Journal of Experimental Social Psychology 82 (2019) 6–15
Contents lists available at ScienceDirect
Journal of Experimental Social Psychology journal homepage: www.elsevier.com/locate/jesp
Case Report
With a little help from my friends (and strangers): Closeness as a moderator of the underestimation-of-compliance effect
T
Sebastian Deria, , Daniel H. Steinb, Vanessa K. Bohnsc ⁎
a b c
G78 Uris Hall, Department of Psychology, College of Arts & Sciences, Cornell University, Ithaca, NY 14853, United States of America 2220 Piedmont Ave, Department of Management of Organizations, Haas School of Business, University of California, Berkeley, CA 94720, United States of America 394 Ives Hall, Department of Organizational Behavior, ILR School, Cornell University, Ithaca, NY 14853, United States of America
ARTICLE INFO
ABSTRACT
Handling editor: Shaul Shalvi
Those seeking help systematically underestimate the likelihood that strangers will help them (Bohns, 2016). However, it is not known whether this same error persists when requesting help from people with whom we interact regularly. In three experiments (the last of which was pre-registered), participants (N = 310) predicted the likelihood that either their friends or strangers would agree to a request for help. Participants then approached members of one of these two groups (i.e., friends or strangers) with this request (N = 953). We confirmed our predictions that (1) overall help-seekers would underestimate the likelihood that those they approached for help would agree to their requests and that (2) this underestimation error would be smaller for participants making requests of friends. We also found that (3) the underestimation effect persists even for those making requests of friends and (4) help-seekers expected the rate of helping between the two groups to vary more than it did. We discuss and test several mechanisms that might account for these effects. These findings suggest people may over-rely on their friends, and discount the role of strangers, when seeking help.
Keywords: Compliance Interpersonal closeness Help-seeking Social influence
Imagine you are short a few dollars at the grocery store, and you are considering asking someone you do not know (i.e., a stranger) in line for change. Do you have an accurate sense of how likely this person would be to help? Now imagine you are in this same situation, but you are considering making this request of someone you know well (i.e., a friend). Would you be any more accurate when predicting the likelihood your friend would help? Furthermore, do you have an accurate sense of how much more likely this friend would be to agree to help than a stranger? In the current research, we examine these questions, first testing whether individuals seeking help more accurately predict how much help they will receive from their friends, as opposed to strangers, when they directly request assistance. Second, we compare predictions about how much more likely friends are to agree to help than strangers to actual differences in compliance between these two groups. 1. Do help-seekers predict compliance more accurately for friends versus strangers? Previous research has found that help-seekers systematically underestimate the extent to which strangers comply with direct, face-to-
⁎
face requests for help (e.g., Bohns et al., 2011; Bohns, Newark, & Xu, 2016; Flynn & Lake, 2008; Roghanizad & Bohns, 2017). Help-seekers in these studies have asked > 14,000 potential helpers various requests, and the difference between help-seekers' predictions of compliance and actual compliance—the “underestimation-of-compliance effect”—is sizable (help-seekers underestimate compliance by 48%, on average; Bohns, 2016). However, the large number of potential helpers that help-seekers have approached in these studies has almost exclusively been strangers (Bohns, 2016). Thus, whether this underestimation-ofcompliance effect persists for requests made of friends remains an open question, and an important one, given the lion's share of requests for help in everyday life are made of people help-seekers know well (BarTal, Bar-Zohar, Greenberg, & Hermon, 1977). There are several reasons to expect help-seekers will be more accurate when predicting the compliance of friends versus strangers. First, help-seekers' underestimation of the likelihood that strangers will comply with their requests is primarily attributed to a perspectivetaking error made by help-seekers who fail to appreciate how difficult (i.e., awkward and uncomfortable) strangers would find rejecting a direct request for help (Bohns, 2016; Flynn & Lake, 2008). However,
Corresponding author. E-mail address:
[email protected] (S. Deri).
https://doi.org/10.1016/j.jesp.2018.11.002 Received 17 April 2018; Received in revised form 13 November 2018; Accepted 13 November 2018 0022-1031/ Published by Elsevier Inc.
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al.
people may have better insight into the minds and emotions of close others (Anderson, Keltner, & John, 2003; Colvin, Vogt, & Ickes, 1997; Funder & Colvin, 1998; Stinson & Ickes, 1992). People also have more experience both seeking and being asked for help from friends than strangers (Bar-Tal et al., 1977). Greater insight into the minds of friends and more experience requesting help from them should increase helpseekers' ability to take their perspectives and estimate their likelihood of agreeing to a request for help. Second, help-seekers likely expect their friends—with whom they repeatedly interact—have a substantially greater vested interest in helping them than strangers whom they may never meet again (Rand & Nowak, 2013). In general, people share stronger emotional and interpersonal bonds with close others than strangers (Aron, Aron, & Smollan, 1992), and people expect friends to empathize with and respond to their needs more than strangers (Shapiro, 1980). Thus, help-seekers' estimates of how likely friends would be to agree to a request for help should be higher than their estimates for strangers. Barring a large difference in the actual rates of compliance between friends and strangers (which is unlikely, as discussed below), this should mitigate the tendency to underestimate compliance for those making requests of friends.
whom one interacts repeatedly. However, people are often surprisingly willing to help and offer resources to those they do not know well or will never meet again (Andreoni & Miller, 2002; Henrich et al., 2001; Kahneman, Knetsch, & Thaler, 1986). In fact, there is direct evidence that people think self-interest plays a larger role in driving others' behavior than it does (Miller, 1999; Miller & Ratner, 1996). In one series of studies, participants thought a person's group membership would lead to greater support for issues in which the person had a vested interest than it did (Miller & Ratner, 1998). Similarly, help-seekers may expect provincial bonds of familiarity to play a larger role in others' decisions regarding whether to help than they do. Altogether, previous research suggests help-seekers will predict a larger difference between strangers' and friends' likelihood of agreeing to a request for help than should be apparent in the actual rates of compliance between the two groups. 3. Overview of studies In sum, we hypothesize (1) consistent with previous findings, helpseekers will underestimate the likelihood people will agree to their help requests (hypothesis 1), and (2) this underestimation will be smaller for those making requests of friends vs. strangers (hypothesis 2). Further, we explore whether (3) this underestimation-of-compliance holds specifically for those making requests of friends (exploratory research question 1), and (4) help-seekers expect the difference in the rates of helping between friends and strangers to be larger than the observed difference in helping between the two groups (exploratory research question 2). In three experiments, we examine these claims by having participants (N = 310) predict how much help they will receive from friends or strangers. After predicting compliance, participants leave the laboratory and actually make these requests of members of one of the two groups (N = 953), allowing us to compare expectations to reality. To examine the robustness of our findings, we use different help requests across studies, i.e., requests to fill out questionnaires (Study 1a; Study 2) or count beans in jars (Study 1b). Various measures aimed at examining the underlying perspective-taking failures predicted to account for these effects are explored (e.g., underestimating how uncomfortable helpers feel rejecting requests, overestimating helpers' concerns with self-interest). In all three experiments, we report how we determined sample sizes, and all data exclusions, manipulations, and measures (Simmons, Nelson, & Simonsohn, 2013). Measures not directly relevant to the current topic, as well as all data, code, and survey materials are available in the Open Science Framework (OSF) repository for this project (https://osf.io/8jy6r). Additional analyses are also conducted throughout the manuscript; these are summarized in the Supplementary Analyses (SA), available through the journal's online portal and in the OSF repository.
2. Do help-seekers know how much more likely friends are than strangers to agree to a help request? We have thus far argued that the gap between predicted and actual compliance will be smaller for those making requests of friends than strangers. However, there is another important difference to examine: the difference between the amount of help that help-seekers predict friends versus strangers will agree to and the amount of help that friends versus strangers actually grant. Help-seekers may predict friends are four times more likely to help than strangers when they are actually only two times more likely to help. In contrast, help-seekers may predict friends are two times more likely to help than strangers, and they may actually be two times more likely to help. In both cases, the gap between predicted and actual compliance might be smaller for requests made of friends than strangers. However, in the former case, help-seekers' predictions about the difference in helping behavior between friends versus strangers is systematically mis-calibrated, while in the latter case, it is not. Here we present several arguments for why the predicted difference will be larger than the actual difference in helping between friends and strangers. First, the psychological forces driving predictions of compliance differ meaningfully from the actual forces governing the behavior of friends and strangers confronted with a direct request for help. For help-seekers making predictions, one's relationship to a potential helper is likely to impact their expectations of how likely that person is to agree to a request for help (Buchsbaum, Seiver, Bridgers, & Gopnik, 2012). However, the actual behavior of individuals confronted with a request for help is likely to be impacted by more immediate, and universal, social norms and motivations. A primary concern when interacting with others is avoiding embarrassment or offense (Goffman, 1971; Sabini, Siepmann, & Stein, 2001). For example, people tend to display more trust towards strangers than they privately feel, which is partially the result of efforts to avoid offending others (Dunning, Anderson, Schlösser, Ehlebracht, & Fetchenhauer, 2014). Denying a help request is likely to be seen as an offense (Brown & Levinson, 1987), whether this request comes from a friend or stranger. Thus, a potential helper is unlikely to reject such a request—especially if it is relatively minor—regardless of his or her relationship to the help-seeker. Since these forces likely act strongly on both friends and strangers, they should push the compliance rates for both groups closer together. Second, help-seekers likely expect self-interest to play a larger role in determining whether potential helpers will agree to a help request than it does. From the perspective of maximizing self-interest, it makes sense that a stranger would be less willing to help than a friend with
4. Study 1a Studies 1a and 1b used a 2 × 2 mixed design, where the first factor varied between-subjects (target population: friends vs. strangers) and the second factor was a repeated measure within-subjects factor (rate of helping: predicted vs. actual). 5. Method 5.1. Participants We recruited 52 undergraduate primary participants (i.e., helpseekers; 35 female), 49 of whom returned for the second session. They, in turn, approached 179 secondary participants (i.e., potential helpers). One primary participant was excluded for making an impossible estimate (see Table S1 for information on exclusions). A sample size of 25 primary participants per condition has 80% power to detect 7
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al.
underestimation-of-compliance effects of d = 0.40 or greater. Underestimation-of-compliance effects are typically larger than this, e.g., d = 1.10 in Study 1 of Flynn and Bohns (2008), d = 1.00 in Study 1 of Roghanizad and Bohns (2017). In this experiment (and Studies 1b and 2), primary participants received $15.00 in two installments ($5 at the first lab session; $10 at the second lab session). Secondary participants were not paid.
# of people approached to get 3 to help
10
5.2. Procedure Primary participants completed the study in two sessions, which were exactly one week apart. In the first session, primary participants were told they would be making requests of others. Specifically, they needed to get three people (i.e., secondary participants) to do a simple task—complete a 22-item personality inventory—as well as answer several questions probing their feelings about and reasons for compliance (i.e., feelings of discomfort, pressure, and willingness to help; see OSF repository for exact items). Primary participants were assigned to make these requests of either friends (specifying these are people with whom they interact regularly and see at least once a week) or strangers (specifying these are people they do not know in any way). To ensure primary participants made requests the same way in both conditions, they were instructed to ask each person the following: “Will you do me a favor? Will you fill out this questionnaire?” Further, they were told to approach each person individually and in-person. Primary participants were then asked to estimate how many people they would need to approach to get three to agree to complete the task. This number constitutes our predicted compliance dependent variable. They then completed several questions meant to probe our mechanism (see Table S2 of Supplementary Analyses for exact items). Finally, they were given a tally sheet to record their interactions with every person they approached and to document whether each person agreed. In the second session, primary participants returned one week later with the three questionnaires completed by targets in sealed envelopes and their tally sheets, which reported how many people they had to approach—our actual compliance dependent variable. They then answered several other exploratory questions (see OSF repository) and were debriefed.
9
8.24
8 7 6 5
4.22
4
Predicted 4.12
Actual 3.28
3 2 1 0 Strangers
Friends
Fig. 1. Predicted and actual helping in Study 1a, by condition. For all graphs reported in this paper, error bars represent un-pooled standard errors in each cell.
t(22) = 6.95, p < .001, dav = 1.15, 95% CI = [0.65, 1.65]. Primary participants in the friend condition also predicted they would need to approach significantly more people (M = 4.12, SD = 1.62) than they had to (M = 3.28, SD = 0.61), t(24) = 3.46, p = .002, dav = 0.69, 95% CI = [0.19, 1.19]. This result addresses our first exploratory research question, suggesting the underestimation effect holds for requests of friends. Next, we examined whether the underestimation effect was smaller for those making requests of friends than strangers (our second hypothesis) and whether help-seekers expected the difference in rates of helping between friends and strangers to be larger than it was (our second exploratory research question). To examine these questions, we conducted a 2 (target population: friends vs. stranger) × 2 (rate of helping: predicted vs. actual) mixed-model ANOVA. There was a significant interaction between target population and helping, F(1, 46) = 27.27, p < .001, η2p = 0.37. The pattern of means in this interaction supported our second hypothesis and addressed our second exploratory research question. Supporting our second hypothesis, the difference between actual and predicted helping (the underestimationof-compliance effect) was smaller for those soliciting help from friends (Mpredicted = 4.2 v. Mactual = 3.3), t(47) = 1.99, p = .052,4 dz = 0.29, 95% CI = [−0.05, 0.63],5 than those soliciting help from strangers (Mpredicted = 8.0 v. Mactual = 4.2), t(47) = 9.14, p < .001, dz = 1.32, 95% CI = [0.95, 1.69]. In answer to our second exploratory research question, the difference in predicted helping (Mstrangers = 8.0 v. Mfriends = 4.2), t(47) = 4.45, p < .001, d = 1.29, 95% CI = [0.92, 1.66], was larger than the difference in actual helping (Mstrangers = 4.2 v. Mfriends = 3.3), t(47) = 1.88, p = .066, d = 0.54, 95% CI = [0.20, 0.88]. Help-seekers thought they would need to approach 90% more strangers than friends to get the same amount of help, when in fact they only had to approach 27% more people (Fig. 1). To test the robustness of our results to various assumptions in the models above, we also analyzed the data in several alternate ways. First, to deal with potential issues of non-normality, we log-transformed our measures (predicted and actual compliance) and re-ran all analyses
6. Results and discussion First, we tested whether primary participants underestimated the likelihood secondary participants would agree to their help requests (our first hypothesis).1 A paired samples t-test comparing predicted to actual compliance revealed that help-seekers predicted they would need to approach significantly more people (M = 6.09, SD = 3.79) than they had to (M = 3.73, SD = 1.77), t(47) = 6.22,2 p < .001, dav = 0.80, 95% CI = [0.45, 1.15],3 confirming our first hypothesis. Help-seekers underestimated the likelihood people would agree to their help requests. We also wanted to explore whether the underestimation effect held for each target population. Replicating prior work (Bohns, 2016), a paired samples t-test revealed help-seekers in the stranger condition predicted they would need to approach significantly more people (M = 8.24, SD = 4.31) than they had to (M = 4.22, SD = 2.41), 1 All analyses are done on participants who showed up to both sessions. A few other exclusions are made (e.g. a prediction of having to ask 0 people), all of which are noted in Table S1 (in the Supplementary Analyses document). 2 For the sake of simplicity and to maintain consistency with our later preregistration, we report this as a simple paired samples t-test across all participants (i.e. across those approaching friends and those approaching strangers). But, this can also be examined as the main effect of helping in 2 (helping: predicted v. actual) x 2 (target population: friends v. strangers) ANOVA. This too is significant, F(1, 46) = 63.66, p < .001, η2p = 0.58. 3 The code used for computing 95% Confidence Intervals for Cohen's d is made available in the OSF repository.
4 These simple slope analyses examining the underestimation of compliance effect in the friends and strangers conditions test the same hypotheses as the two sets of follow-up paired samples t-tests reported in the paragraph above. For the sake of completeness, we report the results of both analyses. Note that analyses do not reveal the exact same results (i.e. p-values) – as the standard errors are pooled differently in the two analyses (i.e. not at all for the t-tests and across all four cells in the simple slopes analyses). Nonetheless, these results are broadly consistent across the three studies. 5 All d's are reported in terms of absolute magnitude (i.e. unsigned). However, the 95% CI for the corresponding d may include zero, as here, to indicate when the CI overlaps zero.
8
9
Note Note Note Note Note Note Note Note
1. 2. 3. 4. 5. 6. 7. 8.
3.6 (1.3)
3.1 (1.5)
3.8 (1.4)
4.1 (1.4)
3.6 (0.9)
Mean (stranger condition)
2.4 (1.6)
3.7 (1.6)
4.8 (1.2)
5.4 (1.1)
4.5 (1.3)
Mean (friend condition)
t(46) = 2.88, p = .006
t(46) = −1.15, p = .257
t(46) = −2.43, p = .019
t(46) = −3.38, p = .001
t(46) = −2.94, p = .005
Comparison
0.43 [0.01, 0.85]*
−0.64 [−1.16, −0.11]* −0.86 [−1.29, −0.42]* −0.57 [−1.03, −0.11]* 0.09 [−0.31, 0.48]
Med. → DV (b path) −2.57 [−3.84, −1.30]* −2.13 [−3.23, −0.94]* −2.66 [−3.89, −1.43]* −3.23 [−4.48, −1.97]* −2.67 [−3.96, −1.38]*
IV → DV (c’ path)
−0.51 [−1.68, 0.01]
−0.62 [−1.30, −0.21]* −1.05 [−2.32, −0.30]* −0.52 [−1.59, −0.07] 0.05 [−0.12, 0.55]
Indirect effect
Individual mediation analysis
0.29 [−0.07, 0.65]
−0.71 [−1.12, −0.30]* −0.61 [−1.07, −0.15]* 0.45 [0.09, 0.80]*
−0.24 [−0.71, 0.23]
Med. → DV (b path) −1.41 [−2.56, −0.25]*
IV → DV (c’ path)
−0.35 [−1.18, 0.05]
−0.87 [−2.09, −0.24]* −0.56 [−1.57, −0.07]* 0.24 [−0.08, 0.97]
−0.23 [−0.82, 0.16]
Indirect effect
Simultaneous mediation analysis
−1.19 [−2.03, −0.36]*
0.53 [−0.40, 1.46]
0.91 [0.16, 1.67]*
1.23 [0.50, 1.96]*
0.97 [0.31–1.63]*
IV → med. (a path)
−3.18 [−4.41, −1.96]*
Total effect (c path)
Same for both individual and simultaneous mediation
All analyses are done with exclusions noted in Table S1. Condition (IV) always coded as: Stranger = 0, Friend = 1. Dependent variable (DV) is the extent of over-estimation (i.e. # of people participants predicted they'd need to approach – # of people participants actually had to approach). Mediator composites (Med.) are always calculated such that higher value signifies “more” of that construct (i.e. more difficulty saying no, more willingness to help, etc.) The total effect refers to the c-path, i.e. the coefficient for effect of condition (IV) on the extent of overestimation (DV) with no mediators or controls in the regression model. 95% confidence intervals for the indirect effects computed in PROCESS (see Hayes, 2013) using 10,000 bootstrapped resamples. For mediation analysis, coefficients and their corresponding 95% CI are given. * indicates p < .05, or in case of indirect effect simply that 95% CI doesn't include zero.
0.88
Discomfort asking
0.81
Feelings of obligation
0.84
0.88
Willingness to help
Understand position
0.58
Difficulty saying no
Study 1a
α
Construct
Study
Table 1 Mediation analyses for Study 1a.
S. Deri et al.
Journal of Experimental Social Psychology 82 (2019) 6–15
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al.
reported above on these log-transformed measures. The pattern and significance of all results reported above were replicated. Second, to deal with the possibility that our variables could be considered count data, we also ran two Poisson regressions—one with random intercepts for each participant, one without any random effects—where, mirroring the ANOVA above, target population (friends vs. stranger), rate of helping (predicted v. actual) and their interaction were entered in the model as predictors. Again, the pattern and significance of our findings were the same as above (under both models, with and without random effects). We report all additional analyses in full detail in the “Alternative Analyses of Main Findings” section of the Supplementary Analyses document. In addition, we have made all data and analysis scripts available in the OSF repository (https://osf.io/8jy6r). Lastly, results for five possible perspective-taking constructs that might mediate the difference in size of the underestimation effect between the friend and stranger conditions are shown in Table 1. Several of these constructs emerged as individual mediators (helpers' feelings of awkwardness), simultaneous mediators (helpers' feelings of obligation), or both (helpers' willingness to help). However, these results are interpreted with some caution because the study was powered to find typical underestimation-of-compliance effects but not for mediation analysis. We explore mechanisms with a well-powered experiment in Study 2.
# of people approached to get 2 to help
10 9 8 6.84 7 6 5 4
Predicted
4.19
Actual
3.36 2.65
3 2 1 0 Strangers
Friends
Fig. 2. Predicted and actual helping in Study 1b, by condition.
9. Results and discussion We first tested whether primary participants underestimated the likelihood secondary participants would agree to their help requests.6 Supporting our first hypothesis, a paired samples t-test comparing predicted to actual compliance revealed help-seekers predicted they would need to approach significantly more people (M = 5.49, SD = 3.45) than they had to (M = 3.00, SD = 1.43), t(50) = 5.24,7 p < .001, dav = 0.94, 95% CI = [0.60, 1.28]. Again, this underestimation effect held for each target population. A paired samples t-test revealed help-seekers in the stranger condition predicted they would need to approach significantly more people (M = 6.84, SD = 4.01) than they had to (M = 3.36, SD = 1.73), t(24) = 4.17, p < .001, dav = 1.13, 95% CI = [0.63, 1.63]. Likewise, primary participants in the friend condition predicted they would need to approach significantly more people (M = 4.19, SD = 2.19) than they had to (M = 2.65, SD = 0.98), t(25) = 3.73, p = .001, dav = 0.91 95% CI = [0.43, 1.39]. Once again, the underestimation-of-compliance effect held for requests made of friends (our first exploratory research question). To test our remaining hypothesis and exploratory research question, we conducted a 2 (target population: friends vs. stranger) × 2 (rate of helping: predicted vs. actual) mixed-model ANOVA. There was a significant interaction between target population and helping, F(1, 49) = 4.45, p = .040, η2p = 0.083. The pattern of means in this interaction supported our second hypothesis and addressed our second exploratory research question. Supporting our second hypothesis, the difference between actual and predicted helping (the underestimationof-compliance effect) was smaller for those soliciting help from friends (Mpredicted = 4.19 v. Mactual = 2.65), t(50) = 2.39, p = .021, dz = 0.33, 95% CI = [0.00, 0.66] than those soliciting help from strangers (Mpredicted = 6.84 v. Mactual = 3.36), t(50) = 5.30, p < .001, dz = 0.74, 95% CI = [0.40, 1.08]. Addressing our second exploratory research question, the difference in predicted helping (Mstrangers = 6.84 v. Mfriends = 4.19), t(50) = 2.94, p = .005, d = 0.82, 95% CI = [0.48, 1.16], was larger than the difference in actual helping (Mstrangers = 3.36 v. Mfriends = 2.65), t(50) = 1.81, p = .077, d = 0.51, 95% CI = [0.18, 0.84]. Help-seekers thought they would need to approach 63% more strangers than friends to get the same amount of help, when in fact they only had to approach 27% more people (Fig. 2). To test for robustness, we re-analyzed these data under the same three alternative models used for the re-analysis of Study 1a. These
7. Study 1b We made several slight adjustments in Study 1b to examine the robustness of our effect. Namely, the help request was changed to ensure our effect was not specific to one type of request. Further, we addressed a potential concern that primary participants in Study 1a were approaching friends in different contexts (e.g., relaxing in one's dorm) than strangers (e.g., rushing by on campus). As much as possible, we wanted to ensure differences in compliance between friends and strangers were due to targets' relationship to participants and not other factors, such as the location or circumstances under which they were approached. 8. Method 8.1. Participants We again aimed to recruit 25 primary participants per betweensubjects condition (50 total). A sample of this size gives us 80% power to detect underestimation-of-compliance effects of d = 0.40 or greater. Primary participants were recruited from dormitories housing high school students attending a university summer program. Our final sample included 54 primary participants (31 females, Mdn age = 17), 52 of whom returned for the second session (the exclusion of a participant whose responses were 5 SDs above the mean is noted in Table S1). These participants approached 153 secondary participants. 8.2. Procedure The procedures and measures in Study 1b were the same as in Study 1a, with exceptions now noted. First, we changed the request task for generalizability. Participants were given a jar of 600–650 pinto beans and asked targets to count the number of beans in the jar (literal beancounting). Since this task took longer for both primary and secondary participants to complete, primary participants were asked to get two secondary participants (instead of three in Study 1a) to complete it. Second, we modified the task instructions to control for the context in which primary participants approached secondary participants. Specifically, primary participants (all of whom were living in a university dormitory) were instructed to only make requests in their dormitory.
6
As in Study 1a, all analyses are done on participants who showed up to both sessions and all exclusions are noted in Table S1. 7 Again, this can be examined as the main effect of helping in 2 (helping: predicted v. actual) × 2 (target population: friends v. strangers) ANOVA. This too is significant, F(1, 49) = 29.75, p < .001, η2p = 0.378. 10
11
0.71
0.86
Discomfort asking
Perceived difficulty
4.1 (1.6)
3.4 (1.5)
3.8 (1.3)
4.2 (1.1)
3.6 (1.1)
Mean (stranger condition)
4.4 (1.4)
3.5 (1.3)
3.5 (1.5)
4.3 (1.2)
4.1 (1.1)
Mean (friend condition)
t(49) = −0.63, p = .532
t(49) = −0.22, p = .823
t(49) = 0.84, p = .405
t(49) = −0.27, p = .791
t(49) = −1.61, p = .114
Comparison
0.52 [−0.10, 1.14]
0.61 [−0.05, 1.26]
−0.16 [−0.86, 0.54]
−0.64 [−1.44, 0.16]
−0.34 [−1.19, 0.51]
Med. → DV (b path)
−1.89 [−3.71, −0.06]* −1.94 [−3.87, −0.11]* −2.00 [−3.80, −0.19]* −2.08 [−3.90, −0.26]*
−1.77 [−3.68, 0.13]
IV → DV (c’ path) −0.17 [−1.11, 0.13] −0.06 [−0.80, 0.31] 0.05 [−0.12, 0.64] 0.05 [−0.40, 0.74] 0.14 [−0.22, 0.98]
Indirect effect
Individual mediation analysis
0.31 [−0.38, 0.99]
0.37 [−0.38, 1.11]
0.02 [−0.88, 092]
−0.45 [−1.30, 0.40]
−0.28 [−1.39, 0.83]
Med. → DV (b path) −1.87 [−3.88, 0.15]
IV → DV (c’ path)
−0.14 [−1.34, 0.33] −0.04 [−0.67, 0.23] −0.01 [−0.61, 0.39] 0.03 [−0.24, 0.71] 0.08 [−0.15, 0.94]
Indirect effect
Simultaneous mediation analysis
0.27 [−0.58, 1.11]
0.09 [−0.70, 0.88]
−0.32 [−1.08, 0.44]
0.09 [−0.57, 0.74]
0.50 [−0.13, 1.13]
IV → med. (a path)
−1.94 [−3.79, −0.09]*
Total effect (c path)
Same for both individual and simultaneous mediation
All analyses are done with exclusions noted in Table S1. Condition (IV) always coded as: Stranger = 0, Friend = 1. Dependent variable (DV) is the extent of over-estimation (i.e. # of people participants predicted they'd need to approach – # of people participants actually had to approach). Mediator composites (Med.) are always calculate such that higher value signify “more” of that construct (i.e. more difficulty saying no, more willingness to help, etc.) 95% confidence intervals for the indirect effects computed in PROCESS (see Hayes, 2013) using 10,000 bootstrapped resamples. For mediation analysis, coefficients and their corresponding 95% CI are given. * indicates p < .05, or in case of indirect effect simply that 95% CI doesn't include zero.
0.81
Feelings of obligation
1. 2. 3. 4. 5. 6. 7.
0.54
Willingness to help
Note Note Note Note Note Note Note
0.59
Difficulty saying no
Study 1b
α
Construct
Study
Table 2 Mediation analyses for Study 1b.
S. Deri et al.
Journal of Experimental Social Psychology 82 (2019) 6–15
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al.
results were highly consistent—all results were the same for the log analysis, and three of the four primary findings were the same for two Poisson regressions. The one result that did not reach standard levels of significance was the interaction between target population and rate of helping. This may have been due to the sample size, an issue we address in the next study. These analyses are reported in full detail in the “Alternative Analyses of Main Findings” section of the Supplementary Analyses document; data and analysis scripts are available in the OSF repository (https://osf.io/8jy6r). Mediating mechanism results are shown in Table 2, although here no consistent mediators emerged; the same cautions are noted regarding this analysis as in Study 1a.
# of people approached to get 3 to help
11 10
9.39
9 8 7 6
Predicted
5 4
3.81
Actual
3.88 3.07
3 2 1 0 Strangers
10. Study 2
Friends
Fig. 3. Predicted and actual helping in Study 2, by condition.
In Study 2, we aimed to increase the credibility of our findings by replicating them in a pre-registered study (https://aspredicted.org/ fq8hw.pdf). In addition, we collected a larger sample to conduct a wellpowered mediation analysis. As with the previous studies, Study 2 used a 2 (target population: friends vs. strangers) x 2 (rate of helping: predicted vs. actual) mixed design.
(M = 3.81, SD = 1.80), t(162) = 6.65,9 p < .001, dav = 0.67, 95% CI = [0.48, 0.86]. This underestimation effect again held for each target population. A paired samples t-test revealed help-seekers in the stranger condition predicted they would need to approach significantly more people (M = 9.39, SD = 6.97) than they had to (M = 4.54, SD = 2.31), t(81) = 6.30, p < .001, dav = 0.93, 95% CI = [0.66, 1.20]. Likewise, primary participants in the friend condition predicted they would need to approach significantly more people (M = 3.88, SD = 1.59) than they had to (M = 3.07, SD = 0.26), t(80) = 4.54, p < .001, dav = 0.70, 95% CI = [0.43, 0.97]. The underestimation effect again held for requests of friends (our first exploratory research question). To test our remaining hypothesis and exploratory research question, we conducted a 2 (helping: predicted vs. actual) × 2 (target population: friends vs. stranger) mixed-model ANOVA. There was a significant interaction between helping and target population, F(1, 161) = 25.96, p < .001, η2p = 0.139. The pattern of means in this interaction again supported our second hypothesis and addressed our second exploratory research question. Supporting our second hypothesis, the difference between predicted and actual helping (the underestimation-of-compliance effect) was smaller for those soliciting help from friends (Mpredicted = 3.88 v. Mactual = 3.07), t(162) = 1.42, p = .157, dz = 0.11, 95% CI = [−0.07, 0.29], than for those soliciting help from strangers (Mpredicted = 9.39 v. Mactual = 4.54), t(162) = 8.67, p < .001, dz = 0.68, 95% CI = [0.49, 0.87]. Answering our second exploratory research question, the difference in predicted helping (Mstrangers = 9.39 v. Mfriends = 3.88), t(162) = 6.94, p < .001, d = 1.09, 95% CI = [0.90, 1.28], was larger than the difference in actual helping (Mstrangers = 4.54 v. Mfriends = 3.07), t(162) = 5.67, p < .001, d = 0.89, 95% CI = [0.70, 1.08]. Help-seekers thought they would need to approach 142% more strangers than friends to get the same amount of help, when in fact they only had to approach 48% more people (Fig. 3). To test for robustness, we re-analyzed these data under the same three alternative models used for the re-analysis of Studies 1a–1b. These results were highly consistent—all results were the same for the log analysis as well as for the Poisson regression without any random effects; three of the four primary analyses were the same for the Poisson regression with random intercepts, with the only exception that the difference between predicted and actual compliance for those making requests of friends did not reach standard levels of significance. These analyses are reported in full detail in the “Alternative Analyses of Main Findings” section of the Supplementary Analyses document; data and analysis scripts are available in the OSF repository (https://osf.io/ 8jy6r).
11. Method 11.1. Participants We aimed for 100 primary participants per between-subjects condition (200 total), doubling the heuristic suggested by Simmons et al. (2013) for detecting medium-size effects. Ultimately, we recruited 204 primary participants (127 females, Mdn age = 19), 168 of whom returned for the second session (82.3%). These participants approached 621 secondary participants. Accounting for dropouts, this sample has 80% power to detect underestimation-of-compliance effects of d = 0.22. 11.2. Procedure The procedures and measures in this study were the same as in Study 1a, with the exceptions now noted. First, primary participants had one day between sessions, rather than one week, due to availability of lab space. Second, the mediation indices for primary participants were altered to measure not only perspective-taking failures but also help-seekers' expectations about structural factors that might vary between requests of friends and strangers. The first two indices aimed to capture the perspective-taking failures we theorized to be most central: inferences about (1) helpers' discomfort saying “no” and (2) helpers' willingness versus reluctance to help (i.e., the role of helpers' self-interests). The other two indices aimed to measure help-seekers' expectations about situational and structural factors that might vary between interactions with friends and strangers: (3) perceived social norm violations (Burgoon & Hale, 1988) and (4) expected reciprocity (Clark & Mils, 1993). (See Table S2 for exact items.) 12. Results and discussion We first tested whether primary participants underestimated the likelihood secondary participants would agree to their help requests.8 Supporting our first hypothesis, consistent with the first two studies, and as specified in our pre-registration (https://aspredicted.org/fq8hw. pdf), a paired samples t-test comparing predicted to actual compliance revealed help-seekers predicted they would need to approach significantly more people (M = 6.65, SD = 5.76) than they had to
9 Again, this can be examined as the main effect of helping in 2 (helping: predicted v. actual) × 2 (target population: friends v. strangers) ANOVA. This too is significant, F(1, 161) = 50.61, p < .001, η2p = 0.239.
8
As in Studies 1a-1b, all analyses are done on participants who showed up to both sessions and all exclusions are noted in Table S1. 12
Note Note Note Note Note Note Note Note
1. 2. 3. 4. 5. 6. 7. 8.
13
0.92
Reciprocity
3.3 (1.9)
3.6 (1.5)
4.1 (1.1)
3.4 (1.1)
Mean (stranger condition)
5.8 (1.2)
2.9 (1.3)
5.2 (1.1)
3.7 (1.1)
Mean (friend condition)
t(160) = −10.27, p < .001
t(160) = 3.34, p = .001
t(160) = −6.35, p < .001
t(160) = −1.71, p = .090
Comparison
−0.56 [−1.06, −0.06]*
−1.27 [−1.95, −0.58]* −1.22 [−1.90, −0.54]* 0.19 [−0.37, 0.75]
Med. → DV (b path) −3.66 [−5.20, −2.12]* −2.67 [−4.38, −0.96]* −3.90 [−5.54, −2.26]* −2.63 [−4.64, −0.62]*
IV → DV (c’ path)
−1.41 [−3.17, −0.22]*
−0.38 [−1.37, −0.01]* −1.37 [−3.89, −0.38]* −0.14 [−0.69, 0.28]
Indirect effect
Individual mediation analysis
−0.22 [−0.72, 0.29]
−1.01 [−1.71, −0.31]* −0.90 [−1.62, −0.19]* 0.11 [−0.43, 0.65]
Med. → DV (b path) −2.10 [−4.10, −0.09]*
IV → DV (c’ path)
−0.55 [−1.74, 0.51]
−0.30 [−1.06, 0.00]* −1.01 [−2.67, −0.12]* −0.08 [−0.58, 0.36]
Indirect effect
Simultaneous mediation analysis
2.54 [2.05, 3.02]*
−0.74 [−1.18, −0.30]*
1.12 [0.77, 1.47]*
0.30 [−0.05, 0.65]
IV → Med. (a path)
−4.04 [−5.62, −2.46]*
Total effect (c path)
Same for both individual and simultaneous mediation
All analyses are done with exclusions noted in Table S1. Condition (IV) always coded as: Stranger = 0, Friend = 1. Dependent variable (DV) is the extent of over-estimation (i.e. # of people participants predicted they'd need to approach – # of people participants actually had to approach). Mediator composites (Med.) are always calculate such that higher value signify “more” of that construct (i.e. more difficulty saying no, more willingness to help, etc.) Condition (IV) always coded as: Stranger = 0, Friend = 1. 95% confidence intervals for the indirect effects computed in PROCESS (see Hayes, 2013) using 10,000 bootstrapped resamples. For mediation analysis, coefficients and their corresponding 95% CI are given. * indicates p < .05, or in case of indirect effect simply that 95% CI doesn't include zero.
0.83
0.74
Willingness to help
Norm violation
0.44
Difficulty saying no
Study 2
α
Construct
Study
Table 3 Mediation analyses for Study 2.
S. Deri et al.
Journal of Experimental Social Psychology 82 (2019) 6–15
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al.
To examine what might account for the fact that underestimation was smaller for friends than strangers, we conducted both individual and simultaneous mediation analyses on our four proposed constructs. The two most consistent mediating constructs were those measuring perspective-taking failures (see Table 3). Inferences about willingness to help showed a significant indirect effect in both individual and simultaneous mediation. In addition, inferred awkwardness had a significant indirect effect and mediated in both the individual and simultaneous mediation analyses, although its b-path was just below standard levels of significance. Overall, the results from a well-powered and pre-registered experiment are consistent with the results from the previous two.
larger requests than those used in our experiments, the difference in compliance between intimates and strangers may be larger, a difference that requesters may be more or less accurately attuned to. Before dismissing as obvious that compliance would diverge to a greater extent in the case of larger requests, one might take note of the many instances in which others are willing to pay an extraordinary price to help those they do not know at all—donating kidneys to complete strangers, sacrificing huge portions of their income, or committing to raising their children (Anonymous, 2017; MacFarquhar, 2015, 2016). In his book Touching Strangers, Richard Renaldi (2014) has people who do not know each other pose together as intimates. The images are striking because of how easily total strangers can pass for people who have known each other their whole lives—as father and daughter, sister and brother, and so on. Yet, perhaps, our surprise at such photos is unjustified. As our results suggest, it seems in strangers we have a friend.
13. General discussion Across three experiments, we found help-seekers underestimate the extent to which both strangers and friends will agree to direct requests for help. Thus, the underestimation-of-compliance effect, previously documented only for strangers (Bohns, 2016), persists for friends, although the effect is smaller. Furthermore, although help-seekers are more accurate at predicting how much friends (compared to strangers) will help them, they expect the difference in helping between these two groups to be larger than it is. In Studies 1a, 1b, and 2, primary participants expected friends to agree to help 90%, 63%, and 142% more than strangers, respectively, when in fact they did agree to help more, but only at rates of 27%, 27%, and 48%. We also present some evidence that these results involve perspective-taking failures on the part of helpseekers, who anticipate self-interest will play a larger role driving the behavior of others, and who may also make mistaken inferences about how awkward it would be for helpers to reject requests for help. More broadly, these findings corroborate several other lines of research related to influence and helping behavior. They lend additional credence to research on the strength of “weak ties,” which has shown others with whom we are only minimally acquainted can influence our lives in meaningful ways. In addition to helping us find a job (Granovetter, 1973, 1974), and exerting a measurable impact on our well-being (Epley & Schroeder, 2014; Sandstrom & Dunn, 2014a, 2014b), weak ties are resources help-seekers can tap into when seeking help. Help-seekers underappreciate just how similar these weaker connections (strangers) are to our closer connections (friends) regarding the likelihood they will help us out when we are in need. These findings also bear on the social support literature. Perceptions of support availability are associated with greater well-being and adjustment to stressful life events (Bolger, Zuckerman, & Kessler, 2000; House, Landis, & Umberson, 1988). However, the actual provision of support within close relationships does not always have positive effects, and sometimes even results in greater distress (Bolger et al., 2000; Gleason, Iida, Shrout, & Bolger, 2008), for reasons that are poorly understood (McClure et al., 2014). The present research suggests perceptions of social support availability may be chronically mis-calibrated, leading people to believe they have less access to help, should they need it, than they do. Further, our research identifies an alternative, underutilized outlet for support-seeking: seeking help from strangers. Receiving support from strangers may not bear the same costs as accepting support from close others. While accepting help from a stranger can be awkward, this is typically an acute, transitory form of distress. Accepting support from a relationship partner, on the other hand, risks disrupting the “supportive equity” of one's relationship, creating feelings of dependency that may result in more significant, enduring personal distress (Gleason, Iida, Bolger, & Shrout, 2003). Future research in this domain should explore the generalizability of these results, particularly with regard to the effect of request magnitude. While our research suggests people underestimate their breadth of support resources (i.e., number of supporters), we have not touched on perceptions of depth (i.e., amount of support from each supporter; Armstrong & Kammrath, 2015; Newark, Bohns, & Flynn, 2017). For
Open practices Materials and data for the studies are available at https://osf.io/ 8jy6r. The pre-registration for Study 2 is available at: https:// aspredicted.org/fq8hw.pdf. Additional analyses are reported in detail in the Supplementary Analyses document linked in the online version of the article and the OSF project repository (https://osf.io/8jy6r). Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.jesp.2018.11.002. References Anderson, C., Keltner, D., & John, O. P. (2003). Emotional convergence between people over time. Journal of Personality and Social Psychology, 84(5), 1054–1068. https://doi. org/10.1037/0022-3514.84.5.1054. Andreoni, J., & Miller, J. (2002). Giving according to GARP: An experimental test of the consistency of preferences for altruism. Econometrica, 70(2), 737–753. Anonymous (2017, May 20). Why I donated one of my kidneys to a stranger. The guardian. Retrieved from https://www.theguardian.com/society/2017/may/20/why-idecided-to-donate-one-of-my-kidneys-to-a-stranger. Armstrong, B. F., III, & Kammrath, L. K. (2015). Depth and breadth tactics in support seeking. Social Psychological and Personality Science, 6(1), 39–46. Aron, A., Aron, E. N., & Smollan, D. (1992). Inclusion of other in the self scale and the structure of interpersonal closeness. Journal of Personality and Social Psychology, 63(4), 596. Bar-Tal, D., Bar-Zohar, Y., Greenberg, M. S., & Hermon, M. (1977). Reciprocity behavior in the relationship between donor and recipient and between harm-doer and victim. Sociometry, 293–298. Bohns, V. K. (2016). (Mis) understanding our influence over others: A review of the underestimation-of-compliance effect. Current Directions in Psychological Science, 25(2), 119–123. Bohns, V. K., Handgraaf, M. J., Sun, J., Aaldering, H., Mao, C., & Logg, J. (2011). Are social prediction errors universal? Predicting compliance with a direct request across cultures. Journal of Experimental Social Psychology, 47(3), 676–680. Bohns, V. K., Newark, D., & Xu, A. (2016). For a dollar, would you…? How (we think) money influences compliance with our requests. Organizational Behavior and Human Decision Processes, 134, 45–62. Bolger, N., Zuckerman, A., & Kessler, R. C. (2000). Invisible support and adjustment to stress. Journal of Personality and Social Psychology, 79(6), 953. Brown, P., & Levinson, S. (1987). Politeness: Some universals in language usage. Cambridge, England: Cambridge University Press. Buchsbaum, D., Seiver, E., Bridgers, S., & Gopnik, A. (2012). Learning about causes from people and about people as causes: probabilistic models and social causal reasoning. Advances in Child Development, 125. Burgoon, J. K., & Hale, J. L. (1988). Nonverbal expectancy violations: Model elaboration and application to immediacy behaviors. Communication Monographs, 55(1), 58–79. Clark, M. S., & Mils, J. (1993). The difference between communal and exchange relationships: What it is and is not. Personality and Social Psychology Bulletin, 19(6), 684–691. Colvin, C. R., Vogt, D. S., & Ickes, W. (1997). Why do friends understand each other better than strangers do? In W. Ickes (Ed.). Empathic accuracy (pp. 169–193). New York: Guilford Press. Dunning, D., Anderson, J. E., Schlösser, T., Ehlebracht, D., & Fetchenhauer, D. (2014). Trust at zero acquaintance: More a matter of respect than expectation of reward. Journal of Personality and Social Psychology, 107(1), 122. Epley, N., & Schroeder, J. (2014). Mistakenly seeking solitude. Journal of Experimental
14
Journal of Experimental Social Psychology 82 (2019) 6–15
S. Deri et al. Psychology: General, 143(5), 1980. Flynn, F. J., & Lake, V. K. B. (2008). If you need help, just ask: Underestimating compliance with direct requests for help. Journal of Personality and Social Psychology, 95, 128–143. Funder, D. C., & Colvin, C. R. (1988). Friends and strangers: Acquaintanceship, agreement, and the accuracy of personality judgment. Journal of Personality and Social Psychology, 55(1), 149–158. https://doi.org/10.1037/0022-3514.55.1.149. Gleason, M. E., Iida, M., Bolger, N., & Shrout, P. E. (2003). Daily supportive equity in close relationships. Personality and Social Psychology Bulletin, 29(8), 1036–1045. Gleason, M. E., Iida, M., Shrout, P. E., & Bolger, N. (2008). Receiving support as a mixed blessing: Evidence for dual effects of support on psychological outcomes. Journal of Personality and Social Psychology, 94(5), 824. Goffman, E. (1971). Relations in public. Harmondsworth, England: Penguin. Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360–1380. Granovetter, M. S. (1974). Getting a job: A study of contacts and careers. Cambridge, mass. Harvard University Press. Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., & McElreath, R. (2001). In search of homo economicus: Behavioral experiments in 15 small-scale societies. The American Economic Review, 91(2), 73–78. House, J. S., Landis, K. R., & Umberson, D. (1988). Social relationships and health. Science, 241, 540–545. Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1986). Fairness and the assumptions of economics. Journal of Business, S285–S300. MacFarquhar, L. (2015, September 22). Extreme altruism: Should you care for strangers at the expense of your family? The Guardian. Retrieved from https://www. theguardian.com/world/2015/sep/22/extreme-altruism-should-you-care-forstrangers-as-much-as-family. MacFarquhar, L. (2016). Strangers drowning: Impossible idealism, drastic choices, and the urge to help. Penguin.
McClure, M. J., Xu, J. H., Craw, J. P., Lane, S. P., Bolger, N., & Shrout, P. E. (2014). Understanding the costs of support transactions in daily life. Journal of Personality, 82(6), 563–574. Miller, D. T. (1999). The norm of self-interest. American Psychologist, 54(12), 1053–1060. Miller, D. T., & Ratner, R. K. (1996). The power of the myth of self-interest. Current societal concerns about justice (pp. 25–48). US: Springer. Miller, D. T., & Ratner, R. K. (1998). The disparity between the actual and assumed power of self-interest. Journal of Personality and Social Psychology, 74(1), 53–62. Newark, D. A., Bohns, V. K., & Flynn, F. J. (2017). The value of a helping hand: Helpseekers' predictions of help quality. Organizational Behavior and Human Decision Processes, 139, 18–29. Rand, D. G., & Nowak, M. A. (2013). Human cooperation. Trends in Cognitive Sciences, 17(8), 413–425. Renaldi, R. (2014). Richard Renaldi: Touching strangers. Aperture. Roghanizad, M. M., & Bohns, V. K. (2017). Ask in person: You're less persuasive than you think over email. Journal of Experimental Social Psychology, 69, 223–226. Sabini, J., Siepmann, M., & Stein, J. (2001). The really fundamental attribution error in social psychological research. Psychological Inquiry, 12, 1–15. Sandstrom, G. M., & Dunn, E. W. (2014a). Is efficiency overrated? Minimal social interactions lead to belonging and positive affect. Social Psychological and Personality Science, 5(4), 437–442. Sandstrom, G. M., & Dunn, E. W. (2014b). Social interactions and well-being: The surprising power of weak ties. Personality and Social Psychology Bulletin, 40(7), 910–922. Shapiro, E. G. (1980). Is seeking help from a friend like seeking help from a stranger? Social Psychology Quarterly, 259–263. Simmons, J., Nelson, L., & Simonsohn, U. (2013, February). Life after p-hacking. Talk presented at annual meeting of the society of personality and social psychology in New Orleans, LA. Stinson, L., & Ickes, W. (1992). Empathic accuracy in the interactions of male friends versus male strangers. Journal of Personality and Social Psychology, 62(5), 787.
15