. As a result, previous researchers have not directly observed risky strategies, but have inferred them based on final outcomes (e.g., agreement versus impasse, high versus low final outcome). Unfortunately, final outcomes may be influenced by factors other than the riskiness of the strategies adopted by negotiators, such as hostile interactions or biased perceptions (Morris, Larrick, & Su, 1999). In addition, two of the principal studies that have explored the effect of goals on negotiation performance have used a goal manipulation in which participants were told that it was “against company policy” to accept an outcome worse than the assigned goal (Huber & Neale, 1987) or that they “should not under any circumstances” make a deal worse than the assigned goal (Neale & Bazerman, 1985). (Huber & Neale (1986) provided incomplete information about the goal manipulation, but used the same task as Huber and Neale (1987)). These unconventional goal instructions manipulated more than the goal—they also changed external payoffs by imposing a severe organizational penalty for performance below the goal. To provide a straightforward test of how mere goals affect strategy selection in bargaining, we used a simple distributive bargaining task in which two players conduct a single, simultaneous, anonymous exchange based on strategies they choose in advance. The task is known as the ultimatum bargaining game and it has been widely studied in the economics literature (Guth, Schmittberger, & Schwarze, 1982; Kahneman, Knetsch, & Thaler, 1986; see review in Camerer & Thaler, 1995). The ultimatum game has two participants: The Proposer, who proposes a division of a sum of money (an offer), and the Responder, who decides whether to accept or reject the proposal. According to economic theory, the unique subgame perfect equilibrium requires that the Proposer offers the smallest possible amount to the Responder, and the Responder accepts it (since receiving some money is better than receiving none). In reality, the modal response for both players is consistently an equal division (Camerer & Thaler, 1995). A substantial proportion of Responders place some value on receiving a fair outcome and thus reject amounts that are unequal. Proposers typically oblige by offering an equal division; some Goal-Induced Risk Taking 21 do so because of an intrinsic desire to be fair and some do so because of a strategic desire to maximize their expected payoff by completing the transaction (Kahneman, Knetsch, & Thaler, 1986). The ultimatum game allows us to detect risky bargaining behavior directly by measuring the strategies that are stated by the two players: the amount that the Proposer keeps, and the minimum offer that the Responder demands. Previous studies have shown that it is risky to propose or demand more than an equal share. For example, in Larrick and Blount (1997), lowering an offer from $3.50 to $3.00 (of a $7.00 pool) lowered the Proposer’s expected value from $3.50 (the maximum expected value) to $3.00 (see also Kahneman, Knetsch, & Thaler, 1986). In addition, demands by Responders greater than equality were nearly certain to fail, since only about 5 percent of offers allocated more than half of the pool to the Responder. In Study 3, we predicted that participants who set a specific, challenging goal would be less satisfied with smaller amounts and would be more likely to request an unequal share. As a result, they would decrease the probability of agreement and reduce the expected value of the deal. Method Participants. Participants were 152 MBA students who took part in a class exercise. All participants were paid based on the outcome of their decisions. Materials. All participants were told, "In this exercise, you will be randomly paired with a student in another section to divide $7. You will play this exercise only once, and you will never know the identity of the other person. We will supply the $7.00." In the do your best condition, participants were encouraged to do their best to make money in the exercise. In the goal condition, participants were encouraged to set a specific, challenging goal for the amount of money they would like to make in the exercise. They were told that they could set any goal they liked, as long as it was specific and challenging. Goal-Induced Risk Taking 22 Participants were assigned to the role of Proposer or Responder. In the descriptions of the tasks, the two roles were referred to as “Player 1” and “Player 2.” Participants assigned to the Proposer role were told: You will be randomly paired with a student in another section to divide $7. You will be asked to propose a division of the $7.00 between yourself and the other person (whom we'll call Player 2), such as $X for yourself and $Y for Player 2. Then Player 2 will be asked whether he/she accepts or rejects this proposal. If Player 2 accepts the proposal, each player will get the amount you proposed. If Player 2 rejects the proposal, neither student will receive any money. How much do you propose for yourself and for Player 2? The Proposer then decided which division to propose from a long list of options (by $0.50 increments): _______ $7.00 for Player 1 and $0.00 for Player 2 _______ $6.50 for Player 1 and $0.50 for Player 2 _______ $6.00 for Player 1 and $1.00 for Player 2 . . . _______ $0.00 for Player 1 and $7.00 for Player 2 Participants in the Responder role were given identical instructions, except they were told, "The other student (whom we'll call Player 1) will propose a division of the $7.00, such as $X for Player 1 and $Y for you." The phrase Player 2 was replaced with you throughout the remainder of the instructions. Responders were presented with each possible proposal and indicated for each whether they would accept or reject it. Results Overall, goal participants set an average goal of $4.99. By chance, Proposers in the goal condition happened to set a lower goal (M = 4.74) than did Responders (M = 5.25), t(69) = -1.63, p = .11. Our prediction was that participants who set high goals would be more likely to engage in a risky strategy. Based on the research we reviewed in the introduction (Larrick & Blount, 1997), requesting more than half of the pool can be a risky strategy, particularly for Responders. Goal-Induced Risk Taking 23 Table 1 displays the proportion of participants in the two roles who requested more than half the pool. As predicted, the goal manipulation increased the tendency to ask for more than half. Although the effect was not significant for Proposers, it was substantial and significant for Responders. We expected that the riskier requests in the goal condition would reduce the average proportion of completed transactions compared to the do your best condition. To test this, we created a variable that measured the proportion of actual proposals that were acceptable to each Responder. This variable reflects the proportion of all possible transactions involving a given Responder that would end in a completed transaction. For example, imagine a Responder in the goal condition who has stated a minimum acceptable outcome of $2.50. Because 73% of the Proposers in the goal condition gave the Responder an amount equal to or greater than this, 73% of the possible transactions involving this Responder would be successfully completed. For each condition, we calculated the average proportion of Proposer offers that were accepted by the Responders. As expected, fewer offers were acceptable on average in the goal condition (M = .53, sd = .37) than in the do your best condition (M = .68, sd = .24), t(74) = 2.13, p < .05, two-tailed. This 15% difference translates into an expected payoff that was $1.05 less for goal dyads than for do your best dyads. Not surprisingly, the 15% difference was largely due to the risky requests of the Responders in the goal condition. (The burden of the $1.05 decrement, however, was borne roughly equally by Proposers and Responders.) Discussion Study 3 demonstrated that goal-setting leads to more risk-seeking in a simple distributive bargaining task. Contrary to our expectations, goals had no effect on Proposer requests. The likely reason for this null effect is that Proposers already have high aspirations in the ultimatum game so there is little room for the goal manipulation to work. For example, the average goal of $4.74 among Proposers in the goal condition was not substantially greater than the average request by Proposers in the do your best condition, $4.01. Responders, by contrast, are ordinarily focused on receiving an amount between $0 and an even split ($3.50 in this case), so Goal-Induced Risk Taking 24 there is more room for high goals to affect risk-seeking. The average goal of Responders in the goal condition, $5.25, was substantially higher than the average amount requested by Responders in the do your best condition, $2.97. Consistent with this interpretation, the goal manipulation did significantly increase the riskiness of demands among participants in the Responder role. It is important to note that high requests have no strategic value in these games. Because there is only one, anonymous, simultaneous exchange, participants cannot influence each other with their responses. In this respect, the demand of a Responder is a “reservation price”—the price at which the Responder would prefer to accept the alternative of $0. Ordinarily, Responders are quite content to receive an equal share. What is striking is that nearly a third of Responders with a specific, challenging goal preferred $0 to an equal division. By setting such high reservation prices, these players significantly reduced the number of completed transactions. As in previous research, Study 3 demonstrates that there may be dangers in setting a specific, challenging goal in a bargaining situation. Other studies have noted that goals may be dangerous in integrative bargaining situations (where partners have unequal values for different issues and they can create joint gains by making trades). For example, Huber and Neale (1987) found that when both parties had difficult goals, they were less likely to trade concessions and to reach fully integrative agreements. Similarly, Polzer and Neale (1995) found that having difficult goals on specific issues blocked integrative trades. Our study demonstrates that high goals can also lead to poor performance in distributive bargaining where the payoff is fixed and participants must merely agree on how to distribute it. General Discussion We used the Prospect Theory value function to predict that people would be significantly more risk-seeking when they set a specific, challenging goal. Consistent with this prediction, we found that goal participants were more likely to choose risky strategies than participants who were trying to do their best. And among goal participants, those who selected higher goals chose riskier strategies. These results held for three tasks performed for actual money--a skill task, a Goal-Induced Risk Taking 25 financial decision, and a bargaining exercise. By pursuing riskier strategies in these tasks, goal participants in all three studies accepted lower expected values for their final outcomes. It is interesting to note that economists have also found a link between goals and risk taking. For example, Chevalier & Ellison (1997) found that mutual fund managers below specific performance thresholds tend to take greater risks toward the end of the year. Mutual fund companies, in effect, receive a reward for achieving such thresholds (e.g., increased media exposure if a fund finishes among the top ten performers in an investment category), so such risk taking may be a rational response to environmental payoffs. The studies reported here are striking because they suggest that salient organizational goals can encourage risk-seeking even in the absence of external incentives. In the conclusion, we discuss some additional theoretical predictions of the Prospect Theory value function and then we consider the implications of risk-seeking for organizational decisions. Implications for Goal-Setting Theory: The Dynamics of Search We have argued that goals serve as reference points and they change the value of outcomes in a way predicted by the Prospect Theory value function. As a result, someone who is behind a goal will feel much better by moving toward their goal than they will feel bad in falling further behind. When a person is behind a goal, this simple asymmetry in value is likely to explain many of the previous results on search and the current results on risk-seeking. The convexity of value below a goal makes both risk-taking and experimentation more valuable than sure-fire strategies that yield small gains. In the case of risky decisions (as in the studies reported here), people will risk lower outcomes to try for higher outcomes. In the case of experimentation, people will risk some dead ends and false starts to try to find a more effective approach. At base, our analysis assumes that all of these behaviors result because goals change the perceived value of various outcomes. This argument may shed light on previous theoretical analyses of how goals and targets affect strategies. March and Simon (1958) assumed that Goal-Induced Risk Taking 26 people search for better options if they are sufficiently below their aspiration. According to them, individuals and organizations start by pursuing strategies that are readily available in their repertoires (by engaging in local search); only if they are unsatisfied with their available routines do they elaborate their pre-existing routines or invent new ones. Wood and Locke (1990) proposed a similar schema with additional psychological detail. They assumed that when individuals are behind a goal, they initially apply well-learned strategies like “work harder,” “persist longer,” or “pay more attention.” If these "universal task strategies" seem unlikely to allow them to reach the goal, individuals reassess their efforts and try to develop new strategies that are specific to the domain. Thus, previous researchers have typically agreed that people first engage in local search (through previously stored strategies) and then they expand their search if the products of local search do not suffice to reach the goal. The value function provides additional theoretical underpinnings for these accounts. Interestingly, the same asymmetry in value that leads people to be risk-seeking when they are solidly in the domain of losses may make them extremely conservative if their current strategy is sufficient to extricate them from their current “loss” position behind their goal. Because the slope of the value function is steeper just before the goal than just after (the principle of loss aversion), any strategy that is sure to reach the goal will be more attractive than an equal expected value strategy that has a probabilistic chance of exceeding the goal or falling short. This prediction of the value function is consistent with the process explanations of search which predict that people will first try to reach the goal using standard, local strategies, and then will consider riskier strategies if their old strategies are insufficient. The intuition of the value function is illustrated by the problem below which was given to 31 MBA students: Terry works in sales and has decided to do his best this month. With two days left, he has completed 26 sales. He is considering two strategies: A) He knows that if he concentrates his remaining two days on the 4 clients most ready to buy, he can close those deals. Goal-Induced Risk Taking 27 B) He can spend his time spot-calling his entire remaining base of 20 clients. On average, this strategy in the past has produced anywhere from 2 to 8 sales. Which strategy will Terry prefer? Here, Terry has no particular goal and participants thought he would be mildly risk averse: 54% said Terry would prefer the certain strategy, 46% said Terry would prefer the risky one. A second set of participants (n=33) were given a version in which the first two sentences were changed to “Terry works in sales and has set a goal of making 30 sales this month. With two days left, he has completed 26 sales.” Here, Terry can reach his goal, with certainty, by using strategy A. As predicted by the value function, the goal makes Terry much more riskaverse. Ninety-one percent of the participants believed Terry would be risk-averse and would prefer the certain strategy. (The two versions differed significantly, p < .01.) Finally, another group of participants (n = 31) was given a version where Terry’s safe strategy would leave him below his goal (mirroring the circumstance created in Studies 1 through 3). In this version, Terry had a goal of making 30 sales and had completed only 21. As predicted by the convexity of the value function in this region, 71% of participants believed Terry would prefer the risky strategy (which is significantly different from the other two conditions, p < .05). Combined, these results suggest that the value function may explain why people are exceptionally conservative when they have a strategy that will reach their goal, and risk-seeking when they do not. This function allows us to understand why people may adopt a specific dynamic pattern of behavior that has been predicted by a number of theorists. Although the value function only considers how people evaluate outcomes, its simple principles are powerful enough to predict that people will be more conservative when one strategy allows them to reach their goal with certainty and more risky when they do not know whether a given strategy will Goal-Induced Risk Taking 28 reach their goal. The value function thus gives a unified psychological account for two predictions made by previous models on other bases. Implications for Decision Theory Other theories of decision making have proposed that concepts similar to goals, such as aspirations (Lopes, 1987; March & Shapira, 1992; Schneider & Lopes, 1986), will induce risk seeking. These models tend to emphasize a motivational component that directs attention toward high or low outcomes, and an aspiration against which outcomes are compared. Although March and Shapira (1992) draw on the Prospect Theory value function in their analysis, most aspiration level research has not. Letting goals serve as reference points provides one possibility for integrating Prospect Theory and aspiration level theories. The fact that challenging goals increase risk seeking also suggests that other possible comparisons, when made salient to a decision maker (Larrick, 1993), might have a similar effect on risk preference. In support of this claim, previous research on anticipated regret has shown that people also become more risk seeking when they expect to confront a salient foregone alternative (Larrick & Boles, 1995). In these studies, the foregone alternative was made salient by leading people to expect feedback about its outcome. When participants in a simulated job negotiation expected inevitable feedback on a competing job offer, they set a higher bottom-line for their current job negotiation, negotiated more aggressively, and were more likely to reach an impasse; when participants could avoid feedback on a competing offer, they were far less risk taking, demanded less, and nearly always reached an agreement. We suspect that goals and foregone alternatives are only two of many comparisons that might induce greater risk taking. Similar effects should result from interpersonal comparisons, such as feedback on the performance of peers. If people lag behind the group average or even one salient peer, they may take greater risks even in the absence of extrinsic incentives (Loewenstein, Thompson, & Bazerman, 1989). In general, therefore, we propose that goals are one of many non-status quo comparisons that distort value according to the Prospect Theory value function and that change motivation and behavior in a systematic way. Goal-Induced Risk Taking 29 Alternative Value Functions In a recent review, we provided evidence that the three principles underlying the Prospect Theory value function describe the effects of goals on motivation (Heath et al., 1999). For example, people believe that falling short of a goal produces negative affect (consistent with goals serving as a reference point that divides outcomes into regions of gain and loss) and that the negative affect of falling short of a goal is more intense than the positive affect of exceeding a goal (consistent with loss aversion). People also believe that the motivation to perform an incremental unit of work increases as one approaches a goal and then declines after achieving it (consistent with diminishing sensitivity). The Prospect Theory value function also offers a parsimonious explanation for many well-established goal-setting effects, such as increased persistence and effort (see Heath et al., 1999), and makes several new predictions, such as increased risk taking. Together, the findings of Heath et al. (1999) and Studies 1 through 3 indicate that the Prospect Theory value function captures the relationship between performance and value in the presence of a goal. There are, however, other plausible relationships. We briefly consider two types of value functions that have been proposed in previous theoretical work on goals: Linear functions and step-utility functions. Locke and Latham (1990, p. 77-78) propose that satisfaction is a linear relationship of performance. As with the Prospect Theory value function, the linear function maps performance below the goal to feelings of dissatisfaction and performance above the goal to feelings of satisfaction (the reference point principle). Also, as with the Prospect Theory value function, a higher goal shifts the function to the right, creating larger regions of dissatisfaction and smaller regions of satisfaction. However, Locke and Latham’s linear value function includes neither loss aversion (a steeper slope below the goal) nor diminishing sensitivity (flattening of the curves away from the goal). Differences in risk preference can only arise from differences in inflections, discontinuities, or curvature in utility. Consequently, linear value functions cannot explain, among other things, goal-induced changes in risk preference. Goal-Induced Risk Taking 30 Other researchers have implied that goals create a discontinuous “jump” in utility at the goal (e.g., Bandura, 1986; Bandura & Schunk, 1981), e.g., v x x k x k ( ) , , = + > > 0 0 and v x x x ( ) , = < 0 . Unlike a linear value function, a discontinuous function could explain risk seeking (although no one in the goal-setting literature, to our knowledge, has addressed this implication). Under the step function explanation, the constant added to all levels of performance equal or higher than the goal would increase the expected utility for any risky strategy that attained the goal by pk, where p is the probability of achieving the goal. 3 Although this proposal is plausible, the problem with a step utility function is that it cannot account for prior goal results demonstrating diminishing sensitivity. In our previous research (Heath et al., 1999), people believed that the motivation to perform an incremental unit of work increases with proximity to the goal, even when the goal is unattainable. A step function cannot explain this result because there would be no increase in motivation from pk (since p would be 0). Moreover, a simple step function would not predict additional performance above a goal, unless additional assumptions about slope were made besides a discontinuity. Goal-setting studies routinely find that people do in fact continue performing once they have surpassed the goal, albeit at a lower level. The Prospect Theory value function predicts continued but sharply diminished performance above a goal. Thus, we would argue that the Prospect Theory value function is preferable to a step function on three grounds. First, the value function has been used to predict and explain behavior in wide a variety of domains, such as social psychology (Kramer, 1989), medicine (McNeil, Pauker, & Tversky, 1988), and the law (Rachlinski, 1996). Second, it requires just three principles to describe the evaluation process. And, third, it explains a wide variety of both old and new effects in the goal-setting literature. By contrast, step functions have not been widely applied in behavioral research, and, depending on the nature of the step function, would have to sacrifice either breadth or parsimony: A simple step function that included just a discontinuity would not be able to explain most goal-setting phenomena; a complex step function that made assumptions about slope, curvature, and a discontinuity would not be as parsimonious Goal-Induced Risk Taking 31 as the Prospect Theory value function. Thus, we believe that the Prospect Theory value function should be favored over a step function on the grounds of precedence, breadth, and parsimony. Organizational Implications The goal-setting literature has demonstrated many times that goals improve performance in simple tasks where people can improve their performance by exerting greater effort and persistence. However, goals may not improve performance if they make people more willing to take risks. For example, the existing literature on strategy development indicates that goals make people more willing to make sizeable and frequent changes in their strategies. Previous researchers have noted that, in doing so, performance may suffer (Early, Connolly, & Ekegren, 1989; Early, Connolly, & Lee, 1989; Hogarth, Gibbs, McKenzie, & Marquis, 1991). Earley, Connolly, & Lee (1989) summarize their paper by saying that "challenging goals may not be beneficial when effective task strategies are not readily identifiable. In such settings, goals may stimulate excessive strategy search, degrading overall performance" (p. 589). In our studies, people who set goals engaged in risky strategies despite the fact that the risky strategies reduced their cash payoffs (e.g., by over 15% in Studies 2 and 3). However, these observations should merely serve as cautions. In organizations, the value of taking risks may vary across contexts. Financial models, for example, assume risk and reward are positively correlated (Brealey & Meyers, 1996). The key task for organizations is using goals to motivate taking appropriate risk. If organizations desire employees to take greater risks, our research suggests that goals may be helpful. Ordinarily, psychological and organizational factors may conspire to make individuals avoid risk in organizations. On their own, individuals tend to be risk averse because of loss aversion (Kahneman & Tversky, 1979; Benartzi & Thaler, 1995), concavity for gains (Kahneman & Tversky, 1979), and the absence of feedback on foregone alternatives (Guthrie, 1998; Larrick & Boles, 1995). Organizations may compound these individual tendencies by basing performance evaluations on small samples of performance or on short-run outcomes. Although individual and organizational factors may lead individuals to avoid risk, organizations Goal-Induced Risk Taking 32 may be more successful when individuals take risks (Hammond, 1967; Swalm, 1966). If organizations encourage individuals to set goals, individuals may be more willing to take risks. Compared to the status quo, a challenging goal can harness loss aversion and diminishing sensitivity to make it 1) extremely satisfying to approach higher levels of performance and 2) relatively painless to fail. Debbi Fields, the founder of Mrs. Fields Cookies says that, in her start-up days, goals helped encourage greater creativity. She says, “Rather than trying to do $50,000 for the week, which to me sounded incredibly difficult, I said to the folks I worked with, ‘That’s $50 right now this hour—can we do it?’” According to Fields, this kind of challenge led her and her employees to take creative risks. “When you say, ‘Hey, we’re behind our goal, it’s raining outside, nobody’s here, what are we going to do?’, then we are creative—we go outside with umbrellas and give people samples in the rain. That’s when we get outrageous. Sometimes the impossible happens because you're willing to challenge it” (Longswirth, 1991, p. xi). In this situation, the asymmetry in value produced by a specific goal helped people to appreciate progress toward the goal and helped them overcome inhibitions that would otherwise prompt them to avoid seemingly “impossible” situations. Also, diminishing sensitivity took the sting out of the potential lower performance that may have resulted from “outrageous” behavior. On the other hand, by setting a challenging goals, an organization can make movements toward the goal seem so attractive that individuals may be tempted to engage in risky or even reckless behavior to move toward the goals. In the early 1990s, the CEO of Bausch and Lomb set extremely high double-digit growth targets for his sales managers. In return, many of them responded by engaging in risky and, in some cases, unethical behaviors: According to Business Week, these managers responded to the challenging goals by pursuing riskier strategies, “often at the expense of sound business practice or ethical behavior. They gave customers extraordinarily long payment terms, knowingly fed gray markets, and threatened to cut off distributors unless they took on huge quantities of unwanted products. Some also shipped goods before customers ordered them and booked the shipments as sales, a possible violation of recognized accounting practices.” Interestingly, when these practices came to light, Bausch and Lomb’s Board of Goal-Induced Risk Taking 33 Directors did not blame the specific individuals who chose the risky and sometimes unethical strategies, instead, it placed the blame on the person who set the stretch goals in the first place: they fired the CEO (Maremont, 1995). Thus, our argument and the data in this paper merely points out a psychological truth about value: Challenging goals make decision makers sensitive to performance near the goal but insensitive to “downsides” or “worst case scenarios.” Depending on the context and how the process is managed, challenging goals may encourage innovation and creativity or they may encourage reckless and harmful behavior. Goal-Induced Risk Taking 34 ReferencesBandura, A. (1986). Social foundations of thought and action: A social cognitivetheory. Englewood Cliffs, NJ: Prentice Hall International.Bandura, A., & Schunk, D. H. (1981). Cultivating competence, self-efficacy, andintrinsic interest through proximal self-motivation. Journal of Personality and SocialPsychology, 45, 1017-1028.Bandura, A. & Wood, R. E. (1989). Effects of perceived controllability and performancestandards on self-regulation of complex decision making. Journal of Personality and SocialPsychology, 56, 805-814.Bavelas, J. B. & Lee, E. S. (1978). Effect of goal level on performance: A trade-off ofquantity and quality. Canadian Journal of Psychology, 32, 219-40.Benartzi, S. & Thaler, R.H. (1995). Myopic loss-aversion and the equity premium puzzle.Quarterly Journal of Economics, 110, 75-92.Brealey, R. A., & Meyers, S. C. (1996). Principles of corporate finance. New York:McGraw Hill.Camerer, C., & Thaler, R. H. (1995). Ultimatums, dictators and manners. Journal ofEconomic Perspectives, 9, 209-219.Chevalier, J. & Ellison, G. (1997). Risk taking by mutual funds as a response toincentives. Journal of Political Economy, 105, 1167-1200.Cialdini, R. B. (1993). Influence: The psychology of persuasion. New York: WilliamMorrow.Davidson, D., Suppes, P., & Siegel, S. (1957). Decision making: An experimentalapproach. Stanford, CA: Stanford University Press.Earley, P. C., Connolly, T. & Ekegren, G. (1989). Goals, strategy development, and taskperformance: Some limits on the efficacy of goal setting. Journal of Applied Psychology, 74,24-33. Goal-Induced Risk Taking 35 Earley, P. C., Connolly, T. & Lee, C. (1989). Task strategy interventions in goal setting:The importance of search in strategy development. Journal of Management, 15, 589-602.Earley, P. C, Northcraft, G. B., Lee, C., & Lituchy, T. R. (1990). Impact of process andoutcome feedback on the relation of goal setting to task performance. Academy of ManagementJournal, 33, 87-105.Earley, P. C. & Perry, B. C. (1987). Work plan availability and performance: Anassessment of task strategy priming on subsequent task completion. Organizational Behaviorand Human Decision Processes, 39, 279-302.Fishburn, P. C. & Kochenberger, G. A. (1979). Two-piece Von Neumann-Morgensternutility functions. Decision Sciences, 10, 503-518.Grice, H.P. (1975). Logic and conversation. In P. Cole (ed.) Syntax and semantics: Vol.9: Pragmatics. New York, Academic Press, p. 225-242.Guth, W., Schmittberger, R., & Schwarze, B. (1982). An experimental anaysis ofultimatum bargaining. Journal of Economic Behavior and Organization, 3, 367-388.Guthrie, C. (1999). Better settle than sorry: The regret aversion theory of litigationbehavior. University of Illinois Law Review, 43-90.Hammond, J. S. (1967). Better decisions with preference theory. Harvard BusinessReview, 45, 123-141.Heath, C., Larrick, R.P., Wu, G. (1999). Goals as reference points. CognitivePsychology, 38, 79-109.Hogarth, R. M., Gibbs, B. J., McKenzie, C. R., & Marquis, M. A. (1991). Learning fromfeedback: Exactingness and incentives. Journal of Experimental Psychology: Learning,Memory, & Cognition, 17, 734-752.Huber, V. L., & Neale, M. A. (1986). Effects of cognitive heuristics and goals onnegotiator performance and subsequent goal setting. Organizational Behavior and HumanDecision Processes, 36, 342-365. Goal-Induced Risk Taking 36 Huber, V. L., & Neale, M. A. (1987). Effects of selfand competitor goals onperformance in an interdependent bargaining task. Journal of Applied Psychology, 72, 197-203.Kahneman, D. (1992). Reference points, anchors, norms, and mixed feelings.Organizational Behavior and Human Decision Processes, 51, 296-312.Kahneman, D., Knetsch, J., & Thaler, R. H. (1986). Fairness and the assumption ofeconomics. Journal of Business, 59, S285-300.Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision underrisk. Econometrica, 47, 263-291. Kramer, R. M. (1989). Windows of vulnerability or cognitive illusions? Cognitiveprocesses and the nuclear arms race. Journal of Experimental Social Psychology, 25, 79-100. Larrick, R. P. (1993). Motivational factors in decision theories: The role of self-protection. Psychological Bulletin, 113, 440-450. Larrick, R. P., & Blount, S. (1997). The claiming effect: Why players are more generousin social dilemmas than in ultimatum games. Journal of Personality and Social Psychology, 72,810-825. Larrick R. P., & Boles, T. L. (1995). Avoiding regret in decisions with feedback: Anegotiation example. Organizational Behavior and Human Decision Processes, 63, 87-97. Lattimore, P. K., Baker, J. R., & Witte, A. D. (1992) The influence of probability onrisky choice: A parametric investigation. Journal of Economic Behavior and Organization, 17,377-400.Lewin, K., Dembo, T., Festinger, L., & Sears, P. (1944). Level of aspiration. In J.M.Hunt (Ed.) Personality and the behavior disorders (Vol. 1, pp. 333-378). New York: Ronald. Locke, E. A. & Latham, G. P. (1990). A theory of goal setting and task performance.Englewood Cliffs, NJ: Prentice-Hall. Locke, E. A. & Latham, G. P. (1991). Self-regulation through goal setting.Organizational Behavior and Human Decision Processes, 50, 212-247. Goal-Induced Risk Taking 37 Loewenstein G. F., Thompson, L., & Bazerman M. H., (1989). Social utility anddecision-making in interpersonal contexts. Journal of Personality and Social Psychology, 57,426-441. Longswirth, E. K (1991). Anatomy of a start-up. New York: Inc. Magazine Publishing.Lopes, L. L. (1987). Between hope and fear: The psychology of risk. Advances inExperimental Social Psychology, 20, 255-295.March, J. G. & Shapira, Z. (1992). Variable risk preferences and the focus of attention.Psychological Review, 99, 172-183.March, J. G., & Simon, H. A. (1958). Organizations. New York: Wiley.Maremont, M. (1995). Blind ambition: How the pursuit of results got out of hand atBausch & Lomb. Businessweek, October 23, .McNeil, B. J., Pauker, S. G., & Tversky, A. (1988). On the framing of medical decisions.In D. E. Bell, H. Raiffa, & A. Tversky (Eds.), Decision making: Descriptive, normative, andprescriptive interactions (pp. 562-568). Cambridge: Cambridge Univ. Press.Meyer, J.P. & Gellatly, I.R. (1988). Perceived performance norm as a mediator in theeffect of assigned goal on personal goal and task performance. Journal of Applied Psychology,73, 410-420.Morris, M. W., Larrick, R. P., & Su, S. K. (1999). Misperceiving negotiationcounterparts: When situationally-determined bargaining behaviors are attributed to personalitytraits. Journal of Personality and Social Psychology.Neale, M. A., & Bazerman, M. H. (1985). The effect of externally set goals on reachingintegrative agreements in competitive markets. Journal of Occupational Behavior, 6, 19-32.Neale, M. A., & Bazerman, M. H. (1991). Cognition and rationality in negotiation. NewYork: Free Press. Goal-Induced Risk Taking 38 Northcraft, G. B., & Neale, M. A. (1987). Experts, amateurs, and real estate: Ananchoring-and-adjustment perspective on property pricing decisions. Organizational Behaviorand Human Decision Processes, 39, 228-241.Payne, J. W., Laughhunn, D. J., & Crum, R. L. (1980). Translation of gambles andaspiration effects in risky choice behavior. Management Science, 26, 1039-1060.Payne, J. W., Laughhunn, D. J., & Crum, R. L. (1981). Further tests of aspiration leveleffects in risky choice behavior. Management Science, 27, 953-958.Polzer, J. T., & Neale, M. A. (1995). Constraints or catalysts? Reexamining goal settingwith the context of negotiation. Human Performance, 8, 3-26.Rachlinski, J. J. (1996). Gains, losses, and the psychology of litigation. SouthernCalifornia Law Review, 70, 113-185.Rothkopf, E. Z. & Billington, M. J. (1979). Goal-guided learning from text: Inferring adescriptive processing model from inspection times and eye movements. Journal of EducationalPsychology, 71, 310-327.Schneider, S. L., & Lopes, L. L. (1986). Reflection in preferences under risk: Who andwhen may suggest why. Journal of Experimental Psychology: Human Perception andPerformance, 12, 535-548.Shapira, Z. (1989). Task choice and assigned goals as determinants of taskmotivation and performance. Organizational Behavior and Human Decision Processes, 44, 141-165.Swalm, R. O. (1966). Utility theory: Insights into risk taking. Harvard BusinessReview, 44, 123-136. Thaler, R. H., & Johnson, E. J. (1990). Gambling with the house money and trying tobreak even: The effects of prior outcomes on risky choice. Management Science, 36, 643-660. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics andbiases. Science, 185, 1124-1131. Goal-Induced Risk Taking 39 Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulativerepresentations of uncertainty. Journal of Risk and Uncertainty, 5, 297-323.White, S. B., & Neale, M. A. (1994). The role of negotiator aspirations and settlementexpectancies in bargaining outcomes. Organizational Behavior and Human Decision Processes,57, 303-317.Wood, R. E., & Bandura, A. (1989). Impact of conceptions of ability on self-regulatorymechanisms and complex decision making. Journal of Personality and Social Psychology, 56,407-415.Wood, R. E., Bandura, A., & Bailey, T. (1990). Mechanisms governing organizationalperformance in complex decision making environments. Organizational Behavior and HumanDecision Processes, 46, 181-201.Wood, R. E., & Locke, E.A. (1990). Goal setting and strategy effects on complex tasks.In L. Cummings & B. Staw (Eds.), Research in Organizational Behavior (pp. 73-109).Greenwich, CT: JAI Press.Wood, R. E., Mento, A. J., & Locke, E. A. (1987). Task complexity as a moderator ofgoal effects: A meta-analysis. Journal of Applied Psychology, 72, 416-425. Goal-Induced Risk Taking 40 AppendixThe following are the anagrams used in Study 1. (No lines or labels appeared in theoriginal materials.)A. $0.40 anagramB. $0.20 anagramsATSDMURETOUR (Example Set 1)RLERAQULKILS 1. GACEAPK1. COHRP (Example Set 2)2. DUTRANO2. SSIHP 3. ACEILMT3. HOTLC (Example Set 3)4. GORMAII4. EQUNE 5. PYTOERV5. THCAM (Other Items)6. HIRFEFS6. OOFRP7. ENOTLUI7. AGSLO8. WTNARAR8. AHESD9. OPDIESE9. BDARE10. PENRATH10. DSSEE11. SHISREE11. OTMAS12. TAGYNSM12. DMAAR13. VILRACHY13. DEIRP14. INNPUGE14. GERAC15. GETTANN15. UETSI Goal-Induced Risk Taking 41 Author NoteRichard P. Larrick, University of Chicago, Graduate School of Business. Chip Heath,Fuqua School of Business, Duke University. George Wu, University of Chicago, GraduateSchool of Business. Goal-Induced Risk Taking 42 Footnotes1 We believe another approach used to explain goal effects is a variant of the spuriousexpectancies explanation. In a series of negotiation studies, Neale and colleagues proposed thatgoals serve as “anchors” (Neale & Bazerman, 1991). As originally proposed by Tversky andKahneman (1974), anchoring is a cognitive process used to make numerical judgments, in whichpeople start with an initial estimate and then adjust. The critical result is that people’s initialestimates are easily influenced by arbitrary numeric information and their subsequentadjustments are small. For example, Northcraft and Neale (1987) found that real estate agentswere highly influenced by arbitrary “listing prices” when estimating the appraisal value and thelowest acceptable offer for a residential property. As Kahneman (1992) has argued, anchorsneed to be carefully distinguished from reference points: Anchors influence what peopleperceive to be normal, likely, or realistic, whereas reference points determine how people valueoutcomes, sharply dividing whether outcomes are perceived as gains or as losses. Althoughgoals could increase risk taking through either process, we will present evidence supporting avalue rather than expectancy interpretation.2 The same proportions can be calculated by generating every possible combination ofProposers’ offers and Responders’ minimums for each condition. With 1,599 (39 x 41) and1,295 (37 x 35) possible deals in the do your best and goal conditions, respectively, thedifference in proportions between conditions is highly significant. The t-test reported in the textuses fewer degrees of freedom and is a more conservative test.3 One type of step function approach assumes that there is an external evaluator (such as amanager) who dispenses rewards for achieving a goal (e.g., Naylor & Ilgen, 1984). Economistshave demonstrated that such external incentives do increase risk taking for those performingbelow the goal (e.g., Chevalier & Ellison, 1997). In this case, however, the goal is no longer a“mere” goal: The increase in utility, k, is a function of a discrete increase in a payoff. Since werestrict our analysis to goals unaccompanied by extrinsic incentives, we are concerned with justthe claim that goals produce a purely “psychological” jump in utility. Goal-Induced Risk Taking 43 Table 1Proportion of Participants in Each Condition Requesting More than Half of the $7 Pool (CellCounts in Parentheses)._______________________________________________________________________Instruction Condition________________________________RoleDo Your BestGoal Test Between Conditions_______________________________________________________________________Proposer46%51% χ (1, n = 76) = .05, ns(18/39)(19/37)Responder5%29% χ 2 (1, n = 76) = 6.29, p < .01(2/41)(10/35)_______________________________________________________________________Note. Chi-Square tests correct for continuity. Goal-Induced Risk Taking 44