Self - Prediction Based on Strength of Intention Derek

When trying to predict one’s future behavior, a natural starting point is an evaluation of one’s intentions. If self-predictions regarding some target behavior begin with an assessment of the strength of one’s intention to carry out the behavior, previous research on the psychology of judgment implies that (a) predictions will tend to be as extreme as the impression of intention strength on which they are based, even when intention strength is imperfectly correlated with behavior; and that (b) adjustment in light of other factors known to influence behavior will tend to be insufficient. Consequently, self-predictions are expected to overweight the strength of current intentions when predicting future behavior, and to underweight factors that influence the ease with which intentions are translated into action as well as instability in intention strength over time. Because intentions typically reflect one’s goals and aspirations, this analysis implies that self-predictions based on strength of intention will tend to be overly optimistic. Previous research as well as several new preliminary studies consistent with this analysis are described. Self-Prediction Based on Strength of Intention 3 People’s predictions regarding their future behavior play a critical role in everyday decision making. In deciding whether to open a self-directed retirement savings account, for example, an individual must anticipate whether he or she will indeed invest an annual lump sum as intended. Likewise, a patient’s decision to begin a course of antibiotic medication (as opposed to an alternative course of treatment) may hinge on a self-assessment of the likelihood of adhering to the medication schedule (Kaplan & Simon, 1990). Although individuals obviously have access to an enormous amount of information about themselves, self-predictions of future behavior are not always accurate (e.g., Osberg & Shrauger, 1986; Vallone, Griffin, Lin, & Ross, 1990), even for types of behavior that are largely under the individual’s control (e.g., Epley & Dunning, 2001), and poor decisions can be the result. To predict one’s future behavior, it is natural to begin with an evaluation of one’s intentions with regard to the behavior. All else being equal, the stronger one’s intention to carry out a course of action, the more likely it is that the action in fact will be carried out. Reliance on intention strength as a basis for self-prediction, then, will often yield accurate predictions. But it can also produce systematic bias. In particular, it is argued in this paper that self-predictions typically place too much weight on strength of intention, and consequently underweight other factors, unrelated to intention strength, that influence behavior. As a result, because intentions generally reflect goals and aspirations, self-predictions are often too optimistic. To motivate the analysis which follows, several examples are first offered in which reliance on intention strength as a basis for self-prediction may lead to poor decision outcomes. All of these observations are drawn from the domain of consumer behavior, where the evaluation of products and services often implicitly requires an assessment of the consumer’s likely future behavior, but examples from other domains (e.g., health decisions) are also readily generated. Many product purchase decisions are contingent on a prediction of how likely and frequently one is to use the product if purchased. It appears that consumers frequently base such predictions on good intentions that do not always translate into behavior. Exercise equipment is the classic example. According to a 1997 survey by the Sporting Goods Manufacturers Association, of the nearly 50 million American households in which home exercise equipment can be found, the equipment is not regularly used in over one-third of the households. In one Wall Street Journal survey (Dolan, 1989), nearly three-quarters of people who owned home exercise equipment reported that they did not use the equipment as much as they planned. Self-Prediction Based on Strength of Intention 4 Anecdotal evidence suggests that, despite the great expectations they produce, various kitchen gadgets and appliances also often go unused and are eventually relegated to the basement to preserve counter space. Purchasing decisions can be influenced by product features that are never used, such as a camcorders’s fade-in and fade-out controls and superimposed title capabilities (Simonson, Carmen, & O’Curry, 1994). Consumers have also been observed to overestimate their future use of services (e.g., local telephone service; internet home food delivery), and consequently prefer flat-rate billing (Kridel, Lehman, & Weisman, 1993; Nunes, 2000) even when they would be better off, given their actual level of use, paying on a per-use (i.e., measured service) basis. Another example of the consequences of erroneous self-predictions is found in mail-in product rebates. Manufacturers report that offering such rebates frequently stimulates retail sales, but it seems that a surprisingly small fraction of the rebate coupons are actually redeemed (estimates range from 5% up to 20% for small to moderate-sized rebates; Bulkeley, 1998). One possible explanation for this observation, consistent with the present analysis, is that consumers overestimate at the time of purchase the likelihood that they will complete and mail the rebate coupon before it expires (Loewenstein & Schkade, 1999). A final example concerns credit card adoption decisions. The apparent failure of competition in the credit card industry to drive down high interest rates is, arguably, puzzling in light of the number of firms competing in the market. Ausubel (1991) argues that the “stickiness” of credit card interest rates is attributable to suboptimal consumer decision making in which individuals underestimate the probability that they will carry an outstanding balance and consequently incur interest charges. As a result, they are not sufficiently sensitive to interest rates when choosing a credit card. Ausubel supports this claim with a comparison of data from two surveys. A consumer survey reveals that nearly half (47%) of surveyed consumers state that they tend to pay their credit card balance in full each month, but a survey of credit card companies indicates that nearly three-quarters of active credit card accounts incur interest charges in any given month, suggesting that there is a substantial subset of consumers who fail to pay off their outstanding balance each month despite their intentions to do so. Consistent with this analysis, a recent longitudinal study of credit card adoption and use (Yang, Livia, & Qui, 2001) identified a subset of “wishful thinkers” (consumers who indicate that they intend to pay off their outstanding balance each month but who also report carrying such balances on a regular Self-Prediction Based on Strength of Intention 5 basis) and showed that their credit card adoption decisions are driven primarily by annual fees and are largely insensitive to interest rates even though this group routinely carries outstanding balances. Such examples suggest that, in the domain of consumer decision making and presumably other domains as well, systematic biases in self-prediction of future behavior can impose substantial costs. Intention-Based Prediction For behaviors that are largely under an individual’s control, it stands to reason that whether or not the individual intends to engage in some behavior should be a primary determinant of whether or not the behavior occurs. Thus in predicting whether the individual will engage in the behavior, it is only natural to base the prediction on an evaluation of the individual’s intention to carry out the behavior. Indeed, the intuition that the stronger the intention, the more likely the behavior is generally borne out by an extensive program of research based on the theory of planned behavior (e.g., Ajzen, 1991; Ajzen & Madden, 1986). In the present analysis, intentions are assumed to be tied to specific behaviors or activities, and can vary in strength, that is, the extent to which the individual feels committed (at a given point in time) to carrying out that behavior or activity. In this sense, an intention (e.g., to exercise 3 days a week) is more specific than a personal goal (e.g., to lose weight) in that it is tied to the particular means by which the goal will be achieved. Such intentions are generally less specific than plans, which consistent of a temporal chain of situated events (e.g., going to the health club after work three days a week, with aerobic exercise on Mondays and Fridays and weight-lifting on Wednesdays) leading to the target outcome with which the intention is associated (cf. Gollwitzer & Brandstaetter, 1997). The accuracy of predictions based on strength of intention depends on the predictive validity of intention strength, that is, the correlation between intention strength and actual behavior. Common sense may dictate that one’s actions are generally determined by one’s intentions, but folk wisdom also tells us that certain actions may occur (or fail to occur) despite one’s best intentions. Researchers studying attitudes have long been frustrated by low attitude-behavior correlations (e.g., Wicker, 1969). One possible explanation for this observation is that intentions, as determined by attitudes, may not always translate readily into behavior. Indeed, while the theory of planned behavior implicates intentions as the mediator between attitudes and behavior, research conducted within this framework indicates that intentions typically account Self-Prediction Based on Strength of Intention 6 for only about 20% to 40% of variance in actual behavior (Ajzen, 1991; Gollwitzer, 1999; Ouellette & Wood, 1998; Sutton, 1998). In one study (Netemeyer & Burton, 1990), for example, self-rated intentions to participate in an election to be held one week later accounted for 35% of variance in actual election participation. In another study (Ajzen & Madden, 1986), the correlation between students’ measured intentions to regularly attend a course and the number of lectures actually attended over the next eight weeks was only .36. As a consequence of the often low predictive validity of intentions, manipulations intended to persuade people to change their behavior will often have much more pronounced effects on stated intentions than on actual behavior. Employees in one study who were shown a documentary that made a strong case for the use of seatbelts, for example, reported stronger intentions than did a control group to use their seatbelts regularly, but the two groups did not differ on follow-up measures of actual (selfreported) seatbelt use (Sutton & Hallett, 1989). The predictive validity of intention strength can vary widely depending on the behavior being predicted. For example, voters’ stated intentions regarding candidate choice are very highly correlated with their actual votes (unpublished study by Watters, 1989, cited in Ajzen, 1991). For other behaviors, however, the intention-behavior correlation is much lower. For example, self-rated intentions regarding election participation (i.e., whether the person will vote at all) are only moderately correlated with actual participation (Watters, 1989). What factors influence the predictive validity of intention strength? As illustrated in Figure 1, the validity with which intention strength (evaluated at T1, the time of prediction) can be used to predict later behavior (at T2, the time at which the opportunity arises to carry out the target behavior) may be usefully separated into two components: (1) the temporal stability of intention strength over the period elapsing between T1 and T2; and (2) the translation reliability of intentions at T2 in terms of how readily they successfully lead to action. That is, intention strength at T1 will be a valid predictor of behavior at T2 only to the extent that intention strength is stable over the period between T1 and T2 and intention strength at T2 is translated reliably into action. The predictive validity of intention strength, then, is governed by factors that influence the temporal stability and translation reliability of intentions. There are numerous situational or contextual factors, which are unrelated to intention strength, that can influence the reliability with which intentions are translated into action. These influences fall into a category that Lewin (1951) described as “channel” factors (see Ross & Self-Prediction Based on Strength of Intention 7 Nisbett, 1991), a classic example being the provision of a map to students following a persuasive appeal to participate in a tetanus inoculation program (Leventhal, Singer, & Jones, 1965). In terms of the present analysis, the persuasive appeal can be viewed as strengthening intentions to participate in the inoculation program, while the provision of a map can be viewed as influencing the ease with which those strengthened intentions are translated into behavior. The results of this study illustrate the power of channel factors in determining whether intentions are translated into action: Although all the students exposed to the persuasive appeal generally reported strong intentions to participate, students provided with a campus map indicating the location of the health center (and a prompt to consider when and by what route they would go to the center) were much more likely to actually go for their inoculation than were students not provided with a map. In a meta-analysis of factors influencing blood donation frequency, as another example, Ferguson (1996) found that the effect size of self-rated intentions to donate blood is nearly matched by that of an important channel factor, namely the waiting period to which potential donors are subject prior to the opportunity to donate blood. Intentions can be mentally elaborated to produce specific plans that are more reliably translated into behavior. Gollwitzer and Brandstaetter (1997) have shown, for example, that prompting participants to form “implementation intentions” specifying exactly when and how a goal will be achieved produces higher rates of goal achievement than does merely specifying the goal itself, which the authors attribute to the development of an associative link between the goal and the situations in which progress toward the goal might be made. In short, intentions are more or less readily translated into action depending on the extent to which the intention is elaborated in the form of a specific plan, the type of behavior in question, and the presence of channel factors that facilitate or impede that behavior. Results from several previous studies indicate that the temporal stability of intentions also moderates the relationship between intention strength and subsequent behavior (Conner, Sheeran, Norman, & Armitage, 2000; Doll & Ajzen, 1992; Sheeran, Orbell, & Trafimow, 1999), such that the more temporally stable the intention, the stronger the relationship between intention strength and behavior. Conner et al. (2000), for example, found that the likelihood of attending a health screening was more highly predictable from self-rated intentions among patients whose intentions were stable rather than unstable over an earlier 12-month period. Self-Prediction Based on Strength of Intention 8 If we accept that self-predictions of future behavior are likely to be based on an assessment of intention strength, then it follows that the expected accuracy of these predictions will depend on the predictive validity of intentions (as determined by their temporal stability and translation reliability). Previous research, however, indicates that intuitive predictions are often insensitive to such considerations of evidence quality. Kahneman and Tversky (1973) observed that people often engage in “prediction by evaluation,” in which predictions (e.g., of a teacher’s competence five years after leaving teachers college) are just as extreme as the impressions (e.g., of the teacher’s performance in a practice lesson during college) upon which they are based. Because they are insensitive to the validity or reliability of the predictor variable, the resulting predictions are non-regressive and hence too extreme. In further developing this analysis, Griffin and Tversky (1992) suggested that confidence in judgment is based primarily on the strength of the impression conveyed by the available evidence, with relatively little regard to the “weight” of that evidence, that is, its reliability or diagnostic value. Overconfidence results when strength is high and weight is low; underconfidence results when strength is moderate and weight is high. The present analysis applies the strength-weight framework developed by Griffin and Tversky (1992) to self-prediction, where strength in this case refers to intention strength and weight to the validity of intention strength as a predictor of behavior. Under such circumstances, if self-predictions are based primarily on strength of intention, we would expect predictions to be particularly overconfident when intention strength is high and predictive validity is low. In this case, because one’s intentions are typically determined by one’s goals and aspirations, overconfidence yields overly optimistic predictions in the presence of strong intentions. In light of the intuitive usefulness of strength of intention in predicting future behavior, in the present analysis it is assumed that an evaluation of intention strength serves as the starting point or anchor in the prediction process. Adjustments may then be made in light of additional factors that influence behavior to the extent that they are recognized as relevant. Such adjustments are typically insufficient (Tversky & Kahneman, 1974), however, and as a consequence factors unrelated to intention strength are anticipated to be relatively underweighted in the prediction process. In terms of the analysis outlined above, this account implies that the temporal stability of intention strength and the reliability with which intentions are translated into action are likely to be neglected or underweighted relative to their impact on the behavior being predicted. Underweighting such factors will produce an optimistic bias: People will tend Self-Prediction Based on Strength of Intention 9 to overestimate the likelihood of carrying out activities and completing tasks that they intend to complete (i.e., when intention strength is high). Optimistic Misprediction Consistent with the analysis developed above, there is considerable evidence that people’s predictions regarding their personal futures are often optimistically biased (e.g., Armor & Taylor, 1998; Kahneman & Lovallo, 1993; Kunda, 1990; Taylor & Brown, 1988; Weinstein, 1980). In a study of graduating MBA students predicting the number of job interviews and offers they would receive, for example, Hoch (1985) found that students’ predictions were generally too optimistic. Of course, the outcome of a job search can be viewed as largely out of the job candidate’s control. But self-predictions regarding clearly controllable behavior exhibit a similar optimistic bias. For example, college students predicting how many daffodils they would buy in an upcoming campus fundraiser substantially overestimated both the quantity purchased and the probability that they would buy any flowers at all (Epley & Dunning, 2000). College students also systematically overestimated their likelihood of voting in a presidential election to be held one month later (Epley & Dunning, 2001). One form of optimistic bias in self-prediction that has been studied extensively is the planning fallacy (Kahneman & Tversky, 1979), the tendency to underestimate how long it will take to complete a project. Buehler, Griffin, and Ross (1994) provided empirical evidence of the planning fallacy in their investigation of students’ predictions regarding the time it would take them to complete various academic and non-academic tasks. Students working on their undergraduate honors thesis, for example, predicted on average that they would complete the thesis in about 34 days but actually completed it in about 55 days; less than 30% of the students completed their thesis by the date they had predicted. All of these examples of overly optimistic self-predictions are consistent with the notion that such predictions tend to overweight intentions and consequently underweight other factors that influence the likelihood of actually carrying out the course of action in question. As is elaborated below, the present analysis further implies that self-predictions are likely to be highly sensitive to manipulations of factors that influence intention strength, such as incentives, and relatively insensitive to manipulations of other factors that are unrelated to intentions, such as prompts to consider base rates or to generate alternative scenarios. The present account also implies that self-predictions will tend to neglect considerations regarding the temporal instability Self-Prediction Based on Strength of Intention 10 of intention strength even when such instability arises from readily anticipated changes in intention strength over time. The presence of incentives to complete a task generally has a pronounced effect on selfpredictions. In particular, when incentives are introduced that presumably increase the strength of intentions to complete a task, people’s judgments of the likelihood of successfully completing the task become markedly more optimistic, typically beyond that justified by any actual influence on performance of the incentive in question. In a field study, for example, Buehler, Griffin, and MacDonald (1997) found that taxpayers who expected to receive an income tax refund, and were therefore more strongly motivated to file their tax return as soon as possible, estimated that they would file their return 10 days earlier on average than did taxpayers who did not expect a refund. In fact, the two groups did not differ in terms of when they actually filed their returns, which turned out to be later (i.e., closer to the filing deadline) than either group predicted. Several studies have also found that manipulating incentives for completion of laboratory tasks has a similar effect, such that incentives influence predictions to a greater extent than they do actual performance on the task (Buehler et al., 1997; Byram, 1997; Henry, 1994; Henry & Sniezek, 1993). On the assumption that financial incentives influence intentions to complete a task, these results are consistent with the present analysis. Neglect of Other Sources of Information Reliance on intention strength as a basis for self-prediction of future behavior will tend to insulate the prediction from considerations of relevant past behavior and of possible obstacles to carrying out the behavior in question, even when such considerations are made salient. Compared to students’ highly optimistic predictions of when they will complete a current course assignment, for example, their reports of how long they have taken to complete similar assignments in the past prove to be a less biased predictor of actual completion times (Buehler et al., 1994). Despite the relevance of this information, however, prompting the students to assess their completion times for related past assignments before making their predictions regarding the current assignment had no discernible influence on their predictions; only when the students were further required to indicate when the current assignment would be completed if it took as long as past assignments of the same kind, and then to generate a scenario in which this outcome occurred, did their predictions become less optimistic (Buehler et al., 1994). Self-Prediction Based on Strength of Intention 11 The apparent imperviousness of self-predictions to considerations of past behavior is consistent with Kahneman and Tversky’s (1979; also Kahneman & Lovallo, 1993) contention that intuitive predictions tend to be based largely on evidence pertaining to the case at hand and tend to neglect evidence regarding the broader class to which the case belongs. From this view, consideration of the “population base rate” of the behavior in question, that is, how frequently others engage in the behavior, would also be expected to be largely neglected in self predictions, even though this source of information could improve predictive accuracy relative to that expected from exclusive reliance on strength of intention. Indeed, Vallone et al. (1990) have shown that self-predictions regarding future behavior are particularly prone to error when they depart from the relevant population base rate, even though such predictions are frequently made with great confidence. Epley and Dunning (2000, 2001) found that, relative to self-predictions, assessments of how likely others are to buy flowers as part of a fundraiser or to vote in an upcoming election are less subject to optimistic bias. Similarly, entrepreneurs provide more realistic (i.e., less optimistically biased) estimates of other entrepreneurs’ chances of success than they do in evaluating their own chances (Cooper, Woo, & Dunkelberg, 1988). These results are consistent with the present analysis in that predictions regarding one’s own behavior, but not predictions regarding others, are expected to be dominated by an evaluation of intention strength. Self-predictions are also frequently insensitive to possible obstacles to completion of a project, which influence the reliability with which intentions translate into action, again even when the possibility of encountering the obstacles is made salient. Although predictions appear to be based on highly optimistic, best-case scenarios regarding how the project will be completed (Buehler et al., 1994), asking participants to generate worst-case scenarios in which things do not go as planned appears to have little or no effect on the magnitude of optimistic bias observed in subsequent predictions (Newby-Clark, Ross, Buehler, Koehler, & Griffin, 2000). In the presence of a strong intention to complete a project well ahead of a deadline, apparently, predictions regarding their completion are largely unaffected by the availability of scenarios in which events do not unfold as planned. The suggestion that currently-activated intentions block consideration of relevant past behavior and possible obstacles to task completion in self-predictions is supported by the finding that predictions by impartial observers are more likely than selfpredictions to incorporate information regarding past behavior (Buehler et al., 1994) and worstcase scenarios (Newby-Clark et al., 2000). Self-Prediction Based on Strength of Intention 12 According to the present analysis, self-predictions based on strength of intention are also expected to be largely insensitive to predictable changes in intention strength over time. Indeed, research on intertemporal choice suggests that people’s predictions and decisions tend to overweight the current state of critical predictor variables even when those variables are known to be temporally unstable. Loewenstein (1996; Loewenstein & Schkade, 1999) has compiled substantial evidence that people tend to underpredict the impact of being in a “hot” visceral state (e.g., hungry, sexually aroused) when they are in a “cold” state at the time of prediction, and vice versa. Consistent with this view, Read and van Leeuwen (1998) report that participants’ advance choices regarding a snack to be received a week later were heavily influenced by their current state of hunger, producing choices that were later reversed. Simonson (1990; see also Kahneman, Fredrickson, Schreiber, & Redelmeier, 1993) suggests that decisions with consequences extended over time are largely insensitive to the temporal dimension, such that, for example, people schedule a more varied set of snacks in advance (as if they were to eat them all in a single session) than they actually want when the snacks are later consumed over an extended period of time. As applied to the present analysis, this line of research suggests that immediate intentions may be overweighted in predictions even when temporal changes in intention strength (and the likelihood of corresponding behavior) are highly predictable. Indeed, being in the grip of a strong intention might profitably be viewed as a type of visceral state along the lines of those discussed by Loewenstein (1996), even if the physiological correlates are quite different. A somewhat related state, curiosity, has received similar treatment (Loewenstein, 1994; Loewenstein, Prelec, & Shatto, 1996). Recent research by Liberman and Trope (1998) on the construal of future events suggests that, for temporally distant events, people tend to adopt high-level construals that focus primarily on the desirability of those events, that is, the extent to which the events will help them achieve their goals. Feasibility considerations are more likely to be taken into account as the events become more temporally proximal. For example, when asked to give an invited talk at a conference to be held in nine months, a professor may focus largely on the desirability of giving such a talk (e.g., the prestige involved and the people she might meet), but as the conference draws near, the professor may tend to focus more on the effort and time required to prepare a talk that is appropriate for the conference. Self-Prediction Based on Strength of Intention 13 The distinction between considerations of desirability and feasibility made by Liberman and Trope (1998) in some ways parallels the distinction made in the present analysis between one’s intentions and the readiness with which such intentions translate into behavior. It should be noted, though, that intentions may very well incorporate considerations of feasibility as well as desirability. In this sense, Liberman and Trope’s work can be viewed as identifying the determinants of behavioral intentions, that is, people’s decisions to commit to a particular course of action in the future. If that is the case, then basing predictions on strength of intention might be expected to produce greater optimistic bias in predictions as the temporal distance to the target event increases; for temporally proximal events, by contrast, intentions may be more sensitive to considerations of feasibility, and consequently predictions based on those intentions may suffer less from a tendency to neglect possible obstacles. In the present research, we investigate the possibility that considerations of feasibility or difficulty of a target behavior influence predictions only to the extent that they influence intentions regarding that behavior. The present analysis can be formalized and, with some additional assumptions, can be used to develop a model of intention-based prediction. In the remainder of this paper, however, the focus is on the qualitative predictions that follow from the general analysis developed above. Preliminary Studies Several preliminary studies have been conducted in an attempt to investigate some of the implications of the present analysis. The first two studies represent an initial exploration of the relationship between self-predictions and intentions. These studies demonstrate that selfpredictions of future behavior are closely associated with intention strength, which in turn is strongly influenced by the importance of the goals satisfied by completing the behavior. Furthermore, consistent with the present analysis, considerations of the difficulty of the target behavior appear to influence predictions only to the extent that they influence intentions. The final two studies allow an assessment of the accuracy of self-predictions. Taken together, the results of these two studies indicate that, consistent with the present analysis, factors that influence intention strength have a larger impact on predictions than on actual behavior while factors unrelated to intention strength have a larger impact on behavior than on predictions. Study 1 The first study investigates the extent to which self-predictions mirror direct ratings of intention strength. Introductory psychology students were presented with a list of 12 activities, Self-Prediction Based on Strength of Intention 14 each of which was assumed to be generally desirable, although the students were expected to vary substantially in the extent to which they intended to accomplish each. The activities fell into three broad categories: health, academics, and prosocial behaviors. The full set of activities is listed in Table 1. On one page of the questionnaire, participants were instructed to rate the strength of their intention to carry out each activity on a 1 (absolutely no intention) to 9 (very strong intention) scale. On another page of the questionnaire, participants were instructed to judge the probability that they would actually carry out each activity on a probability scale running from 0% (certainly will not) to 100% (certainly will) in 10% increments. The order in which the intention ratings and probability judgments were made was counterbalanced over participants. The relative ease or difficulty with which the activities could be accomplished was manipulated between subjects, with one group (n = 377) receiving a more difficult version of the activities than the other group (n = 388). As shown in Table 1, values specifying the number of times the activity would be carried out in a fixed period of time, or the amount of time over which a sustained activity would be maintained, were varied such that for each activity, there was a logically more or less achievable version (e.g., exercise at least 2 hours per week next term vs. exercise at least 5 hours per week next term). This manipulation might be expected to influence the predictive validity of intention strength, in the sense that good intentions are likely to be less readily translated into action as the activity in question becomes more difficult to carry out. Given the pronounced difference in difficulty of activities across the two versions, it is not surprising to find that both rated intentions and judged probability tended to be lower for the difficult activities (intention M = 5.3, probability M = 49%) than for the easy activities (intention M = 6.2, probability M = 60%). In other words, the students reported having weaker intentions to engage in the more difficult activities, and consequently assigned them a lower probability. In the absence of behavioral outcome data, of course, we cannot know whether this difference is sufficiently large to capture the effect of difficulty on actual activity completion. According to the present account, we would expect the judged probability of engaging in an activity to depend largely on self-rated intentions to engage in that activity, with little sensitivity to the ease or difficulty with which intentions can be translated into action. Hierarchical regression analyses were conducted to test this hypothesis. Results from two analyses are shown Self-Prediction Based on Strength of Intention 15 in Table 2. In both analyses, target activity was entered as the first predictor, and naturally accounted for a substantial proportion of the variance in the probability judgments. Self-rated intention strength and activity difficulty were entered next, with intention entered first in one analysis and difficulty entered first in the other. Both analyses indicate that even after controlling for mean differences across target activities, strength of intention to engage in the activity remains a powerful predictor of the judged probability of actually engaging in that activity, while difficulty level accounts for very little variance in the self-predictions. As is shown in Figure 2, separate regression analyses of judged probability as a function of self-rated intention strength produce nearly identical regression lines. In other words, the judged probability of engaging in an activity depended almost exclusively on intention strength, with little or no adjustment for the difficulty level of the activity. If the difficulty level of an activity affects the likelihood of successfully translating intentions into actions (which seems plausible but is not tested in this study due to the absence of outcome data), then accurate prediction would require that the regression line associated with the difficult activities to have a less steep slope than that associated with the easy activities. The observation of nearly identical regression lines is consistent with the hypothesis that self-predictions are based on—and dominated by—an evaluation of intention strength. Study 2 It could be argued that the results of Study 1 are attributable to participants failing to distinguish the intention rating task from the probability judgment task. That is, participants might have used an evaluation of the likelihood of engaging in an activity as the basis for the intention strength ratings as well as for the probability judgments, perhaps because the distinction between the intention to carry out an activity and the likelihood of actually carrying out that activity was not made clear to them. On this account, it would not be surprising to find that the resulting probability judgments were just as extreme as the intention strength ratings, regardless of the difficulty level of the target activity. This interpretation seems somewhat implausible given the design of the questionnaire, in which participants evaluated both intention strength and judged probability for the same set of target activities: Presumably the contrast between the two types of ratings within a single questionnaire would imply that they were meant to measure different things. Nonetheless, a second questionnaire study was conducted in which this alternative interpretation does not as readily apply. Self-Prediction Based on Strength of Intention 16 In the second study, probability judgments were again elicited for target activities that varied in difficulty level (e.g., exercise at least 2 hours per week next term vs. exercise at least 5 hours per week next term), but the intention ratings concerned broader behavioral goals (e.g., to exercise regularly). Participants were asked to rate how “personally important” each listed goal or activity was to them, which was thought to be more clearly distinct from the probability of actually carrying out a particular activity associated with the goal. Intentions regarding the designated target activities would be expected to be closely related to (though conceptually distinct from) the personal importance of the goals they satisfy. Eight target activities were evaluated in Study 2, four concerning health and four concerning academics. As in the previous study, an easier and more difficult version of each activity was produced by varying the designated intensity or duration of the target activity. Although the difficulty level of the target activities varied between participants, all participants evaluated the personal importance of the same set of general goals. Participants rated the personal importance to them of each of the eight general goals on a 1 (not at all important) to 7 (extremely important) scale, and judged the probability of actually engaging in each of the eight target activities using the same probability scale employed in the previous study. The order in which the importance ratings and probability judgments were made was counterbalanced over participants. Table 3 lists the general goals and specific target activities that were evaluated in the study. Not unexpectedly, the student participants rated the academic goals as more personally important than the health goals, F(1, 435) = 520, MSE = 0.96, p < .001. An analysis of the mean judged probability assigned to the health and academic items showed a comparable domain effect, in which the academic target activities were assigned a higher probability of completion than were the health target activities, F(1, 435) = 215, MSE = 247, p < .001. The analysis also revealed an effect of difficulty level similar to that observed in the previous study, F(1, 435) = 502, MSE = 46.6, p < .001, with participants who evaluated the relatively easy target activities (n = 235) assigning them a higher probability of completion (M = 68%) than participants (n = 202) who evaluated the relatively difficult target activities (M = 58%). The difficulty level by domain interaction was not statistically significant. Figure 3 shows separate regression lines for the easy and difficult activities plotting judged probability as a function of self-rated personal importance. The personal importance ratings appear to be the primary determinant of judged probability, as were the intention strength ratings Self-Prediction Based on Strength of Intention 17 in the previous study. Consistent with the main effect of difficulty on the probability judgments, the regression line associated with the relatively easy activities falls above that associated with the difficult activities. As in the first study, though, the slopes of the two regression lines are nearly identical. Whether the additive adjustment is sufficient cannot be assessed in this study due to the absence of outcome data. If the difficulty level of the activity in fact influences the reliability with which intentions are translated into action, then it could be argued that the regression lines should not, normatively, have equal slopes. Rather, the slope of the line associated with the difficult activities should be flatter than that associated with the easy activities, reflecting the notion that increasing intention strength is less likely to yield activity completion when the activity in question is relatively more difficult to achieve. Summary of Studies 1 and 2 The results of Study 1 indicate that self-predictions closely mirror evaluations of strength of intention to carry out a particular activity. In that study, once specific intentions regarding the designated target activity are taken into account, the difficulty level of the activity exerts no influence on the relationship between self-rated intentions and predictions. In other words, the effect of activity difficulty on self-predictions appears to be mediated by its effect on intentions regarding the target activity. The results of Study 2 indicate that self-predictions regarding a particular target activity are largely determined by the personal importance of the general behavioral goals that are potentially satisfied by completion of the target activity. In both studies, the slope of the regression line relating self-predictions to intention strength ratings was unaffected by the manipulation of target activity difficulty, which is arguably consistent with the claim that self-predictions are insensitive to considerations of the predictive validity of intentions. In Study 2, however, the two regression lines did not fall directly on top of one another as in Study 1, but instead indicated an additive adjustment for activity difficulty as predicted from importance ratings of the general behavioral goals. Taken together, the results of Studies 1 and 2 are consistent with the notion that (a) selfpredictions are based on an evaluation of one’s strength of intention to engage in the target activity, apparently without regard to the predictive validity of intention strength; and (b) intentions to engage in a particular activity are influenced by considerations both of desirability (i.e., the personal importance of the broader goals satisfied by completion of the target activity) and feasibility (i.e., the difficulty of carrying out the target activity) along the lines discussed by Self-Prediction Based on Strength of Intention 18 Liberman and Trope (1998). It is possible, of course, that reliance on strength of intention could produce accurate self-predictions if the evaluation of intention strength on which they are based is sufficiently sensitive to considerations of feasibility or activity difficulty. The observation of nearly identical slopes of regression lines showing judged probability as a function of intention strength, however, might be taken to suggest that considerations of the predictive validity of intentions is not given adequate weight in self-predictions. To test this possibility, as is done in the next two studies, behavioral outcome data are required so that the accuracy of selfpredictions can be assessed. Study 3 The value of the questionnaire studies reported above is limited by the lack of any measure of activity completion, without which it is not possible to determine whether people in fact place unjustified (in light of outcome data) weight on intention strength in their predictions. In Study 3, we compared people’s predictions with their actual behavior. Undergraduates who had just completed an unrelated experiment were asked to consider volunteering to participate in a brief, web-based study that would be operational for a one-week period beginning several weeks in the future. The undergraduates were assured that their participation was completely voluntary, and that it would be left to them to decide whether or not to participate in the web study once it was up and running. They were told, however, that it would be useful to the graduate student who was conducting the web study to have a realistic estimate of how many participants she could expect, and as such they were asked to estimate the probability that they would participate in the study during the one-week window in which the study would be operational. Some of the undergraduates (n = 33) were informed that their participation in the web study was critical for the graduate student conducting the study to be able to receive her degree, as she was having difficulty finding enough participants for the study. Others (n = 35) were told that their participation would be helpful but not absolutely necessary as the graduate student had already recruited a sufficient number of participants for the study. We assumed that the former cover story (high importance condition) would induce stronger intentions to take part in the web study than would the latter (low importance condition). We also manipulated the ease with which intentions to participate in the study could be translated into action by promising (and actually sending) a reminder e-mail message to one group of undergraduates (reminded condition; n = 35) on the first day of the one-week period in Self-Prediction Based on Strength of Intention 19 which the web study was scheduled to be operational. For the remaining undergraduates (unreminded condition; n = 33), no such reminder was promised (or actually sent). According to the present analysis, predictions regarding participation in the web study were expected to be based largely on an evaluation of intention strength, with relatively little regard to factors influencing the ease with which intentions could be translated into behavior. As such, we expected that the importance manipulation would exert a more pronounced effect on selfpredictions than on actual behavior, and that the reminder manipulation would exert a less pronounced effect on self-predictions than on actual behavior. Participants were undergraduates who had just completed an unrelated experiment in exchange for extra credit in their introductory psychology course. They were informed by the experimenter that another graduate student was conducting a brief study that was originally supposed to included as part of the current experimental session, but that her study had not been ready in time. The experimenter handed them a single-page recruitment form for this other study, which he asked them to read and complete. The recruitment form described the web study (including the importance and reminder manipulations) and indicated the one-week period during which it would be operational. Potential participants were informed that participation was completely voluntary, and that although they would not be offered extra credit in their introductory psychology course, the web study was expected to take less than 5 minutes to complete. All potential participants were asked to provide their e-mail address so that, regardless of whether they chose to participate, they could be sent an informational message at the end of the term explaining the purpose and results of the study. Participants then were asked to provide an estimate of the probability that they would participate in the web study during the one-week window in which it would be operational on a scale running from 0% (certain that you will not participate) to 100% (certain that you will participate) in increments of 5%. It was emphasized that their estimates in no way entailed a commitment to participate, but would instead provide some useful information to the graduate student conducting the study. The bottom part of the recruitment form, which was to be removed and taken home by the respondent, indicated the web address (URL) for the study and when it would be running, and provided an ID number with which to log on to the web study. Participants completed the recruitment form after the experimenter had left the room, detached the bottom portion, and inserted the completed top portion into a sealed reply box. Self-Prediction Based on Strength of Intention 20 As promised, the web study was activated and the e-mail reminders sent (to the reminded group only) shortly after midnight on the first day of the designated one-week period. The reminder indicated the dates during which the study would be running and the website address. The ID number and time and date of participation was logged for each participant who in fact went to the web site during the week in question. These participants were presented with a brief questionnaire (the Tellegen and Atkinson Absorption Scale) consisting of 34 true-false items concerning the vividness of various subjective experiences, data from which were actually used as part of a graduate student research project at the University of Waterloo. At the end of the academic term, all participants were debriefed by e-mail regarding the true purpose of the study. Overall, self-predictions were much too optimistic: The mean judged probability of participating in the web study was 72%, but only 15% of the students actually completed the study several weeks later. As expected, the manipulation of the importance of participating in the study influenced self-predictions, F(1, 62) = 4.9, MSE = 480, p < .05. The judged probability of participating was higher when the graduate student conducting the study was said to be in urgent need of participants (M = 77.4%) than when she was said to have already recruited a sufficient number of participants (M = 65.4%). In terms of actual participation in the web study, the two groups did not differ significantly, F(1, 64) < 1, MSE = 1278, though there was a tendency for the high importance group to participate at a higher rate (M = 18.4%) than the low importance group (M = 11.6%). By contrast, the reminder manipulation had only a small, statistically non-significant effect on predictions, F(1, 62) < 1. Students who were promised a reminder e-mail message assigned a slightly higher probability of participating in the web study (M = 73.6%) than did students who were not promised a reminder (M = 69.3%). In terms of actual participation, students sent a reminder were more likely to participate in the web study (M = 21.3%) than were students who were not sent a reminder (M = 8.7%), an effect that appears quite pronounced but is only marginally statistically significant, F(1, 64) = 2.2, p < .15. In summary, then, the importance manipulation that was designed to influence intention strength exerted a larger effect on predictions (mean probability difference = 12.0%) than it did on actual behavior (mean difference = 6.8%); by contrast, the reminder manipulation that was designed to influence the ease with which intentions are translated into behavior exerted a larger effect on actual behavior (mean difference = 12.6%) than it did on predictions (mean difference Self-Prediction Based on Strength of Intention 21 = 4.3%). These results are consistent with the present analysis, according to which selfpredictions overweight intention strength and underweight the reliability with which intentions can be translated into behavior. Study 4 In the previous study, because of the substantial temporal delay between predictions and the target behavior, the tendency to underweight the predictive validity of intention strength could be attributable to neglect of considerations of either the temporal stability or the translation reliability of intention strength (see Figure 1). The analysis developed in the introduction suggests that the reliability with which intentions are translated into action should be underweighted even when the opportunity to engage in the activity takes place immediately after the self-predictions are elicited, such that the temporal instability of intention strength is unlikely to play a role. This possibility is tested in a fourth experiment, conducted as an undergraduate honours thesis by Holly Fraser, which involved a laboratory task to be completed immediately after self-predictions regarding task performance were elicited. Participants (N = 60) were presented with the task of correctly assembling a “Lego” brand toy model within a 30-minute time limit. The toy model consisted of a car and “launcher” that could be used to propel the car along a flat surface. The model was deemed to be correctly assembled if all the pieces were used and the resulting toy car could be successfully launched with the launcher. Pre-testing indicated that assembling the model in a 30-minute time period was challenging but possible. All participants were provided with an additional, correctlyassembled version of the model to serve as an example against which to evaluate their progress. They were also provided with the box in which the model is packaged, which provides pictures of the assembled model from different vantage points. The reliability with which intentions to complete the task could be translated into successful action was manipulated by providing half the participants with an illustrated assembly instruction booklet that was withheld from the remaining participants. Upon entering the lab room, the participant was seated at a large table upon which was placed the set of Lego pieces from which the model was to be assembled, the box in which the model was packaged, and the assembled example model. The assembly instruction booklet was also placed on the table for those participants assigned to the instructions condition. For those Self-Prediction Based on Strength of Intention 22 participants assigned to the no-instructions condition, there were no assembly instructions to be found anywhere in the lab room. Participants were instructed: We are interested in how people predict their completion times of various tasks. You will be provided with [assembly instructions,] a picture of the completed product, along with an actual completed model to help you with the assembly. Participants were then asked to judge the probability that they would correctly assemble the model within 30 minutes on scale running from 0% (certainly will not finish within 30 minutes) to 100% (certainly will finish within 30 minutes) in 10% increments. Following the probability judgment, participants were asked to rate the strength of their intention to assemble the model within the specified time limit using a 1 (not very strong) to 10 (very strong) scale. After participants completed these measures, the experimenter instructed them to begin the assembly task and started a stopwatch, saying that she would be back at the end of the 30-minute time limit. Thirty minutes later, the experimenter re-entered the room to determine if the participant had correctly assembled the model. At that point, whether or not the model was correctly assembled, participants were asked to complete some follow-up measures, including ratings of how difficult they had found the assembly task, how helpful they found the provided aids, and how much previous experience they had had with this type of assembly task. Predictions were sensitive to the presence or absence of instructions, F(1, 56) = 4.1, p < .05. Participants assigned a higher probability of correctly assembling the model when instructions were provided (M = 75%) than when they were not provided (M = 66%). This result is sensible in light of people’s general expectations that products such as Lego models are usually accompanied by assembly instructions; indeed, several participants commented on the absence of instructions in that condition. The presence or absence of instructions also had a marked effect on the actual likelihood of correctly assembling the model, F(1, 56) = 15.8, p < .01. Participants provided with instructions were much more likely to correctly assemble the model within 30 minutes (M = 53%) than were participants not provided with instructions (M = 10%). The instructions manipulation, then, influenced both predicted and actual performance. The present analysis implies that adjustment of predictions in light of factors (such as instructions) influencing the reliability with which intentions can be translated into action will typically be insufficient. As such, we would expect the instructions manipulation to have a smaller influence on predictions than on performance. To test this possibility, a simple “overconfidence” measure Self-Prediction Based on Strength of Intention 23 was computed as the difference between each participant’s predicted probability of success and his or her actual probability of success, the latter of which assumes a binary (0 or 1) value. Overall, participants’ predictions were too optimistic, with the mean predicted probability of successfully assembling the model (M = 70%) far exceeding their actual success rate (M = 32%). More importantly for current purposes, predictions were more optimistic or overconfident in the absence than in the presence of instructions, F(1, 56) = 12.5, p < .01. In other words, as implied by the present account, participants’ predictions were insufficiently sensitive to the influence of the presence or absence of instructions on their task performance. Predictions implied that the presence of instructions would produce an increase in the probability of success of 9%, while the actual increase was 43%. Table 5 shows the results of regression analyses indicating that predictions were based primarily on strength of intention, with relatively little influence of the presence or absence of instructions. Figure 4 shows the predicted and actual completion probability as a function of intention strength (categorized as high or low based on a median split in each condition), separately for participants with and without instructions. The lines representing the predictions as a function of intention strength exhibit a pattern resembling that found in the questionnaire studies: The two lines have similar slopes, with a small constant difference reflecting an apparent adjustment for the presence or absence of instructions. In discussing the questionnaire studies it was suggested that accurate prediction would require different slopes, reflecting differences in the reliability with which intentions can be translated successfully into action. In this study, we have an outcome variable that allows us to plot actual performance in the same manner to determine if a slope difference is indeed observed. As can be seen from Figure 4, the lines plotting performance probability as a function of intention strength do differ somewhat in slope, with a relatively flatter line in the no instructions condition reflecting the greater difficulty of translating intentions into action than in the instructions condition. General Discussion The results of several preliminary studies are largely consistent with the analysis developed in this paper, according to which self-predictions are based on strength of intention with relatively little regard to the predictive validity of intentions. Study 1 shows that predictions regarding future activities closely follow intention strength ratings for those activities. Study 2 indicates that self-predictions regarding a target activity are highly correlated with ratings of the Self-Prediction Based on Strength of Intention 24 personal importance of the goal achieved by carrying out the target activity. Taken together, the two questionnaire studies suggest that the influence of an activity’s difficulty or feasibility on predictions is mediated by its effect on intentions to carry out the activity. That is, the primary impact of increasing the difficulty level of a designated activity is to decrease the strength of intentions to carry out that activity, which in turn decreases the judged probability of completing the activity. The possibility that increased difficulty might also compromise the predictive validity of intentions does not appear to play a role in intuitive predictions. Studies 3 and 4 included behavioral outcome measures that allow us to examine factors influencing predictive accuracy. In both studies, self-predictions tended to be too optimistic, as would be expected if they are largely based on the strength of currently-activated intentions. In Study 3, a manipulation intended to influence intentions had a more pronounced effect on selfpredictions than on actual behavior; and in Study 4, self-predictions were closely associated with ratings of intention strength. Furthermore, consistent with the present analysis, self-predictions in both studies tended to underweight (relative to their impact on behavior) factors influencing the reliability with which intentions could be translated into action: The presence or absence of a reminder message (Study 3) or assembly instructions (Study 4) had a much more pronounced influence on people’s behavior than they had predicted. Although further research is clearly needed, the results of these preliminary studies are supportive of the analysis developed here. As with any analysis of biases in intuitive predictions, a natural question arises: Over time, shouldn’t people recognize and eventually learn to avoid making biased self-predictions? The paper concludes with a discussion of obstacles to this form of learning in the context of intention-based self-prediction. More general discussions of how predictive biases may be resistant to learning from experience are offered by Armor and Taylor (1998), Loewenstein and Schkade (1999), and Buehler, Griffin, and Ross (in press). First, of course, it should be noted that reliance on strength of intention will often yield selfpredictions that are reasonably accurate. People’s actions are, after all, driven by their intentions, even if other factors frequently also play a role. Furthermore, as mentioned in the discussion of Studies 1 and 2, intentions are typically attuned, at least to some extent, to considerations of the feasibility of an activity as well as the desirability of the goal achieved through that activity. In other words, intentions are quite different from simple desires in that they provide a more realistic basis for the prediction of future behavior. In light of the degree of unpredictability Self-Prediction Based on Strength of Intention 25 inherent in many forms of behavior, intention strength generally provides a useful—if imperfect—cue, as is borne out by the large body of research conducted within the influential framework of the theory of planned behavior (e.g., Ajzen, 1991). The substantial association between intentions and actual behavior may make systematic bias in self-predictions based on strength of intention difficult for an individual to detect. Second, the self-predictions that implicitly guide decisions may only infrequently be rendered in a sufficiently explicit form to allow a clear assessment of their accuracy. Vague, illdefined self-predictions cannot be easily refuted because they can accommodate a broad range of later behavior. If an explicit prediction was never made or is no longer readily available in memory, people may attempt to reconstruct what they predicted on the basis of what they recall their intentions to have been. To the extent that intentions have changed over time, however, such reconstructive attempts may tend to be influenced by current (behavior-consistent) rather than past (behavior-inconsistent) intentions (cf. Loewenstein, 1996), again with the consequence of obscuring systematic biases in intention-based prediction. Third, when explicit predictions are made, the mere act of making the prediction can change subsequent behavior, bringing one’s actions in closer alignment with one’s predictions than would have been the case in the absence of an explicit prediction (Greenwald, Carnot, Beach, & Young, 1987; Sherman, 1980). As a result, an individual could be convinced by the evidence of his apparently accurate (explicit) predictions while regularly making decisions based on biased (implicit) expectations regarding his future behavior. Fourth, assessments of subsequent behavior can be highly subjective, again with the consequence of obscuring mismatches between predicted and actual behavior. Apparent predictive errors can be explained away by attributions regarding exceptional, unforeseeable events or changes in plans (Buehler et al., 1994). If a situation-specific explanation for a particular predictive error can be produced, an individual may feel no need to search for a more general, systematic bias in self-predictions. In some cases, furthermore, people may fail to identify the extent to which their behavior falls short of their intentions: Kruger and Gilovich (2000) have found that people tend to give themselves (but not others) credit for good intentions that just happen not to have been realized in behavior for circumstantial reasons. Finally, individuals (and researchers) may acknowledge that self-predictions based on intentions frequently overestimate the congruence between predicted and actual behavior, but Self-Prediction Based on Strength of Intention 26 defend the apparent bias on the functional argument that biased predictions provide a motivational “boost” that influences future behavior even if it fails to bring it fully into line with predictions. For example, a student may predict that he will study diligently and receive an A in every course next term, despite having consistently failed to do so in the past, in a strategic attempt to use his expectations to influence his future behavior. Investigation of this issue requires an assessment of the positive and negative consequences of biased expectations and predictions (e.g., Armor & Taylor, 1998). It also requires empirical testing. Recent investigations indicate that the consequences of biased predictions for actual behavior can be mixed. Buehler and Griffin (1997) found that an anchoring manipulation that influenced the relative optimism or pessimism of predictions regarding upcoming projects also influenced actual completion times for certain projects that were small in scope and not subject to obstacles outside the individual’s control. For larger-scale projects and projects with avoidable obstacles, however, biases in prediction exerted no influence on the actual completion time of the project, though it did have some effect on when work on the project was started. One ideal method of reconciling the costs and benefits of biased predictions is to employ different modes of prediction when choosing among possible courses of action (where accurate predictions would be beneficial) than when carrying out the chosen course of action (where optimistic predictions may facilitate performance). Gollwitzer and his colleagues (e.g., Gollwitzer, Heckhausen, & Steller, 1990; Gollwitzer & Kinney, 1989) describe these two modes of prediction as arising from “deliberative” and “implemental” mindsets, respectively. Armor and Taylor (1996) found that participants in an implemental mindset made more optimistic predictions regarding a target task (a scavenger hunt) than did participants in a deliberative mindset, and also found that the former group actually outperformed the latter group on the task. This research suggests that individuals may have available to them a mode of prediction for use in deliberative decision making that is less subject to optimistic bias than the alternative implemental mode that is accessed during execution of a selected course of action. It is possible that the influence of intentions on self-predictions documented in the present paper characterizes what Gollwitzer calls an implemental mindset (although predictions in Study 3 were elicited without asking participants to make a decision regarding their participation in the target activity). The role of intentions in self-predictions made in the course of deliberative decision making merits further exploration. Self-Prediction Based on Strength of Intention27 ReferencesAjzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human DecisionProcesses, 50, 179-211.Ajzen, I., & Madden, T. J. (1986). Prediction of goal-directed behavior: Attitudes, intentions, andperceived behavioral control. Journal of Experimental Social Psychology, 22, 453-474.Armor, D. A., & Taylor, S. E. (1996). Mindset, prediction, and performance: The causes andconsequences of optimism. Unpublished manuscript, UCLA.Armor, D. A., & Taylor, S. E. (1998). Situated optimism: Specific outcome expectancies andself-regulation. In M. P. Zanna (ed.), Advances in Experimental Social Psychology, vol. 30(pp. 309-379). New York: Academic Press.Ausubel, L. M. (1991). The failure of competition in the credit card market. American EconomicReview, 81, 50-81.Buehler, R., & Griffin, D. (1997). Do optimistic time predictions have a functional basis?Unpublished manuscript, Wilfrid Laurier University.Buehler, R., Griffin, D., & MacDonald, H. (1997). The role of motivated reasoning in optimistictime predictions. Personality and Social Psychology Bulletin, 23, 238-247.Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the “planning fallacy”: Why peopleunderestimate their task completion times. Journal of Personality and Social Psychology, 67,366-381.Buehler, R., Griffin, D., & Ross, M. (in press). Inside the Planning Fallacy: The causes andconsequences of optimistic time predictions. In T. D. Gilovich, D. W. Griffin, & D.Kahneman (Eds.), Heuristics and biases: The psychology of intuitive judgment. New York:Cambridge University Press.Bulkeley, W. M. (February 10, 1998). Rebates’ secret appeal to manufacturers: Few consumersactually redeem them. Wall Street Journal, p. B1.Byram, S. (1997). Cognitive and motivational factors influencing time prediction. Journal ofExperimental Psychology: Applied, 3, 216-239.Conner, M., Sheeran, P., Norman, P., & Armitage, C. J. (2000). Temporal stability as amoderator of relationships in the theory of planned behaviour. British Journal of SocialPsychology, 39, 469-493.Cooper, A., Woo, C., & Dunkelberg, W. (1988). Entrepreneurs’ perceived chances for success.Journal of Business Venturing, 3, 97-108.Dolan, C. (October 31, 1989). The American way of buying: Good intentions prop up fitnesssales – Health clubs, costly gear sell but go unused. Wall Street Journal, p. 1.Doll, J., & Ajzen, I. (1992). Accessibility and stability of predictors in the theory of plannedbehavior. Journal of Personality and Social Psychology, 63, 754-765.Epley, N., & Dunning, D. (2000). Feeling “holier than thou”: Are self-serving assessmentsproduced by errors in selfor social prediction? Journal of Personality and SocialPsychology, 79, 861-875.Epley, N., & Dunning, D. (2001). Increasing accuracy and bias: The costs and benefits of self-knowledge in behavioral prediction. Unpublished manuscript, Cornell University.Ferguson, E. (1996). Predictors of future behaviour: A review of the psychological literature onblood donation. British Journal of Health Psychology, 1, 287-308. Self-Prediction Based on Strength of Intention28 Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. AmericanPsychologist, 54, 493-503.Gollwitzer, P. M., & Brandstaetter, V. (1997). Implementation intentions and effective goalpursuit. Journal of Personality and Social Psychology, 73, 186-199.Gollwitzer, P. M., Heckhausen, H., & Steller, B. (1989). Deliberative and implemental mind-sets: Cognitive tuning toward congruous thoughts and information. Journal of Personalityand Social Psychology, 59, 1119-1127.Gollwitzer, P. M., & Kinney, R. F. (1989). Effects of deliberative and implemental mind-sets onillusion of control. Journal of Personality and Social Psychology, 56, 531-542.Greenwald, A. G., Carnot, C. G., Beach, R., & Young, B. (1987). Increasing voting behavior byasking people if they expect to vote. Journal of Applied Psychology, 72, 315-318.Griffin, D., & Tversky, A. (1992). The weighing of evidence and the determinants of confidence.Cognitive Psychology, 24, 411-435.Henry, R. A. (1994). The effects of choice and incentives on the overestimation of futureperformance. Organizational Behavior and Human Decision Processes, 57, 210-225.Henry, R. A., & Sniezek, J. A. (1993). Situational factors affecting judgments of futureperformance. Organizational Behavior and Human Decision Processes, 54, 104-132.Hoch, S. J. (1985). Counterfactual reasoning and accuracy in predicting personal events. Journalof Experimental Psychology: Learning, Memory, and Cognition, 11, 719-731.Kahneman, D., Fredrickson, B., Schreiber, C. M., & Redelmeier, D. (1993). When more pain ispreferred to less: Adding a better end. Psychological Science, 4, 401-405.Kahneman, D. and Lovallo, D. (1993) Timid choices and bold forecasts. A cognitive perspectiveon risk taking. Management Science, 39, 17-31.Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review,80, 237-251.Kahneman, D., & Tversky, A. (1979). Intuitive prediction: Biases and corrective procedures.Management Science, 12, 313-327.Kaplan, R. M., & Simon, H. J. (1990). Compliance in medical care: Reconsideration of self-predictions. Annals of Behavioral Medicine, 12, 66-71.Kridel, D. J., Lehman, D. E., & Weisman, D. L. (1993). Option value, telecommunicationsdemand, and policy. Informational Economics and Policy, 5, 125-144.Kruger, J., & Gilovich, T. (2000). Actions, intentions, and trait assessment: The road to self-enhancement is paved with good intentions. Unpublished manuscript, Cornell University.Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480-498.Leventhal, H., Singer, R. P., & Jones, S. H. (1965). The effects of fear and specificity ofrecommendation. Journal of Abnormal and Social Psychology, 37, 688-714.Lewin, K. (1951). Field theory in social science. New York: Harper & Row.Liberman, N., & Trope, Y. (1998). The role of feasibility and desirability considerations in nearand distant future decisions: A test of temporal construal theory. Journal of Personality andSocial Psychology, 75, 5-18.Loewenstein, G. (1994). The psychology of curiosity: A review and reinterpretation.Psychological Bulletin, 116, 75-98. Self-Prediction Based on Strength of Intention29 Loewenstein, G. (1996). Out of control: Visceral influences on behavior. OrganizationalBehavior and Human Decision Processes, 65, 272-292.Loewenstein, G., Prelec, D., & Shatto, C. (1996). Hot/cold empathy intrapersonal gaps and theprediction of curiosity. Unpublished manuscript, Carnegie Mellon University.Loewenstein, G., & Schkade, D. (1999). Wouldn’t it be nice? Predicting future feelings. In D.Kahneman, E. Diener, & N. Schwarz (eds.), Well-being: The foundations of hedonicpsychology (pp. 85-105). New York: Russell Sage.Netemeyer, R. G., & Burton, S. (1990). Examining the relationships between voting behavior,intention, perceived behavioral control, and expectation. Journal of Applied SocialPsychology, 20, 661-680.Newby-Clark, I. R., Ross, M., Buehler, R., Koehler, D. J., & Griffin, D. (2000). People focus onoptimistic and ignore pessimistic scenarios while predicting their task completion times.Journal of Experimental Psychology: Applied, 6, 171-182.Nunes, J. C. (2000). A cognitive model of people’s usage estimations. Journal of MarketingResearch, 37, 397-409.Osberg, T. M., & Shrauger, J. S. (1986). Self-prediction: Exploring the parameters of accuracy.Journal of Personality and Social Psychology, 51, 1044-1057.Oullette, J. A., & Wood, W. (1998). Habit and intention in everyday life: The multiple processesby which past behavior predicts future behavior. Psychological Bulletin, 124, 54-74.Read, D., and van Leeuwen, B. (1998). The effects of appetite and delay on choice.Organizational Behavior and Human Decision Processes, 76, 189-205.Ross, L., & Nisbett, R. E. (1991). The person and the situation. New York: McGraw-Hill.Sheeran, P., Orbell, S., & Trafimow, D. (1999). Does the temporal stability of behavioralintentions moderate intention-behavior and past behavior-future behavior relations ?Personality and Social Psychology Bulletin, 25, 721-730.Sherman, S. J. (1980). On the self-erasing nature of errors of prediction. Journal of Personalityand Social Psychology, 39, 211-221.Simonson, I. (1990). The effect of purchase quantity and timing on variety-seeking behavior.Journal of Marketing Research, 27, 150-162.Simonson, I., Carmen, Z., & O’Curry, S. (1994). Experimental evidence of the negative effect ofproduct features and sales promotions on brand choice. Marketing Science, 13, 23-40.Sutton, S. (1998). Predicting and explaining intentions and behavior: How well are we doing?Journal of Applied Social Psychology, 28, 1317-1338.Sutton, S., & Hallett, R. (1989). Understanding seat-belt intentions and behavior: A decision-making approach. Journal of Applied Social Psychology, 19, 1310-1325.Taylor, S. E., & Brown, J. D. (1988). Illusion and well-being: A social psychological perspectiveon mental health. Psychological Bulletin, 103, 193-210.Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases.Science, 185, 1124-1131.Vallone, R. P., Griffin, D. W., Lin, S., & Ross, L. (1990). Overconfident prediction of futureactions and outcomes by self and others. Journal of Personality and Social Psychology, 58,582-592. Self-Prediction Based on Strength of Intention30 Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personalityand Social Psychology, 39, 806-820.Wicker, A. W. (1969). Attitude versus action: The relationship of verbal and overt behavioralresponses to attitude objects. Journal of Social Issues, 25, 41-78.Yang, S., Livia, M., & Qi, M. (2001). Wishful thinking and credit card adoption decisions.Unpublished manuscript, University of California – Riverside. Self-Prediction Based on Strength of Intention31 Footnotes1. For one group of participants (n = 368), the questionnaire was entitled, “Life-EnrichingActivities,” and was described as involving activities that “contribute to a student’s health,success in university, and general sense of self-worth and well-being.” For another group (n =397), the questionnaire was entitled, “Everyday Activities,” and was described as involvingactivities that “are common components of a typical student’s life at university.” This wasintended as a subtle manipulation of the perceived importance of the activities and henceintentions to engage in them. In fact, the manipulation was too subtle, having no influence at allon rated intentions or judged probability, so in the analyses below the data are collapsed over thisvariable. 2. Of course, if people in fact rely heavily on intention strength as a basis for probabilityjudgment, we might expect precisely such a confusion, along the lines of Kahneman and Tversky’s observation that their participants apparently failed to distinguish prediction fromevaluation.3. In an attempt to manipulate intention strength at the time of judgment, the questionnairewas preceded by completion of an ostensibly unrelated questionnaire that asked participants to rate themselves relative to other students at the university of the same gender with respect toeither physical fitness (n = 210) or academic ability (n = 227). The preceding questionnaire was intended to “prime” goals related to one of the two domains evaluated in the main questionnaire.The priming questionnaire consisted of four attributes (for physical fitness: aerobic or cardiovascular fitness, upper body strength, lower body strength, general overall fitness; foracademic ability: verbal ability, mathematical ability, creativity or imagination, overall academicability). For each attribute, participants rated themselves relative to other students by checkingone of five ordered boxes (e.g., much stronger, a bit stronger, about as strong as, a bit less strong,or much less strong than other students). The priming manipulation did not have the anticipatedeffect on the importance ratings. An analysis of the mean rated importance of the health andacademics items showed only a small unexpected main effect in which participants given thephysical fitness priming questionnaire gave slightly higher importance ratings in both the healthand academics domain than did participants given the academic ability priming questionnaire.The priming manipulation also had no effect on the probability judgments. Given its failure tohave the intended effect, subsequent analyses collapse over the priming questionnaire variable. Self-Prediction Based on Strength of Intention32 4. We also attempted to manipulate participants’ intentions to successfully complete the taskby offering half of them a $5 bonus if they could correctly assemble the model within the 30-minute time limit. No financial incentive was offered to the remaining participants. As it turnsout, the financial incentive did not have its anticipated effect, as is revealed by an analysis of theintention strength rating serving as a manipulation check. This measure showed no significanteffect of financial incentive, F(1, 56) = 1.03, n.s., with a mean rated intention strength of 8.0 forthose given the incentive and 8.2 for those not given the incentive. As such, we would notexpect to find a corresponding influence of the financial incentive manipulation on thepredictions. As it turned out, participants assigned a significantly lower probability of correctlyassembling the model when offered a financial incentive for doing so than when not offered afinancial incentive, F(1, 56 ) = 6.97, p < .01. In hindsight, the explanation for this result is fairlyobvious: Participants offered a financial incentive used it as a source of information from which to infer the difficulty of the task. Indeed, several participants commented that if there was a $5reward for correctly assembling the model within 30 minutes, then it must be a challenging feat to accomplish. A small follow-up study confirmed this interpretation: Participants wereassigned to either a large ($5) or small ($1) performance incentive as determined by a coin toss conducted in their presence, which was intended to eliminate any perceived informativeness ofthe incentive. The two groups did not differ in their self-rated intentions or predictions regarding task performance. Because the presence or absence of an incentive in the main study had noeffect on intention strength ratings or actual performance, and did not interact with the instructions variable in its effect on either predictions or performance, it is not considered furtherhere. Self-Prediction Based on Strength of Intention33 Table 1List of target goals and activities in Study 1, with [easy/hard] version in brackets. Health• exercise at least [2/5] hours per week next term• avoid watching more than [10/5] hours of television per week next term• avoid eating more than one unhealthy “fast-food” meal per [week/month] next term• avoid eating more than one “junk-food” snack per [day/week] next term Academics• study for a minimum of [15/30] hours for each final exam during the exam period this term• hand in all course papers by their due dates for the rest of this [term/academic year]• avoid falling more than [2/1] week[s] behind in reading for any course next term• attend at least [one/two] public lecture[s] on campus next term (not required for course) Prosocial Behaviors• phone parents at least once every [month/week] during the rest of the academic year• phone or write to [at least one/several] friend[s] from high school over the winter break• donate blood at least [once/twice] in the next 12 months• work at least one shift at a soup kitchen or food bank in the next [12/2] months Table 2Results of hierarchical regression analyses with judged probability as the criterion and targetactivity, self-rated strength of intention, and difficulty level of the activity as predictors in Study 1. predictorR ∆ R1. target activity.3082. intention strength .708 .4003. difficulty level.711 .003 predictorR ∆ R1. target activity.3082. difficulty level.334 .0263. intention strength .711 .377 Self-Prediction Based on Strength of Intention34 Table 3General goals and specific target activities [easy/hard versions in brackets] in Study 2. Odd-numbered items concern health and exercise; even-numbered items concern academics. General Goals1. exercise regularly2. study hard for final exams3. avoid eating “junk-food” snacks4. hand in course papers by their due dates5. avoid eating unhealthy “fast-food” meals6. avoid falling behind in course reading7. maintain or lose weight8. attend course lectures Specific Activities [easy/hard]1. exercise at least [2/5] hours per week next term2. study for a minimum of [10/20] hours for each final exam during the exam period this term3. avoid eating more than one “junk-food” snack per [day/week] next term4. hand in all course papers by their due dates for the rest of this [term/academic year]5. avoid eating more than one unhealthy “fast-food” meal per [week/month] next term6. avoid falling more than [2 weeks/1 week] behind in reading for any course next term7. [avoid gaining more than 5 pounds/maintain current weight (or lose weight)] between nowand end of Winter term8. [miss very few lectures in courses/attend every lecture in all courses] next term Table 4Results of hierarchical regression analyses with judged probability as the criterion and targetactivity, self-rated goal importance, and difficulty level of the activity as predictors in Study 2. predictorR ∆ R1. target activity.1742. goal importance .387 .2133. difficulty level.406 .019 predictorR ∆ R1. target activity.1742. difficulty level.199 .0263. goal importance .406 .207 Self-Prediction Based on Strength of Intention35 Table 5Results of hierarchical regression analyses with judged probability as the criterion and presence orabsence of instructions and self-rated strength of intention as predictors in Study 4. predictorR ∆ R1. intention strength .2392. instructions.251 .012 predictorR ∆ R1. instructions.0622. intention strength .251 .189 Self-Prediction Based on Strength of Intention36 Figure 1Decomposition of predictive validity of intention strength into temporal stability and translationreliability. intentionstrengthat T1intentionstrengthat T2 behaviorat T2temporal stability translation reliabilitypredictive validity Self-Prediction Based on Strength of Intention37 Figure 2Regression lines and conditional means showing judged probability as a function of self-ratedintention strength in Study 1 for easy and difficult target activities. 1 2 3 4 5 6 7 8 90102030405060708090100easy activitieshard activitieseasy regression linehard regression line strength of intentionjudgedprobability(%) Self-Prediction Based on Strength of Intention38 Figure 3Regression lines and conditional means showing judged probability as a function of self-ratedpersonal importance in Study 2 for easy and difficult target activities. 12345670102030405060708090100easy activitieshard activitieseasy regression linehard regression line personal importancejudgedprobability(%) Self-Prediction Based on Strength of Intention39 Figure 4Predicted and actual completion (performance) probability as a function of self-rated intentionstrength in Study 4 for participants with and without instructions. Note: Intention strength ratings in each condition were subject to a median split; the figure plotsthe mean predicted and actual completion probability against the mean intention strength ratingof each resulting subgroup.0102030405060708090100 6 7 8 9 10intention strengthcompletionprobability(%) no instructions: predictedno instructions: actualinstructions: predictedinstructions: actual