To help or not to help: Capturing individuals’ decision policies

To help or not to help: Capturing individuals’ decision policies

Fritzsche, Barbara A

The arousal: cost-reward model of bystander intervention developed by Piliavin, Dovidio, Gaertner and Clark in 1981 was tested using a within-subjects “policy capturing” methodology. Four hundred and forty nine participants read 50 scenarios and reported the likelihood they would offer help. Seventy-six percent of the participants’ helping judgments could be reliably described or “captured” with a linear combination of the various costs of helping and costs of not helping specified in the model. In addition, participants were relatively aware of how the costs affected their helping decisions; although female participants may have been more aware than males. These findings provide additional support for the arousal: cost-reward model and extend understanding of the cognitive algebra that occurs before individuals decide to intervene.

In the thirty years since the publication of Latane and Darley’s classic study of the “unresponsive bystander” (Latane & Darley, 1968), various theories have been proposed to explain how individuals make helping decisions in emergencies (see Schroeder, Penner, Dovidio, & Piliavin, 1995). Some of the more popular models view a bystander’s actions as resulting from a cognitive decision process. Among these are the Latane and Darley (1970) decision model of bystander intervention and Piliavin, Dovidio, Gaertner, and Clark’s (1981) arousal: cost-reward model (see also Dovidio, 1984; Dovidio, Piliavin, Gaertner, Schroeder, & Clark, 1991).

The latter model provided the focus of the present study. According to the arousal: cost-reward model, another person’s distress causes physiological arousal in an observer which, in turn, initiates the process of deciding whether to help. Specifically, because people find prolonged physiological arousal aversive, they try to find ways to reduce it. The decision as to what course of arousal reduction to pursue involves weighing the perceived costs to the potential helper of helping (e.g., time, money, effort) and of not helping (e.g., guilt, criticism) and then choosing the response that incurs the smallest net cost. Thus an observer is most likely to offer assistance when the personal costs of helping are low, and the costs of not helping are high. Providing rewards also increases the probability of helping (Piliavin et al., 1981).

There are, of course, circumstances when the decision process becomes more difficult. One of these is when costs of helping are extremely high, but so are the costs of not helping. In these instances, the model posits that people will opt for responses that reduce the cost of either helping or of not helping. For example, if direct helping would produce extremely high costs for the helper, arousal can be reduced by providing indirect assistance (e.g., summoning others to help). Alternatively, when helping costs are extremely high, one may cognitively reinterpret the situation to diminish the perceived costs of not helping. One mechanism for doing this is to diffuse the responsibility for not helping. Thus, if others are present when the need for help arises, a bystander may decide that they will provide the necessary aid (e.g., Gaertner & Dovidio, 1977). In doing this, the observer reduces his/her personal responsibility for not helping, and thus the costs of inaction.

In the years since the model was originally proposed, it has undergone some changes and modifications. For example, very early in the model’s development Piliavin et al. acknowledged that helping in some circumstances may be impulsive and preceded by little, if any, conscious or nonconscious decision processes (e.g., a parent seeing a child in grave danger). Somewhat later the model was revised to include a less egoistic or “selfish” characterization of the arousal that precedes the decision process. More specifically, the work of Batson (1991) and others (see Batson, 1997; Schroeder et al., 1995) led Piliavin et al. to acknowledge that the subjective experience of arousal can take many forms, ranging from fear for one’s own safety, to other-oriented empathic concern for the welfare of the person in need. Other work (e.g., Otten, Penner & Waugh, 1988) has shown that the cost/ reward portion operates in the manner suggested by Piliavin et al. even when there is no prolonged physiological arousal. Despite these qualifications and modifications, the crux of the model remains intact- that bystanders’ helping decisions result from a consideration of two general categories of costs.

The results of literally hundreds of experiments provide independent empirical validation of the basic premises of the arousal: cost-reward model (see Dovidio et al., 1991; Schroeder et al., 1995). However, in almost all instances this support comes from between-subject tests of the model. In these studies, researchers have created different levels of costs of helping (or withholding help) and examined the effects of this manipulation on participants’ helping decisions. For example, in one of the first studies designed to test predictions generated by the arousal: reward-cost model, Piliavin, Piliavin, and Rodin (1975) employed a between-subjects methodology and manipulated: a) the cost of helping by varying the physical appearance of the potential recipient of help (normal-low cost versus physically disfigured- high cost) and b) the cost of not helping by varying the number of people present when the request for help occurred (other potential helpers present low-cost versus no other people present-high cost). Consistent with the model’s predictions about the interaction between these two kinds of costs, the presence of others affected helping only when the costs of helping were high.

Although such results are clearly consistent with the model, they do not speak directly to one of its central tenets; namely, before deciding to intervene or not intervene the individual weighs the different costs of helping and not helping. That is, Piliavin et al.’s results show that different levels of costs or combinations of the levels do produce different amounts of helping. However, the methodology used does not permit one to directly examine the manner in which individual participants evaluated the effects of these costs, or the relative influence of these costs, on a decision to offer help.

Thus, it would seem worthwhile to test the arousal: cost-reward model by directly examining the manner in which a person weighs the information he or she receives about the costs of helping and of not helping. When the same individual is systematically presented with varying levels of the costs of interest, the relative impact of these costs on the helping decision can be assessed. As far as the authors are aware, this methodology has not been used to study the arousal: cost-reward model; and this was the primary focus of the present study. Using a within-subject methodology, several costs of helping – and of not helping – were manipulated, helping decisions were obtained, and attempts were made to generate a statistical description of how the participants reached a decision about helping.

The specific methodology used to do this is called “policy capturing.” It has been widely and effectively applied to studies of clinical judgments, financial decision-making, and social policy judgments (see Hammond & Wascoe, 1980; Stevenson, Busemeyer, & Naylor, 1990), but the approach has received little attention from social/personality psychologists (c.f., Finkelstein & Brannick, 1997). Therefore, a somewhat detailed description of the procedure is presented below.


Policy capturing provides a methodology for modeling how individuals or groups of individuals weigh and combine bits of information to make a judgment (e.g., Donnelly & Bownas, 1984). The first step in a policy capturing study is to identify the set of variables (or cues) that are thought to affect the decision of interest. These variables are called “cues;” they may be selected from experts’ analysis of the decision or derived from some theory or model. To test the relative importance of the cues, the cue values are manipulated and combined into unique configurations or “profiles.” Each profile thus contains the same categories of cues, with their levels varying across profiles. The participant’s task is to make repeated judgments based on the information contained in each profile.

Profiles can be developed in one of two ways. In a representative design, the profiles comprise a representative sample of the stimulus configurations that one would encounter in the “real world;” the covariation among the cues reflects their natural covariation. Systematic designs, however, are more often used to study decision-making. In a systematic design, all possible combinations of cue levels are presented in separate profiles. Although some configurations may not be representative of real-world profiles, the values of the cues are orthogonal to one another. In a systematic design, each cue’s regression weight represents its independent contribution to the decision.

Given a sufficient number of profiles (and judgments), one can calculate a withinsubject multiple regression in which the criterion measure is the judgment and the predictor variables are the cues. If the R^sup 2^ reaches some specified value (often >= 0.50), then the decision policy is said to have been “captured.” The beta coefficients in the regression equation provide statistically-derived measures of the importance of each cue to the decision (Keely & Doherty, 1972). Policy capturing does not require that people actually use a linear additive model to make judgments, only that their judgments can be reliably modeled with multiple regression; the underlying process may be quite different (e.g., Hoffman, 1960). Even when the decision process is nonlinear, judgments are often more accurately described by simple linear models than by rater intuition (e.g., Dawes, 1979).

The policy capturing methodology was used here to address three sets of questions about the impact of costs on helping decisions. First, will the basic premises of the model hold when the impact of costs on helping is directly assessed from an intraindividual perspective? That is, do people use the costs of helping and of not helping when they make decisions about whether or not to intervene? It was expected that they would. To test this prediction the authors examined whether it was possible to reliably predict helping decisions from a linear combination of helping costs.


The second major set of questions concerned the participants’ knowledge of their own helping policies. An extensive literature in cognitive and social psychology suggests that people may have little insight into their cognitive processes (e.g., Nisbett & Wilson, 1977; Plous, 1993; Ross & Nisbett, 1991). Here the question was whether or not a similar lack of insight characterizes helping decisions. There is some reason to expect that this would be the case.

Several studies have compared the policies produced by regression equations based on how individuals actually made decisions to the policies produced by equations based on how participants believed they made the same decisions. In most instances, the empirically derived regression weights better predicted the actual decisions than did the participants’ subjective self-reports of how important each of the cues was in their decisions (e.g., Arnold & Feldman, 1981; Brehmer & Brehmer, 1988; Donnelly & Bownas, 1984; Hobson, Mendel & Gibson, 1981; Slovic & Lichtenstein, 1971; Zedeck & Kafry, 1977).

For example, Donnelly and Bownas (1984) used policy capturing to obtain performance appraisals of firefighters. Supervisors read profiles containing sixteen cues (performance dimensions) and rated overall performance based on the information in each profile. Subsequently, the supervisors ranked the cues according to the perceived importance of each to ratings of performance. The subjective rankings of the importance of the cues differed from the statistically or objectively-derived estimates of the relative importance of the same cues. Moreover, when the two sets of rankings were used to predict the supervisors’ actual ratings, the statistically-derived ones provided significantly better predictions. In the present study, the question of insight about one’s policies was moved from the area of performance appraisals to the area of helping decisions.

To examine the issue of knowledge of one’s own policies, two regression equations were generated for each participant. The first used the weights that were statistically-derived from the participants’ policy capturing data. The second used weights based on the same participants’ subjective estimates of the importance of each cue. The prior policy-capturing literature led the authors to predict that the objectively-derived (i.e., statistically-derived) policies would better model helping judgments than would the subjective estimates of the importance of the individual cues.


The final issue of interest in the present study concerned comparisons of the helping policies of the male and female participants. It is well established that even though men and women do not differ overall in their willingness to help, they do differ in the kinds of help they offer. Men are more likely to help strangers with physical problems (e.g., a disabled vehicle). Women, on the other hand, are more likely to offer emotional help to people they know. (e.g., Otten et al., 1988; Penner, Dertke, & Achenbach, 1973; Piliavin & Unger, 1985). Eagley and Crowley (1986) attribute such differences in helping to gender roles. Women are expected to care for the personal and emotional needs of others, to deliver routine forms of personal service, and more generally, to facilitate the progress of others toward their goals. The demand to serve others is especially strong within the family and, to a certain extent, in other close relationships such as friendships.

In the present study, the participants were asked if they would provide comfort to a friend who had lost a romantic partner. Thus, it was expected that female participants would be more likely to offer help than would males. The use of a policy capturing methodology also allowed the authors to investigate another more subtle kind of gender difference in helping – the helping policies of men and women. To address this issue the regression weights (both those derived from the regression equations and those derived from subjective ratings) for the policies of male and female participants were compared. Although gender differences in policies were expected, the authors did not make any specific predictions about the nature of these differences.



The policy capturing task was completed by 449 students1 (80 males, 362 females, and 7 who did not indicate gender) enrolled in undergraduate psychology courses at a large, southeastern university. One hundred and eighty-four of these participants (33 males, 151 females) also completed the subjective rating task.


Development of the Helping Scenarios. In a prior study concerned with the arousal: cost-reward model, Otten et al. (1988) developed scenarios that concerned helping someone in a non-emergency situation. The helping scenarios describe a person who is despondent and distressed over the end of a romantic relationship, and who calls a same-sex friend for help. The potential helper is studying for an upcoming examination, but is asked by the distressed person to come over because he/she is “really upset and needs someone to talk to.” Following this, differing kinds of information are provided regarding the person requesting help, the amount of help requested, and the conditions surrounding the provision of help. Participants place themselves in the role of the potential helper and indicate how likely it is that they would offer help. These scenarios were modified slightly and used in the present study.

Specifically, the scenarios contained three cues concerned with costs of helping and three cues concerned with costs of not helping. Within the scenarios, each cue was set at one of two levels of cost (low or high). The six costs and the levels associated with each of them are presented below.

Costs of Helping:

1. Time Required. The distressed friend asked for either one hour (low cost) or three hours (high cost) of help.

2. Discomfort in Helping. The potential helper was either uncomfortable (high cost) or comfortable talking to people in situations “like this” (low cost)

3. Nearness of the Examination. The potential helper had an exam either the next day (high cost) or in two days’ time (low cost).

Costs of Not Helping:

4. Requestor’s Responsibility for the Problem. The relationship had ended because the boyfriend/girlfriend had discovered that the friend was secretly seeing other people (low cost) or because the friend had to spend a lot of time studying due to illness and the boyfriend/girlfriend preferred someone who had more time for a relationship (high cost).

5. Ability to Diffuse Responsibility. There were other friends who might come over to comfort the distressed friend (low cost) or there were no other friends who were available (high cost).

6. Requestor’s Deservingness. The distressed friend had previously not helped others in a similar situation (low cost) or had helped others in a similar situation (high cost).

These costs and levels were manipulated in a 2 x 2 x 2 x 2 x 2 x 2 withinsubjects factorial design, creating 64 scenarios. Because there is a tradeoff between the ability of the participants to reasonably cope with the lengthy judgment task and the need for testing enough scenarios to produce stable regression coefficients, guidelines have been developed for choosing the number of scenarios needed. A minimum profile-to-cue ratio for stable regression estimates is 5 to I (Cooksey, 1996). In this study, the 5 to 1 ratio suggests that at least 30 scenarios should be used. For this study, 50 scenarios were randomly selected.2

Policy Capturing Task. Each participant read each of the 50 scenarios and played the role of the person who had been asked for help. For each profile, the participants indicated the likelihood that they would help in the situation on a scale that ranged from “1” (Not at all) to “5” (Extremely).

Subjective Rating Task. Following the policy capturing task, a subset of participants provided their own estimates of the weights given to each of the six cues in making their helping judgments. The procedure for obtaining these subjective weights was taken from prior policy capturing studies (e.g., Arnold & Feldman, 1981; Carkenord & Stephens, 1994; Pargament, Sullivan, Balzer, Van Haitsma, & Raymark, 1995). Specifically, the participants were provided with a list of the cues and asked to distribute 100 points to reflect the importance of each cue. They were instructed to allocate the greatest number of points to the cue that most influenced their decisions to help and the lowest number of points to the least influential cue, making sure that these added up to 100.


Overall, the results were consistent with the arousal: cost-reward model. For example, the model predicts that helping should increase as the costs of helping decrease, and the costs of not helping increase. Across all 50 scenarios, the mean likelihood of helping was 2.85 (SD=-l.22). As expected, the least amount of helping (M = 1.63; SD = .75) was offered in response to the scenario containing the highest costs of helping and the lowest costs of not helping. In addition, the greatest amount of helping (M = 4.29; SD =.86) was offered in response to the scenario containing the lowest costs of helping and the highest costs of not helping.

To further examine the overall amount of helping within the context of the arousal: cost-reward model, each scenario was coded from 0 (i.e., all 3 costs of helping were high and all 3 costs of not helping were low) to 6 (i.e., all 3 costs of helping were low and all 3 costs of not helping were high) and this overall cost variable was correlated with mean likelihood of helping ratings for each scenario. The resulting correlation was .95 (p


The primary purpose of this study was to examine how individuals weigh various costs of helping and not helping in their decisions to intervene (or not intervene). The relative contribution of each of the six costs to an individual’s helping judgments was determined by a within-subjects regression analysis. Each participant’s 50 likelihood of helping judgments were regressed onto the 6 costs. The resulting regression equation represented the participant’s “helping policy.” A policy was considered statistically “captured” if the R^sup 2^ for the regression equation was equal to, or greater than, .50. That is, the linear combination of the six cues could explain at least 50% of the variance in a participant’s helping decision.

Because the F-test for testing the statistical significance of R^sup 2^ requires information about the amount of variance accounted for, the number of cues (in this case, 6), and the sample size (in this case, 50 profiles), even an R^sup 2^ as small as .25 would be statistically significant for this judgment task. However, statistical significance and meaningfulness are not necessarily synonymous, and judgments of meaningfulness should be based on the particular area of research being studied (e.g., see Bobko, 1995; Pedhazur, 1997). Hence, the criterion set for concluding that a participant’s helping policies had been captured was an R^sup 2^ of .50. That is, the criterion ensured that the R^sup 2^ was significant (with 6 and 43 degrees of freedom) and that it accounted for at least 50% of the variance in a participant’s judgments. This decision rule was consistent with that employed in other studies of policy capturing that use a similar number of cues and judgments (see Cooksey, 1996).

The R^sup 2^ for the entire sample ranged from .04 to .97 (M = .60; SD = .18). Based on the .50 criterion, it was possible to capture the policies of 339 out of 449 (76%) participants. The fact that a substantial percentage of the participants’ policies could be captured allowed the authors to proceed to the other questions of interest in this study. In most of the analyses that address these questions, participants with an R^sup 2^ of less than .50 were excluded. The rationale for excluding these individuals was that comparisons that included policies that were unreliable, inconsistent, or could not be captured with a linear regression would not be meaningful.

To examine the relative weights of the costs on participants’ judgments, the beta weights derived from the within-subject regressions were averaged across the participants and regressed on the costs of helping and costs of not helping. The resulting R^sup 2^ was .98 (F (6, 43) = 293.71, p


The second issue of interest was the participants’ knowledge of their own helping policies. It was expected that participants would have little insight into their helping policies. To compare the participants’ statistically-derived policies (based on responses to each profile) with their subjective judgments of these weights, a procedure described by Arnold and Feldman (1981) was used. Beta weights for the cues in the statistically-captured policies were converted into relative weights by multiplying them by their validity coefficients and dividing by the squared multiple correlation between the participants’ judgments and the six cues. Relative weights represent the proportion of predictable linear variance in judgments that each cue provides. Hence, across the six cues, the relative weights sum to one. Then, relative weights were multiplied by 100 so that they could be directly compared to the subjective weights provided by the participants.

After the statistically-derived weights were converted to the same scale as the subjective weights, three analyses were conducted to evaluate the participants’ self-insight into their helping policies. First, a t-test was calculated to test the difference between the statistically-derived and subjective weights for each cue. Second, the correlation between the two kinds of weights was calculated. Third, the subjective weights were used as beta weights in a regression equation to derive predicted likelihood of helping judgments. These predicted judgments were then correlated with the participants’ actual helping judgments for each scenario. The resulting correlation indicates the degree of convergence between the statistically– derived and subjective helping policies.

The means and standard deviations of the subjectively-derived and statistically– derived weights are presented in Table 1. Matched-pair t tests were calculated to compare the two kinds of weights. Nonsignificant t’s would indicate agreement between the participants’ statistically-derived and subjective policies. Because of the large number of comparisons conducted in this study, a Bonferroni correction was applied to all comparisons of regression weight to control for experiment-wise error rate. Applying this correction, the alpha level for the t statistic was .008. The same correction and alpha level were used for all subsequent t tests.

As seen in Table 1, significant differences between self-reported and statistical weights were found for all cues except one, the amount of discomfort that helping would produce (p = .02). Participants gave the requestor’s deservingness, amount of time required, and the ability to diffuse responsibility significantly more importance in their helping judgments than their statistically-derived policies indicated. On the other hand, they gave the requestor’s responsibility for the problem and the nearness of the exam less importance than their statistical policies indicated.

Whereas the mean comparisons provided information on whether or not the subjectively-derived regression weights were significantly displaced from the statistically-derived weights, the next analysis addressed the extent to which the two sets of scores covaried. As can be seen in Table 1, the same two costs were the most important for statistically and subjectively derived weights- victim’s responsibility for the problem and time. However, responsibility for the problem was the most important statistically-derived cost and time required was the most important subjectively-derived cost. Overall, the covariation between the rank orders of the weights derived by statistically-derived and subjective methods was rather substantial; the Spearman rank order correlation was .89 (p

In the final analysis concerned with the two kinds of weights, the subjectively– derived weights were entered into a regression equation to produce predictions of the participants’ actual responses for each scenario. The predicted responses correlated .95 (p


The nature of helping that was requested in the scenarios (providing support and counseling to a friend) appeared to be more consistent with Eagly and Crowley’s (1986) description of female-oriented helping than male-oriented helping. Therefore, women were expected to be more willing to help than men. This hypothesis was not supported. Although the mean likelihood of helping among the women (2.88; SD = .70) was higher than that among the men (2.75; SD = .68), this difference was not significant, t (333) = 1.24, p > .05.

The next set of analyses concerned gender differences in statistically-derived helping policies. Specifically, the authors sought to determine whether men and women differed in the way they made decisions about offering help. As noted earlier, only the men and women with statistically-derived policies that produced R^sup 2^ s of .50 or more were used in the analyses that follow. Mean comparisons were conducted, a rho correlation was computed, and the actual responses of the men were correlated with those predicted by a regression equation using the regression weights of the women and vice-versa.

The standardized beta weights for men and women are presented in Table 2. These weights were compared using t-tests for independent samples. Using an alpha level of .008 (i.e., applying the Bonferroni correction), none of the six comparisons reached statistical significance. Similarly, there was a substantial degree of covariation in regression weights across gender (rho = .71; p > .05), and helping judgments of one gender were almost perfectly correlated with those produced by the other gender’s regression weights (r = .98; p

Finally, gender differences in subjectively-derived helping policies were examined. Specifically, the authors sought to determine whether men and women differed in the way they estimated the relative weight of each cost to their helping judgments. Once again, mean comparisons were conducted, a rho correlation was computed, and the actual responses of the men were correlated with those predicted by a regression equation using the subjective weights of the women and vice-versa.

Table 3 presents the means and standard deviations of the relative weights and subjective weights for men and women. (Beta weights were converted into relative weights so that the subjectively- and statistically-derived weights are measured on the same 100-point scale.) Applying the Bonferroni correction, there were no significant differences between the two sets of weights for men. However, the covariation between the rank orders of the weights derived by statistically-derived and subjective methods was not significant for men (rho = .31; p > .05). The largest difference was that men ranked responsibility for the problem (a cost of not helping) as the fifth most important cue, but it had the largest beta weight in the statistically-derived regression equation. Time required (a cost of helping) was ranked by men as the most important cue, but the regression weights indicate that it was the third most important cue.

For women, there were significant mean differences between the statistically– derived and subjective weights assigned to three of the cues. Relative to the statistically-derived weights, they overestimated the importance of two costs of not helping (requestor’s deservingness and ability to diffuse responsibility) and underestimated the importance of one cost of helping (nearness of the examination). However, the women’s rank order of the subjective weights was very similar to the rank order of the statistically-derived weights (rho = .89; p

Despite differences in statistically-derived and subjective weights within gender, the correlation between one gender’s statistically-derived helping judgments and predictions of those judgments from the other gender’s subjective weights was high (r = .94 to .95, p


Overall, the results of this study provided additional support for the arousal: cost-reward model of helping. As the model would predict, across the 50 scenarios the least helping occurred in the one where the costs of helping were the lowest and the costs of not helping were the highest. Furthermore, the correlation between the values of the costs and of helping decisions approached unity. Thus, the present study, like earlier studies (e.g., Otten et al., 1988), suggests that the model is valid even in circumstances where no emergency or physical distress is present.

Using a statistically-derived linear regression, the authors were able to adequately capture the helping policies of over 75% of the participants. This suggests that people do weigh the various costs associated with helping in a manner suggested by the model. Helping judgments could be statistically “captured” using only information about the various costs associated with helping and not helping. Of special note is the fact that these results were obtained without using any interaction terms (i.e., one cost times another cost). These findings suggest that people do, indeed, engage in a form of “cognitive algebra” when deciding whether or not to help.

The second issue addressed in this study was the extent to which individuals were aware of their helping policies. In the present study, participants’ estimates of their statistically-derived policies agreed fairly well with their actual policies. Although there were significant differences between several of the statistically– derived and subjectively-derived weights, overall there was substantial agreement between the relative magnitudes of the statistically-derived weights for the cues and the participants’ estimates of the relative importance of the cues. The one exception to this pattern was found among the male participants. In contrast to the strong rank-order correlation between the two kinds of weights among females, the correlation among the males was small and nonsigificant. The authors have no ready explanation for this unexpected finding. However, one possibility is that because this kind of helping is less common among men than among women, the male participants had more difficulty identifying the manner in which they made the helping decisions of interest.

The participants (at least the females) seemed to have greater awareness of their policies than participants in other policy capturing studies (e.g., Pargament et al., 1995; Viswesvaran & Barrick, 1992). This may reflect the relative complexity of the judgment task used in this study vis-vis the other studies. In the present case, each cue had only two levels, and there were only six cues. A less complex judgment task may make it easier for participants to be aware of their policies than a task with many cues and/or many levels of cues. Other policy capturing studies that have used only two levels of cues have also found fairly high agreement between statistically-derived weights and subjective estimates of those weights. For example, Harrison, Ryan, and Moore (1996) used only two levels of each cue and found that college students had good high awareness of their policies when rating the classroom performance of college instructors.

It should be noted also that in many of the social psychological studies of insight into decision-making, there is some deliberate attempt to hide from the participants the reasons for a decision, or even to mislead them somewhat about why they made a certain decision. In contrast, in the present study participants were presented with the variables that were manipulated, and asked in a straightforward way to indicate the extent to which these variables influenced their decisions. Thus, perhaps it is not surprising that most of the participants had insight into their helping decisions.

The final issue examined in this study was gender differences. As noted earlier, the kind of helping requested in the present study, emotional support, is consistent with the female gender role. In the present study, however, women were not more likely than men to help the requestor, and the rank order of the cues that influenced helping was very similar for males and females. For both males and females, responsibility for the problem was the most important cue and discomfort in helping was the least important cue in their statistically-derived judgment policies. Further, the correlation between the relative magnitudes of these statistically-derived weights for men and for women was large and significant; and the same was true for the subjectively-derived weights.


Perhaps the major conclusion that can be drawn from this study is that when the “cognitive algebra” suggested by the arousal: cost-reward model was directly examined in a within-subjects analysis of the participants’ helping policies, the results supported the model. The cues (i.e., the costs) did relate to the helping decisions in a manner that was, for the most part, consistent with the model. Specifically, each of the costs that were operationalized in this study contributed significantly to the participants’ helping decisions; and the contributions appeared to be additive. Moreover, as others (e.g., Otten et al., 1988) have found, the model was predictive of helping decisions in a nonemergency, nonarousing situation.

Besides providing additional support for the arousal: cost-reward model, this study also demonstrated the value of a policy-capturing methodology for social psychologists interested in helping, as well as in other kinds of decisions. The fact that the authors are able to discuss the relative importance of the costs in the participants’ helping decisions illustrates the major advantage of using policy capturing to study helping. It permits a direct comparison of the importance of different costs in decisions to offer (or not offer) help and thus enables direct examination of one aspect of the model of interest. As suggested in the introduction, this is not possible in the between-subject designs typically used in studies of models of bystander intervention.

On the other hand, the primary limitation of this study is that simulated, rather than actual, judgments were studied. It is certainly possible that the participants’ responses might differ from what they actually would do when asked to provide help. This is a problem in policy capturing or any other simulation methodology. However, it must be noted that the participants’ paper-and-pencil responses were highly consistent with the findings from both laboratory and nonlaboratory studies of the Piliavin et al. model, in which the dependent measures were helping behaviors (see Dovidio, et al., 1991; Schroeder et al., 1995). More importantly, the simulation used in the present instance allowed the authors to study directly the decisionmaking processes that may underlie these overt behaviors. Thus, it helps to further explicate the relationship between cost estimates and helping decisions and suggests that policy-capturing may be a valuable tool with which to study helping (and many other social psychological processes).

When interpreting the findings, it is also important to consider possible alternative explanations of the participant’s responses that involve factors other than the costs of helping. For example, Weiner (1995) argued that judgments of responsibility (one of the costs of not helping herein) affect a potential helper’s affective reactions to the person in need of aid, which in turn influence the motivation to help. According to Weiner, if the victim is seen as responsible for his/her plight, this elicits negative affect and a decreased willingness to help. Thus the impact of responsibility on helping may have been due to a mechanism other than the cost of not helping.

Similarly, Feather (1992) argued that judgments of deservingness (another cost of not helping in the present study) and responsibility are dependent upon affective reactions to the balance between valued behaviors and valued outcomes. That is, individuals are perceived as deserving when positive behavior (e.g., high effort) results in valued outcomes (e.g., good grades) and when negative behavior (e.g., laziness) results in negative outcomes (e.g., poor grades). Individuals are judged as undeserving when positive behavior (e.g., high effort) results in negative outcomes (e.g., poor grades) or vice-versa.

More generally, Feather (1992) and Weiner (1995) suggest that there is a complex relationship between responsibility for a problem, deservingness, and behavior. In the present study, the manner in which responsibility and deservingness were manipulated and the measures that were obtained did not allow an examination of this complexity. Hence, future research should further examine these and other costs of not helping before any final conclusions about the effects of costs on helping are drawn.

Also, future research might retrospectively catalogue actual incidents of helping, identify the cues present in these situations, and then use policy capturing to create a model of how these cues affect helping. That is, rather than studying helping with a factorial or systematic design, one could use the more ecologically valid representative design suggested by Egon Brunswik 45 years ago (Brunswik, 1955). This may give additional insight into when people help and when they do not.

1 In addition to completing the policy capturing task, these participants were part of the normative sample for the development of the Prosocial Personality Battery (see Penner, Fritzsche, Craiger & Freifeld, 1995).

2 The intercorrelations of the cues across the 50 scenarios were computed to verify that the cues were uncorrelated with each other. The largest correlation between any two cues was. 16 (p >.05), indicating that the regression weights derived from the policy capturing analysis could be treated as zero-order correlations with the helping judgments.


Arnold, H. J. & Feldman, D. C. (1981). Social desirability response bias in self-report choice situations. Academy of Management Journal, 24, 377-385.

Batson, C. D. (1991). The altruism question: toward a social-psychological answer. Hillsdale, NJ: Erlbaum.

Batson, C. D. (1997). Self-other merging and the empathy-altruism hypothesis: Reply to Neuberg et al. (1997). Journal of Personality and Social Psychology, 73, 517-522.

Bobko, P. (1995). Correlation and regression: Principles and applications for industrial/organizational psychology and management. New York: McGraw-Hill.

Brehmer, A. & Brehmer, B. (1988). What have we learned about human judgement from thirty years of policy capturing? In B. Brehmer, C. R. B. Joyce, et al. (Eds.), Human judgement: The ST view Advances in psychology, 54 (pp. 75-114). Amsterdam: Elsevier Science.

Brunswik, E. (1955). Representative design and probabilistic theory in functional psychology. Psychological Review, 62, 193-217.

Carkenord, D. M. & Stephens, M. G. (1994). Understanding student judgments of teaching effectiveness: A “policy capturing” approach. The Journal of Psychology, 128, 675-682.

Cooksey, R. W. (1996). Judgment analysis: Theory. methods, and applications. Academic Press: San Diego, CA.

Dawes, R. M. (1979). The robust beauty of improper linear models in decision making. American Psychologist, 34, 571-582.

Donnelly, L. & Bownas, D. A. (1984). Policy capturing: An approach to understanding personnel decision-making processes. Public Personnel Management, 13, 81-89.

Dovidio, L. F (1984). Helping behavior and altruism: An empirical and conceptual overview. In L. Berkowitz (Ed.) Advances in experimental social psychology (Vol. 17) (pp. 361-427). New York: Academic Press.

Dovidio. E, Piliavin, J. A., Gaertner, S., Schroeder, D. A. & Clark, R. D. (1991). The arousal: cost– reward model and the process of bystander intervention: A review of the evidence. In M. S. Clark (Ed.), Prosocial Behavior (pp. 86-118). Newbury Park, CA: Sage.

Eagly, A. H. & Crowley, M. (1986). Gender and helping behavior: A meta-analytic review of the social psychological literature. Psychological Bulletin, 100, 283-308.

Feather, N. T. (1992). An attributional and value analysis of deservingness in success and failure situations. British Journal of Social Psychology, 31, 125-145.

Finkelstein, M. A. & Brannick, M. T. (1997). Making decisions about sexual intercourse: Capturing college students’ policies. Basic and Applied Social Psychology, 19, 101-120.

Gaertner, S. L. & Dovidio, J. F. (1977). The subtlety of White racism, arousal, and helping behavior. Journal of Personality and Social Psychology, 35, 691-707.

Hammond, K. R. & Wascoe, N. E. (Eds.). (1980). New directions for methodology of social and behavioral science: No. 3. Realizations of Brunswik’s representative design. San Francisco: JosseyBass.

Harrison, P. D., Ryan, J. M. & Moore, P.S. (1996). College students’ self-insight and common implicit theories in rating of teaching effectiveness. Journal of Educational Psychology, 88, 775782.

Hobson, C. J., Mendel, R. M. & Gibson, F W. (1981). Clarifying performance appraisal criteria. Organizational Behavior and Human Decision Processes, 28, 164-188.

Hoffman, P J. (1960). The paramorphic representation of clinical judgement. Psychological Bulletin, 57, 116-131.

Keely, S. M. & Doherty, M. E. (1972). Bayesian and regression modeling of graduate admission policy. Organizational Behavior and Human Decision Processes, 8, 297-323.

LatanE, B. & Darley, J. M. (1968). Group inhibition of bystander intervention in emergencies. Journal of Personality and Social Psychology, 10, 215-221.

Latant, B. & Darley, J. M. (1970). The unresponsive bystander: Why doesn’t he help? New York: Appleton-Century Crofts.

Nisbett, R. E. & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84, 231-259.

Often, C. A., Penner, L. A. & Waugh, G. (1988). That’s what friends are for: The determinants of psychological helping. Journal of Social and Clinical Psychology, 34-41.

Pargament, K. I., Sullivan, M. S., Balzer, W. K., Van Haitsma, K. S. & Raymark, P H. (1995). The many meanings of religiousness: A policy-capturing approach. Journal of Personality, 63, 953983.

Pedhazur, E. J. (1997). Multiple regression in behavioral research: Explanation and prediction (3rd ed.) Fort Worth, TX: Harcourt Brace.

Penner, L. A., Fritzche, B. A., Craiger, J. P. & Freifeld, T. S. (1995). Measuring the prosocial personality. In J. N. Butcher, C. D. Spielberger, et al (Eds.), Advances in personality assessment (pp. 147-163). Hilldale, NJ: Erlbaum.

Penner, L. A., Dertke, M. C. & Achenbach, C. J. (1973). The “flash” system: A field study of altruism. Journal of Applied Social Psychology, 3, 362-370.

Piliavin, J. A., Dovidio, J. R, Gaertner, S. L. & Clark, R. D., III. (1981). Emergency intervention. New York: Academic Press.

Piliavin, I. M., Piliavin, J. A. & Rodin, J. (1975). Costs, diffusion, and the stigmatized victim. Journal of Personality and Social Psychology, 32, 429-438.

Piliavin, J. A. & Unger, R. K. (1985). The helpful, but not helpless female: Myth or reality? In V. O’ Leary, R. K. Unger, & B. S. Wallston (Eds.) Women, gender, and social psychology (pp. 149-186). Hillsdale, NJ: Erlbaum.

Pious, S. (1993). The psychology of judgement and decision making. Philadelphia: Temple UniverSity Press.

Ross, L. & Nisbett, R. E. (1991). The person and the situation: Perspectives of social psychology. Philadelphia: Temple University Press.

Schroeder, D. A., Penner L. A., Dovidio, J. F. & Piliavin. J. A. (1995). The psychology of helping and altruism: Problems and puzzles. New York: McGraw-Hill.

Slovic, P. & Lichtenstein, S. (1971). Comparison of Bayesian and regression approaches to the study of information processing in judgement. Organizational Behavior and Human Decision Processes. 6 649-744.

Stevenson, M. K., Busemeyer, J. R. & Naylor, J. C. (1990). Judgement and decision-making theory. In M. D. Dunnette, L. M. Hough, et al. (Eds.), Handbook of industrial and organizational psychology Vol. 1 (2nd ed., pp. 283-374). Palo Alto, CA: Consulting Psychologists Press, Inc.

Viswesvaran, C. & Barrick, M. R. (1992). Decision-making effects on compensation surveys: Implications for market wages. Journal of Applied Psychology, 77, 588-597.

Weiner, B. (1995). Judgments of responsibility: A foundation for a theory of social conduct. New York: Guilford.

Zedeck, S. & Kafry, D. (1977). Capturing rater policies for processing evaluation data. Organizational Behavior and Human Decision Processes, 18, 269-294.


University of Central Florida, USA


University of South Florida, USA

Barbara A. Fritzsche, University of Central Florida, USA; and Marcia A. Finkelstein and Louis A. Penner, University of South Florida, USA

The authors wish to thank reviewers including: Dr. Nancy C. Higgins, St Thomas University, Canada, Dr. Brenda Wood, University of Idaho and Dr. Bernard Weiner, University of California, Los Angeles. Please address correspondence and reprint requests to: Barbara A. Fritzsche, Department of Psychology, University of Central Florida, PO Box 161390, Orlando, FL 32816-1390. Email:

Copyright Society for Personality Research, Incorporated 2000

Provided by ProQuest Information and Learning Company. All rights Reserved