RevisitingtheStanfordprisonexperimentagainExaminingdemandcharacteristicsintheguardorientation.pdf

Full Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=vsoc20

The Journal of Social Psychology

ISSN: 0022-4545 (Print) 1940-1183 (Online) Journal homepage: https://www.tandfonline.com/loi/vsoc20

Revisiting the Stanford prison experiment, again:Examining demand characteristics in the guardorientation

Jared Bartels

To cite this article: Jared Bartels (2019) Revisiting the Stanford prison experiment, again:Examining demand characteristics in the guard orientation, The Journal of Social Psychology,159:6, 780-790, DOI: 10.1080/00224545.2019.1596058

To link to this article: https://doi.org/10.1080/00224545.2019.1596058

Published online: 08 Apr 2019.

Submit your article to this journal

Article views: 12166

View related articles

View Crossmark data

Citing articles: 3 View citing articles

This article has been awarded the Centrefor Open Science 'Open Data' badge.

This article has been awarded the Centrefor Open Science 'Open Materials' badge.

Revisiting the Stanford prison experiment, again: Examiningdemand characteristics in the guard orientationJared Bartels

William Jewell College

ABSTRACTThe day before the Stanford prison experiment began, the investigators held anorientation session for the guards in which they communicated expectations forhostile guard behavior, a flippant prisoner mindset, and the possibility of endingthe study prematurely. While the study’s principal investigator has minimizedthe influence of this orientation, critics have speculated that it provided a “script”for guard abuse. In the present studies, participants were presented witha hypothetical prison simulation study and randomly assigned as guards to anorientation session that included these expectations (Stanford orientation) orone providing basic study information. Across three studies, participantsexposed to the Stanford orientation relative to a control orientation, reportedgreater expectations for hostile and oppressive behavior on the part of thestudy’s investigator and from others and themselves as guards. The presentresults provide empirical support for speculation that the language of the guardorientation in the Stanford prison experiment sanctioned abuse among guards.

ARTICLE HISTORYReceived 1 August 2018Accepted 13 March 2019

KEYWORDSDemand characteristics;imprisonment; P.G.Zimbardo; social psychology;Stanford prison experiment

In the Stanford prison experiment, Zimbardo and colleagues (Haney, Banks, & Zimbardo, 1973;Zimbardo, 2007) randomly assigned 21 college-aged men to the role of guard or prisoner in a mockprison. Conflict between the prisoners and guards emerged on the second day of the study andescalated until it was prematurely ended on day six. The authors attributed the abusive behavioramong some guards and psychological breakdown among some prisoners to the power of situationalforces, among them an excessive power differential inherent in the roles of guards and prisoners.Guard aggression, according to Haney et al. (1973), was “a ‘natural’ consequence of being in theuniform of a ‘guard’ and asserting the power inherent in that role” (p. 92). Critics, however, havechallenged this account and the methodological and theoretical soundness of the study (Banuazizi &Movahedi, 1975; Banyard, 2007; Carnahan &McFarland, 2007; Fromm, 1973; Haslam&Reicher, 2007,2017; Haslam, Reicher, & Van Bavel, 2018; McFarland & Carnahan, 2009; Reicher & Haslam, 2006).

In 2001 Reicher and Haslam conducted the British Broadcasting Company (BBC) prison study(Reicher & Haslam, 2006). Like the Stanford prison experiment, BBC study participants wererandomly assigned the role of prisoner or guard in an institution with psychological dynamics(i.e., power and status asymmetry) mimicking a real prison. However, the BBC prison study differedfrom the Stanford prison experiment in several key ways. Unlike the Stanford prison experiment,Reicher and Haslam tested a priori propositions of a theory, social identity theory (Tajfel & Turner,1979), throughout the course of the study. In stark contrast to the Stanford prison experiment,guards in the BBC study were not abusive and prisoners, rather than exhibiting despondency,became increasingly noncompliant with guard directives over the course of the study. The resultsof the BBC study suggested that role adoption, rather than being a natural consequence of thesituation, is determined by the extent to which participants identify with a group. In short, prisoners

CONTACT Jared Bartels [email protected] Department of Psychological Science, William Jewell College, 500College Hill, 120 Jewell Hall, Liberty, MO 64068-1896© 2019 Taylor & Francis

THE JOURNAL OF SOCIAL PSYCHOLOGY2019, VOL. 159, NO. 6, 780–790https://doi.org/10.1080/00224545.2019.1596058

coalesced as a group and guards did not. The opposite pattern appears likely in the Stanford prisonexperiment, as identification with the guard role was actively encouraged by the experimentersthroughout the study, beginning with a guard orientation session prior to the start of the study(Haslam & Reicher, 2007; Haslam, Reicher, & Van Bavel, 2018). While Reicher and Haslam (2006)provided no instructions for guard behavior, demand characteristics via the Stanford prison experi-ment guard orientation may, in part, account for the stark behavioral differences between Stanfordprison and BBC prison study guards.

In his seminal paper on demand characteristics, Orne (1962) speculated that, under mostcircumstances, participants are motivated to be “good subjects” or confirm the experimenter’shypothesis because they value science and want to contribute. Research indicates that figuring outthe point of the study or the experimenter’s hypothesis (Sawyer, 1975), knowledge of being observed(Kiley, 1974; see McCambridge, de Bruin, & Witton, 2012), and suggestive communication from theexperimenter (e.g., Navarick, 2004, 2007; Nichols & Maner, 2008; Stern, Saayman, & Touyz, 1978),influence the behavior of research participants. With respect to the latter, Orne (1962) conjecturedthat implicit suggestions would be more likely to elicit hypothesis-consistent responding comparedto explicit communication. However, Navarick (2004, 2007) found that openly communicating theexperimenter’s preference for a participant’s response and that a certain participant response wouldhelp the researchers with data analysis, produced responding in the advocated direction. Stern et al.(1978) utilized more subtle demand characteristics informing participants in a nocturnal dreamstudy that previous testing indicated a tendency to dream about a particular type of locality (e.g.,urban or outdoor-nature). Analysis of dream journals indicated dream content consistent with thesuggestion.

Critics have long speculated about the presence of demand characteristics in the Stanford prisonexperiment (Banuazizi & Movahedi, 1975; Banyard, 2007; Haslam & Reicher, 2007; Haslam &Reicher, 2012, 2017). Banuazizi and Movahedi (1975) surveyed 150 college students in the Bostonarea about their expectations regarding a prison simulation study. Students were provided informa-tion similar to that given to Stanford prison experiment participants, including the original adsoliciting participants, a description of the rights and privileges one would have to give up asa participant (e.g., stay in prison for one to two weeks; consent to be under surveillance), and anaccount of the arrest and booking procedures (e.g., participants are arrested, blindfolded, and sent tothe prison). After giving participants the above information, Banuazizi and Movahedi probed forawareness of the experimenter’s expectations regarding the outcome and asked participants abouttheir expectations regarding guard behavior. Results indicated that 81% were able to deduce thepoint of the experiment, and nearly 90% predicted hostile and oppressive behavior on the part of theguards (this dropped to roughly 50% when participants were asked how they would respond asguards).

Decades after the Banuazizi and Movahedi (1975) publication, critics would highlight anotherpotential source of demand characteristics in the Stanford prison experiment, the guard orientation.The day before the study began, the principal investigator held an orientation session for the guards,in which he, as prison Superintendent, explained the purpose of the study, what they were able to doas guards, and what to expect in terms of prisoner behavior. The guards were told the following:

We can create boredom. We can create a sense of frustration. We can create fear in them, to some degree. Wecan create a notion of the arbitrariness that governs their lives, which are totally controlled by us, by the system,by you, me. … They’ll have no privacy at all, there will be constant surveillance–nothing they do will gounobserved. They will have no freedom of action. They will be able to do nothing and say nothing that we don’tpermit. We’re going to take away their individuality in various ways. … In general, what all this should create inthem is a sense of powerlessness. We have total power in the situation. They have none. (Zimbardo,2007, p. 55).

Haslam and Reicher (2007) note that the above instructions amount to “details of how to oppress theprisoners” (p. 618). It is not only the guidance itself, but the manner in which it is delivered that isproblematic. Note the references to “we” in the above quote. The effect, according to Haslam and

THE JOURNAL OF SOCIAL PSYCHOLOGY 781

Reicher (2007), is the “invocation and management of a shared social identity” (p. 618). Suggestivelanguage is evident later in the orientation session as well as guards were informed that prisoners arenot going to take the study seriously (i.e., Prisoners think this is all going to be fun and games…)that the study might not last two weeks (i.e., We don’t know how long it’s going to run…) and thatthe investigators want the study to have an impact (i.e., We’d like it [the study] to have a real impactfrom the beginning…; Zimbardo, 1971). The content of guard orientation raises doubt about theextent to which the Stanford prison experiment guards were writing “their own scripts on the blankcanvass of the SPE [Stanford prison experiment]” as opposed to operating according to a “script ofterror” (Banyard, 2007, p. 494).

There is no direct reference to demand characteristics in The Lucifer Effect (Zimbardo, 2007),a book detailing the events of the Stanford prison experiment. However, the author notes that theguard orientation was meant to get the guards “into the mood of the joint” (p. 55). While he doesacknowledge that initially “some of the guards’ reactions were probably influenced by their orienta-tion” (p. 216), he concludes that, with or without the orientation, the situational forces wouldproduce the same outcome.

The only study to examine demand characteristics in the Stanford prison experiment was theaforementioned survey by Banuazizi and Movahedi (1975). While these authors focused on demandcharacteristics in the procedural details of the study given to Stanford prison experiment participants,the present studies examine demand characteristics in the debated guard orientation. Specifically, thedisputably suggestive language in the orientation was varied to examine its potential to influence guardresponding in a simulated prison study. Both Orne (1962) and Sawyer (1975) discuss several meth-odologies for detecting the influence of demand characteristics including experimentally manipulatingthese characteristics while controlling for experimental variables, asking participants to respond toa simulation of the experiment and probing for experimenter expectations after such a simulation.Both elements were utilized in the present studies. In light of the results obtained by Banuazizi andMovahedi (1975), the numerous suggestions by critics that the orientation session sanctioned guardabuse, and the results obtained in the BBC prison study in which both an orientation session for guardsand guard abuse were absent, it was expected that an orientation session fashioned after the onedelivered in the Stanford prison experiment would produce greater negative and less positive responsesrelative to one providing basic study information.

Study 1

Method

ParticipantsParticipation was solicited through a crowdsourcing service in the United Kingdom.1 Participants(N = 128) were paid €1.10 for participation. In order to approximate the SPE sample, participationwas limited to males (Mage = 32.07, SD = 11.00). Ethnicity for the sample was as follows: AfricanAmerican = 2, Caucasian = 97, East Asian = 10, Latino(a) = 6, Middle Eastern = 1, Multiracial = 1,Other = 11. This sample size reached the required N = 64 in each group for power of .80 at theα = .05 level (Cohen, 1992).

ProcedureLike Banuazizi and Movahedi (1975), participants were given basic information about a hypotheticalprison study, information given to participants in the Stanford prison experiment (http://pdf.prisonexp.org/geninfo.pdf). Specifically, participants were told that a simulated prison was being estab-lished to study a number of problems of psychological and sociological relevance. Paid volunteerswould be randomly assigned to play the roles of either prisoners or guards for the duration of thestudy. The time period would vary somewhat from about 5 days to 2 weeks for any one volunteer.Participants were also given information about the ad used to recruit participants.

782 J. BARTELS

Upon consent, participants were informed that they were assigned the role of guards and wouldbe asked to listen to the orientation session. Next, participants were randomly assigned to one of twoorientation sessions. The experimental orientation, henceforth referred to as the Stanford orienta-tion, was modeled after the one delivered by investigators in the Stanford prison experiment (theorientation was transcribed from audio available from the Stanford archives; https://purl.stanford.edu/fx266hb9381). Specifically, the Stanford orientation included expectations for guard behavior(i.e., “We can create boredom, we can create a sense of frustration we can create a sense of fear inthem to some degree”), expectations for prisoner behavior (i.e., “prisoners think this is all going to befun and games”), and expectations for the course of the study (i.e., “If it looks like it’s too serious, wemight have to end earlier; we want this study to have a real impact”), and the desire for the study tohave an impact (i.e., “We’d like it to have a real impact from the beginning and not just be some sillyass study”). In the control condition, participants listened to an orientation that included only basicstudy material (also provided to participants in the Stanford prison experiment), including informa-tion about provisions for prisoners (i.e., “Food and accommodations for the prisoners will beprovided which will meet minimal standard nutrition, health and sanitation requirements”) andhousing of guards and prisoners (i.e., “You guards, along with the warden, will be housed in cellblocks adjacent to prisoner cells with meals and bedding provided for you separately”).

Scripts for both orientations were roughly 220 words in length. A video was created, whichincluded a recording of the script created by a commercially available text-to-speech converter witha standard voice. Both videos included an image of a conference room in which the hypotheticalorientation session was held, subtitles, and were roughly 90 seconds in length. Participants wereasked to pay close attention to the orientation session, as they would be asked subsequent questionsabout the information provided. After viewing the video, participants were asked how they expectedguards to behave toward prisoners in the study, how they would behave toward prisoners as a guard,and the principal investigator’s expectations for guard behavior. Participants were asked to indicatetheir expectations for hostile and oppressive as well as friendly, fair, and lenient guard behavior usinga 7-point Likert scale. These categories were drawn from results of Banuazizi and Movahedi (1975).Hostile and oppressive responses were significantly correlated (r = .74, .77, .55, p < .01) within eachscenario (expectations of guards, yourself as guard, experimenter), as were friendly, fair, and hostileresponses (r ranged from .22 to .70, p ≤ .01). Thus, negative (hostile and oppressive) and positive(friendly, fair, and lenient) variables were utilized in the subsequent analysis. Lastly, participantswere asked if they recognized the study and, if so, to name it. Data from participants who were ableto identify the hypothetical study as the Stanford prison experiment was eliminated.

Results

In order to examine differences in expectations for guard behavior between the two orientationconditions, a one-way ANOVA was conducted. Prior to conducting the ANOVA, the assumption ofhomogeneity of variance was evaluated and satisfied, and Levene’s F test ranged from .23 to 2.21(p from .14 to .64). The ANOVA results, summarized in Table 1, indicate that those in the Stanfordorientation relative to the control orientation expected more negative (M = 5.24, SD = 1.17to M = 3.54, SD = 1.35, Hedges’ g = 1.36) and less positive (M = 2.70, SD = 1.21 to M = 3.85,SD = 1.05, Hedges’ g = 1.01) behavior from the others as guards. Similarly, participants expectedgreater negative (M = 3.73, SD = 1.46 to M = 2.86, SD = 1.23, Hedges’ g = 0.64) and less positive(M = 3.70, SD = 1.18 to M = 4.52, SD = .97, Hedges’ g = 0.75) responses from themselves as guardswhen assigned to the Stanford orientation relative to the control orientation. Additionally, resultsindicated that participants in the Stanford orientation compared to the control orientation held thatthe experimenter expected greater negative (M = 5.21, SD = 1.26 to M = 3.60, SD = 1.38, Hedges’g = 1.23) and less positive behavior (M = 2.61, SD = 1.34 to M = 3.86, SD = 1.11, Hedges’ g = 1.00)emitted from guards.2

THE JOURNAL OF SOCIAL PSYCHOLOGY 783

Discussion

Results from study 1 replicate and extend the findings of Banuazizi and Movahedi (1975). Based onthe procedural information given to participants in the Stanford prison experiment, those respond-ing to the Banuazizi and Movahedi survey expected hostile and oppressive behavior from othersplaying the role of guards and to a lesser extent, themselves in the role of guards. In the presentstudy, results indicated that participants ostensibly assigned to the role of guards in a simulatedprison study expressed different expectations for hostile and oppressive guard behavior dependingon the nature of the orientation session. Those exposed to an orientation modeled on the originalStanford prison experiment guard orientation indicated greater expectations for hostile and oppres-sive guard behavior relative to those in an orientation where basic study information was given (i.e.,absent demand characteristics). These results speak to concerns raised among critics about thedemand characteristics inherent in the guard orientation (e.g., Reicher & Haslam, 2006). Haslamand Reicher (2007) note that the guard orientation not only provides a scheme for guard oppression,but uses language (e.g., “We can create boredom, we can create a sense of frustration we…”) thatfosters a sense of a shared identity and shared mission between himself and the guards. The presentresults suggest that the language of the guard orientation in the Stanford prison experiment mayevoke hostile/oppressive expectations.

Study 2

Study 1 established that the evocative language in the guard orientation of a simulated prison studyinfluenced expectations for positive and negative guard behavior. Study 2 represents an attempt toreplicate the findings in study 1 with a sample that more closely approximates the Stanford prisonexperiment participants. Thus, in study 2, participation was limited to male university students.

Method

ParticipantsParticipation was solicited through a crowdsourcing service in the United Kingdom. Participants (N= 149; Mage = 22.40, SD = 4.40) were paid €0.77 for completing the study. Ethnicity for the samplewas as follows: Caucasian = 102, East Asian = 6, Latino(a) = 12, Middle Eastern = 13, Multiracial = 7,Other = 9. This sample size reached the required N = 64 in each group for power of .80 at the α = .05level (Cohen, 1992).

Table 1. One-way ANOVA results with expectations for guard behavior as the criterion.

Study 1 Study 2

MSPE(SD)a MCon(SD)

b F g 95% CI MSPE(SD)c MCon(SD)

d F g 95% CI

Guard(other)Negative 5.24(1.17) 3.54(1.35) 58.37** 1.36 [0.97,1.75] 5.06(1.21) 3.82(1.32) 35.61** 0.98 [0.64,1.32]Positive 2.70(1.21) 3.85(1.05) 31.59** 1.01 [0.64,1.38] 2.93(1.10) 3.73(1.08) 19.76** 0.73 [0.40,1.07]Guard(self)Negative 3.73(1.46) 2.86(1.23) 12.91** 0.64 [0.28,1.00] 3.74(1.28) 2.99(1.29) 12.55** 0.58 [0.26,0.91]Positive 3.70(1.18) 4.52(0.97) 18.00** 0.75 [0.39,1.11] 4.05(1.04) 4.45(1.09) 5.21* 0.38 [0.07,0.70]InvestigatorNegative 5.21(1.26) 3.60(1.38) 47.51** 1.23 [0.85,1.61] 5.00(1.46) 3.60(1.17) 41.90** 1.05 [0.71,1.40]Positive 2.61(1.34) 3.86(1.11) 31.52** 1.00 [0.63,1.37] 2.98(1.36) 3.97(0.94) 26.07**e 0.84 [0.50,1.18]

MSPE = Mean for Stanford group, MCon = Mean for control group; g = Hedges’ g; Guard (other) = expectations for others as guards;Guard (self) = expectations for oneself as guard; Investigator = expectations of experimenter; Negative (hostile and oppressive);Positive (friendly, fair, and lenient); CI = confidence interval.

an = 72. bn = 56. cn = 78. dn = 71. eBrown-Forsythe = 26.95***p < .05, **p < .01.

784 J. BARTELS

ProcedureThe procedure was the same as in study 1.

Results

As in study 1, a one-way ANOVA was conducted. Prior to conducting the ANOVA, the assumption ofhomogeneity of variance was evaluated and satisfied in all but one case (i.e., investigator-positive; Levene’sF = 20.88, p < .013), Levene’s F test ranged from .01 to 4.81 (ps from .01 to .94). The ANOVA results,summarized in Table 1, indicate that those in the Stanford orientation, relative to the control orientation,expected more negative (M = 5.06, SD = 1.21 to M = 3.82, SD = 1.32, Hedges’ g = 0.98) and less positive(M = 2.93, SD = 1.10 toM = 3.73, SD = 1.08, Hedges’ g = 0.73) behavior from others as guards. Participantsalso expected greater negative (M = 3.74, SD = 1.28 to M = 2.99, SD = 1.28, Hedges’ g = 0.58) and lesspositive (M = 4.05, SD = 1.04 toM = 4.45, SD = 1.09, Hedges’ g = 0.38) responses from themselves as guardswhen assigned to the Stanford orientation relative to the control orientation. Lastly, results indicated thatparticipants in the Stanford orientation compared to the control orientation held that the experimenterexpected greater negative (M = 5.00, SD = 1.46 toM = 3.60, SD = 1.17, Hedges’ g = 1.05) and less positivebehavior (M = 2.98, SD = 1.36 toM = 3.97, SD = 0.94, Hedges’ g = 0.84) from guards.

Discussion

Study 2 replicated the results of study 1 with a sample that more closely approximated that of the Stanfordprison experiment. While the present results do not provide evidence that Stanford prison guards wereresponding to demand characteristics, they provide strong support, along with results from Banuazizi andMovahedi (1975), that demand characteristics were present in the Stanford prison study. Thus, theysubstantiate concerns raised among critics about the suggestive language of the guard orientation (e.g.,Haslam & Reicher, 2017). It should be noted that in study 2 (as in study 1), effect sizes were in the small tomedium range (Cohen, 1992) when participants were asked to indicate how they would respond as guards,while effects were in themedium to large range when participants indicated how others would act as guardsand about the experimenter’s expectations. While these results may appear to weaken the case for demandcharacteristics, they do so only if one mistakenly assumes that the present studies are an attempt to accountfor behavior itself (whether in the Stanford prison experiment or a contemporary prison simulation study)rather than demand characteristics per se. Moreover, the discrepancy between others as guards and oneselfas a guard, a discrepancy that emerged in the Banuazizi and Movahedi (1975) study as well, should beinterpreted in light of the well-established better-than-average effect or tendency to rate oneself as morefavorable relative to others (e.g., Brown, 2012).

Study 3

Studies 1 and 2 indicated that a guard orientation modeled after the Stanford prison experimentorientation influenced the expectations for positive and negative guard behavior among participantsresponding to a hypothetical prison study. However, the possibility remains that the crucial phrases inthe orientation would not have accounted for any hostile or oppressive expectations beyond thatproduced by the situational variables identified by Haney et al. (1973); see also Zimbardo, 2007) ascausally related to the pathological behavior among prisoners and guards. Thus, study 3 sought to controlfor the experimentally relevant independent variables while manipulating demand characteristics;a method recommended by Sawyer (1975). Since the primary contention over the guard orientationhas been the script for negative guard behavior, study 3 focused exclusively on negative (hostile andoppressive) guard behavior. Additionally, study 3 presented participants with an ambiguous open-endedscenario involving guard-prisoner interaction. Essentially, this study represents an attempt to eliminatethe influence of the guard orientation by providing participants with additional information thataccentuates the situational forces presumed to be influential in the Stanford prison experiment.

THE JOURNAL OF SOCIAL PSYCHOLOGY 785

Method

ParticipantsLike study 2, participation was limited to male university students and was solicited througha crowdsourcing service in the UK. Participants (N = 108) were paid €0.88 for participation.Demographic information was as follows: Mage = 24.77, SD = 6.10; ethnicity – Caucasian = 59,East Asian = 16, Latino(a) = 12, African American = 9, Middle Eastern = 1, Multiracial = 4,Other = 7. Data from those who correctly identified the study as the Stanford prison experimentand data from those who failed to correctly answer the manipulation check question was eliminatedfrom subsequent analysis. The sample approached the required N = 64 in each group for power of.80 at the α = .05 level (Cohen, 1992).

ProcedureThe procedure was the same as in study 1, with two exceptions. First, participants in study 3 wereprovided with the following additional information prior to the guard orientation:

Prisoners will wear a loosely fitting muslin smock (i.e., something that resembles a dress). Theprisoner uniforms will have an identification number on the front and the back. Prisoners will referto one another and prison staff will refer to prisoners using this number. Prisoners will have a chainand lock around an ankle, will wear rubber sandals and will wear a nylon stocking cap on their head.Prisoners will be housed in barred cells. Guards will wear khaki shirts and pants and reflectingsunglasses and will carry a whistle and nightstick or baton.

After reading the above information, participants “attended” the orientation session andwere then presented with a scenario involving guard-prisoner interaction similar to an inter-action that occurred on day 2 of the Stanford prison experiment (i.e., Guards woke prisonersup in the morning for breakfast and to review prison rules. While reviewing the rules one ofthe prisoners is laughing and smiling). Participants were asked to indicate three ways in whichthey might respond to the situation and then responded to a question that served as manip-ulation check for their assigned condition (i.e., participants were asked to identify a phrase thatwas included in their orientation session). As in study 1, participants were also asked about theextent to which the principle investigator expected hostile and oppressive behavior among theguards.

Results

Approximating the methods of Banuazizi and Movahedi (1975), two research assistants, blind to thehypothesis and assigned condition of participants, tabulated the number of hostile/oppressiveresponses. The raw agreement between the raters was 80.86% and Cohen’s κ = 60.29 (95% CI,51.47, 69.11) suggested moderate agreement (McHugh, 2012). Scores from the two raters wereaveraged. An independent samples t-test revealed greater hostile and oppressive responses to theprisoner provocation among those in the Stanford orientation (M = 1.53, SD = 0.97, N = 68) relativeto the control orientation (M = 0.69, SD = 0.69, N = 40), t(106) = 4.81, p < .001. Hedges’ g wasestimated at 0.96, 95% CI [0.55, 1.37]. Lastly, results revealed greater principle investigator expecta-tions for hostile/oppressive guard responses among participants in the Stanford orientation(M = 5.06, SD = 1.27, N = 68) relative to the control orientation (M = 4.15, SD = 1.14, N = 40), t(106) = 3.73, p < .001. Hedges’ g = 0.74, 95% CI [0.34, 1.15].

Discussion

Study 3 represents an extension of studies 1 and 2 with an exclusive focus on the expected behaviorof participants as guards and hostile and oppressive conduct. Results indicate that the language of anorientation session for guards in a simulated prison study influences participants’ expectations for

786 J. BARTELS

hostile and oppressive guard responding. Specifically, an orientation session modeled after theStanford prison experiment orientation elicited greater hostile and oppressive responses (e.g.,“scream at him [prisoner] in front of everybody”) to a prison scenario than an orientation sessionproviding basic study information. The guard orientation made a unique contribution to hostile andoppressive guard responding beyond the presumed operative situational variables from the Stanfordprison experiment. Consistent with studies 1 and 2, the experimental group in study 3 indicated thatthe principal investigator had greater expectations for hostile and oppressive guard behaviorcompared to those in the control orientation.

General discussion

No post factum study, including the present ones, can account for the behavior of prison guards in theStanford prison experiment.What the present studies establish is that the guard orientation is imbued withdemand characteristics. In studies 1 and 2, participants expected greater hostile and oppressive responseswhen exposed to a guard orientation in which the principal investigator conveyed expectations for guardbehavior, prisonermentality, and the course and impact of the study (i.e., modeled after the Stanford prisonexperiment) compared to one in which demand characteristics were minimized by providing only basicstudy information. Study 3 represented a more rigorous test of the influence of demand characteristics onexpectations for guard behavior as: (1) participants responded to an open-ended scenario involving anencounter between a prisoner and themselves (2) guard were provided additional study information that,by itself, may have accounted for expectations, and (3) participants were asked to indicate how they wouldrespond as guards rather than how others would respond (which elicited greater negative responses instudies 1 and 2). Yet, consistent with the results of studies 1 and 2, those in the Stanford orientationindicated a readiness to respond to a prisoner with greater hostility and oppressiveness than those in thecontrol orientation. The present results extend the findings of Banuazizi and Movahedi (1975), who foundthat college students with no knowledge of the Stanford prison experiment expected hostile and oppressivebehavior among guards when given the study information Stanford participants received.

The familiar narrative of the Stanford prison experiment (e.g., Haney et al., 1973) is as follows:Over the course of only a few days, ordinary college students randomly assigned the role of guards ina simulated prison study slipped into the role of “tough” guards, displaying increasingly oppressiveand hostile behavior towards fellow college students randomly assigned the role of prisoners. Theanonymity, depersonalization, and the power asymmetry inherent in the roles of prisoner and guardwere the powerful situational forces assumed responsible for the pathological behavior. However, ascritics have noted and the present results corroborate, the role of the research participant is asplausible an explanation for the guard behavior as the social role of prison guard.

An individual assuming the role of research participant is often eager to ascertain the purpose of thestudy and contribute to science by responding in a manner that confirms the experimenter’s hypothesis(Orne, 1962). Of course, not all research participants are predisposed to act as “good subjects” and theextent to which they serve as one may depend on a number of factors including the experimenter (e.g., thereputation of the experimenter), the information provided study recruits, and communication during theexperiment including conveying the scientific importance of the study to participants (Orne, 1962; Orne &Whitehouse, 2000; Silverman & Shulman, 1970). The most salient communication among the “totality ofcues” (Orne, 1962) in the Stanford prison experiment may be the guard orientation. During the orientationsession, prior to the arrest and booking of prisoners, guards were told what behavior was expected ofprisoners and guards (Zimbardo, 2007). Specifically, guards were told that they could create fear andboredom, that prisoners will not take the study seriously, that the study may have to end early, and that theexperimenters wanted the study to have an impact. Previous research suggests that communicating theimportance of the study, beyond the disclosure of the hypothesis alone, can increase the acquiescence todemand characteristics among study participants (Navarick, 2007). The orientation also included language,particularly the use of “we” and “they” that instilled a sense of shared mission on the part of the guards and

THE JOURNAL OF SOCIAL PSYCHOLOGY 787

the superintendent (Haslam et al., 2018). The orientation session, critics charge, set the stage for guardabuse as the language provided a script for guard oppression (e.g., Haslam & Reicher, 2012).

It is important to note that in spite of the demand characteristics present in the guard orientation,not all of the guards in the Stanford prison experiment exhibited abusive behavior. Stanford guardsapparently fell within three categories: good, bad, and by the book (Zimbardo, 2007). Haney et al.(1973) note that “a few [guards] were passive and rarely instigated any coercive control over theprisoners” (p. 81). So striking was this lack of complicity in the presence of strong demandcharacteristics, Fromm (1973) noted that while the authors of the study believed it proved that thesituation was capable of producing ruthless sadists in a matter of days, it “proves, if anything, ratherthe contrary” (p. 81). There are experimental circumstances that may produce responding that runscounter to demand characteristics including situations in which the desire to protect one’s self-imageor project a socially desirable image of oneself is stronger than demand characteristics (Silverman &Shulman, 1970). Additionally, demand discrepant behavior may be attributed to the overly explicitnature of the demands. Replications of the Milgram obedience studies have found a lack ofobedience among participants when told by the experimenter that they have no choice but tocontinue with the study (Haslam, Reicher, & Millard, 2015). While research participants may bedisposed to want to contribute to science, this command is not consistent with one’s schema forproper science.

The present results are intriguing given recent revelation from analyses of Stanford prisonexperiment archives (Le Texier, in press). These analyses call into question the standard narrativeof the study that guards were creative in their evil machinations, guided only by the presence ofpowerful roles and their own imaginations. Rather, tough guard behavior was encouraged through-out the study and acts such as waking prisoners in the middle of the night for counts appears to havebeen influenced by explicit suggestions if not instructions from experimenters rather than represent-ing manifestations of creative evil (Haslam, Reicher, & Van Bavel, in press; Le Texier, in press).Haslam et al. (in press) concluded:

Guards in the SPE [Stanford prison experiment] were not left to their own devices when it cameto making decisions about how to behave and run the prison. On the contrary, they were subjectedto active leadership from the experimenters.

The Banuazizi study established that demand characteristics were present in study protocol, thepresent studies indicate that demand characteristics were present in the guard orientation and recentarchival analyses reveal that demand characteristics continued beyond the guard orientation.Perhaps Stanford prison experiment critic and textbook author Peter Gray (2013) summarized itbest when he noted, “In this experiment, the demands are everywhere” (para. 9).

Notes

1. The materials that support the findings of both studies are openly available in Open Science Framework athttps://osf.io/rdfm3/ (DOI 10.17605/OSF.IO/RDFM3). There is sufficient information for an independentresearcher to reproduce the reported results.

2. The results reported here for study 1 included participants that indicated they recognized the study but couldnot name it. In order to ensure that inclusion of these individuals was not accounting for the differencesbetween the two groups, the analyses were re-run with these participants eliminated (N = 108). This reanalysisproduced similar results with all comparisons statistically significant. For example, ANOVA results indicatedthat those in the Stanford orientation expected greater negative responses from others as guards (F = 48.01,p < .01), from themselves as guards (F = 13.38, p < .01), and believed the experimenter expected greater negativeresponses (F = 45.11, p < .01) relative to the control condition. Thus, participants who indicated that theyrecognized the study but were unable to name it were included in subsequent analyses.

3. Brown-Forsythe statistic is reported in Table 1.

Disclosure statement

No potential conflict of interest was reported by the author.

788 J. BARTELS

Data availability statement

The data described in this article are openly available in the Open Science Framework at https://osf.io/rdfm3/ (DOI10.17605/OSF.IO/RDFM3).

Open Scholarship

This article has earned the Center for Open science badges for Open Data and Open Materials through Open PracticesDisclosure. The data and materials are openly accessible at https://osf.io/rdfm3/ (DOI 10.17605/OSF.IO/RDFM3).

Notes on contributor

Jared Bartels is an Assistant Professor of Psychological Science at William Jewell College. His research interestsinclude achievement motivation and the history of psychology.

References

Banuazizi, A., & Movahedi, S. (1975). Interpersonal dynamics in a simulated prison. American Psychologist, 30,152–160. doi:10.1037/h0076835

Banyard, P. (2007). Tyranny and the tyrant: FROM STANFORD to Abu Ghraib. [Review of the book The Lucifer effect:Understanding how good people turn evil, by P. G. Zimbardo]. The Psychologist, 20, 494–495.

Brown, J. D. (2012). Understanding the better than average effect: Motives (still) matter. Personality and SocialPsychology Bulletin, 38, 209–219. doi:10.1177/0146167211432763

Carnahan, T., & McFarland, S. (2007). Revisiting the Stanford prison experiment: Could participant self-selection haveled to the cruelty? Personality and Social Psychology Bulletin, 33, 603–614. doi:10.1177/0146167206292689

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155–159.Fromm, E. (1973). The anatomy of human destructiveness. New York, NY: Henry Holt & Company.Gray, P. (2013, October 18). Why Zimbardo’s prison experiment isn’t in my textbook [Web log comment]. Retrieved

from http://www.psychologytoday.com/blog/freedom-learn/201310/why-zimbardo-s-prison-experiment-isn-t-in-my-textbook

Haney, C., Banks, C., & Zimbardo, P. (1973). Interpersonal dynamics in a simulated prison. International Journal ofCriminology and Penology, 1, 69–97.

Haslam, S. A., & Reicher, S. (2007). Beyond the banality of evil: Three dynamics of an interactionist social psychologyof tyranny. Personality and Social Psychology Bulletin, 33, 615–622. doi:10.1177/0146167206298570

Haslam, S. A., & Reicher, S. (2012). Contesting the “nature” of conformity. What Milgram and Zimbardo's studiesreally show. Plos Biology, 10, e1001426.

Haslam, S. A., & Reicher, S. (2017). Tyranny: Revisiting Zimbardo’s Stanford prison experiment. In J. R. Smith &S. A. Haslam (Eds.), Social psychology: Revisiting the classic studies (2nd ed., pp. 130–145). Thousand Oaks, CA: Sage.

Haslam, S. A., Reicher, S., & Millard, K. (2015). Shock treatment: Using immersive digital realism to restage and re-examine Milgram’s “obedience to authority” research. PloS one, 10, e109015. doi:10.1371/journal.pone.0109015

Haslam, S. A., Reicher, S., & Van Bavel, J. J. (in press). Brutality is not a product of ‘natural’ conformity to role:Uncovering the role of identity leadership in the Stanford Prison Experiment. The American Psychologist.

Haslam, S. A., Reicher, S. D., & Van Bavel, J. J. (2018, June 30). Doing ill for 'the greater good': Understanding whatreally went on in the Stanford prison experiment. The Inquisitive Mind. Retrieved from http://www.in-mind.org/blog/post/doing-ill-for-the-greater-good-understanding-what-really-went-on-in-the-stanford-prison

Kiley, D. E. (1974). The effect of the research expectancy on the activity level of institutionalized delinquents. Journalof Community Psychology, 2, 242–247. doi:10.1002/(ISSN)1520-6629

Le Texier, T. (in press). Debunking the Stanford prison experiment. The American Psychologist.McCambridge, J., de Bruin, M., & Witton, J. (2012). The effects of demand characteristics on research participant

behaviours in non-laboratory settings: A systematic review. PloS one, 7, e39116. doi:10.1371/journal.pone.0039116McFarland, S., & Carnahan, T. (2009). A situation’s first powers are attracting volunteers and selecting participants:

A reply to Haney and Zimbardo. Personality and Social Psychology Bulletin, 35, 815–818. doi:10.1177/0146167209334781

McHugh, M. L. (2012). Interrater reliability: The kappa statistic. Biochemia Medica, 22, 276–282.

THE JOURNAL OF SOCIAL PSYCHOLOGY 789

Navarick, D. J. (2004). Analysis of impulsive choice: Assessing effects of implicit instructions. The Psychological Record,54, 505–522. doi:10.1007/BF03395489

Navarick, D. J. (2007). Attenuation and enhancement of compliance with experimental demand characteristics. ThePsychological Record, 57, 501–515. doi:10.1007/BF03395590

Nichols, A. L., & Maner, J. K. (2008). The good-subject effect: Investigating participant demand characteristics. Journalof General Psychology, 135, 151–165. doi:10.3200/GENP.135.2.151-166

Orne, M. T. (1962). On the social psychology of the psychological experiment: With particular reference to demandcharacteristics and their implications. American Psychologist, 17, 776–783. doi:10.1037/h0043424

Orne, M. T., & Whitehouse, W. G. (2000). Demand characteristics. In A. E. Kazdin (Ed.), Encyclopedia of psychology(Vol. 2, pp. 469–470). Washington, DC: American Psychological Association.

Reicher, S., & Haslam, S. A. (2006). Rethinking the psychology of tyranny: The BBC prison study. British Journal ofSocial Psychology, 45, 1–40. doi:10.1348/014466605X48998

Sawyer, A. G. (1975). Detecting demand characteristics in laboratory experiments in consumer research: The case ofrepetition-affect research. In M. J. Schlinger (Ed.), Advances in consumer research (Vol. 2, pp. 713–724). Ann Abor,MI: Association for Consumer Research.

Silverman, I., & Shulman, A. D. (1970). A conceptual model of artifact in attitude change studies. Sociometry, 33,97–107. doi:10.2307/2786274

Stern, D. A., Saayman, G. S., & Touyz, S. W. (1978). A methodological study of the effect of experimentally induceddemand characteristics in research on nocturnal dreams. Journal of Abnormal Psychology, 87, 459–462.

Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In W. Austin & S. Worchel (Eds.), Thesocial psychology of intergroup relations (pp. 33–48). Pacific Grove, CA: Brooks/Cole.

Zimbardo, P. G. (1971, August 14). Tape F [sound recording-nonmusical], Philip G. Zimbardo Papers (SC0750).Stanford, CA: Department of Special Collections and University Archives, Stanford University Libraries.

Zimbardo, P. G. (2007). The Lucifer effect: Understanding how good people turn evil. New York, NY: Random House.

790 J. BARTELS