Persuasive and Unpersuasive Critiques of Torture

Torture critics have offered two types of arguments in the hope of swaying public opinion against torture: A pragmatic (consequentialist) argument that “torture doesn’t work” and a moral (deontological) argument about the immorality and cruelty of torture. I present findings from two survey experiments about public support for torture among U.S. adults. The great majority of the respondents in these surveys did not endorse pragmatic arguments. They believed that torture was a quick and effective means of extracting information from detainees who had information about terror attacks. Respondents were unpersuaded by the suggestion that evidence extracted by means of torture might be fragmentary, outdated, or merely corroborative. However, when respondents were informed about the protracted nature of torture, which often requires weeks or months of interrogation before yielding results, their support for torture was lower by 14% in one survey and by 30% in a second survey. Survey participants refused to endorse prolonged torture not because they perceived torture to be ineffective, but because they felt that prolonged torture was cruel. Torture critics would be well advised to steer away from less persuasive arguments about torture inefficacy and instead confront audiences with sobering truths about the cruelty of torture.

C ontemporary critiques of torture can be grouped into two broad camps. In one group are pragmatists (or consequentialists) who object to torture on grounds of efficacy. They argue that torture yields bad intelligence. Innocent detainees will offer spurious confessions to escape their predicament (Pfiffner 2014;Rejali 2007, 462-3, 488-500;Schiemann 2016;Mayer 2005). Guilty detainees, on the other hand, will tell lies that are difficult to distinguish from accurate information. The false leads extracted by means of torture can overwhelm the intelligence apparatus and can consume precious time and resources (Boorman 2011;Costanzo and Gerrity 2009, 183). The pain of torture would cause even the most cooperative of detainees to suffer memory degrada-tion so that they would be unable to recall crucial details (O'Mara 2015). Pragmatic critics further argue that noncoercive interrogation methods that employ psychological insights to create empathy between interrogators and detainees have provided better results than torture (Pfiffner 2014, 138-40 and 145;Alexander and Bruning 2008, 83;Budiansky 2005).
Moral (deontological) critics reject torture for ethical reasons. 1 They argue that torture dehumanizes victims, treating them as means rather than as ends (Scarry 1985;Bagarig and Clarke 2005, 599;Fried and Fried 2010, 55). Torture instrumentalizes pain and undermines the victim's capacity for agency, forcing them to collude against themselves (Sussman 2005, 4-8, 21-29). Victims of torture are defenseless, cannot shield themselves, evade, or retaliate, and cannot know if or when the torture will ever end (Shue 1978, 127-30). It strips its victims of all human dignity, tyrannizes and dominates them (Luban 2007). Moral critics reject torture regardless of how well it might or might not work. 2 My goal in this paper is not to evaluate the merits of these critiques but to evaluate how persuasive U.S. audiences consider these critiques to be. How would U.S. audiences respond to new information about the efficacy of torture compared to new information about the morality of torture? Public opinion data can provide insight into how U.S. audiences evaluate torture policy. U.S. voters indirectly shape that policy, and their views can provide a proxy for the opinions of U.S. leaders who directly shape interrogation policy. 3 To evaluate the relative strength of the pragmatic and the moral arguments, I introduce two new findings on torture that emerge from recent empirical research and that U.S. audiences are less familiar with. The first finding is that torture produces only modest intelligence. Even if torture can yield accurate information from time to time, that information is often fragmentary and unreliable. It tends to be used to confirm or disconfirm prior intelligence, not to reveal new intelligence. The second finding relates to the duration of torture. Contemporary and historical evidence suggest that torture sessions that have yielded information have lasted weeks and months, not hours and days. Torture is a protracted procedure designed to break the detainee's physical and emotional capacity to resist. The purpose of the surveys that I document in this paper is to evaluate which of these two new pieces of information-the modest nature of the intelligence derived from torture or the cruel duration of torture-would prove more effective in reducing respondent support for torture.
Prior surveys have assessed public support for torture but have tended to focus on different forms and settings of torture rather than on different types of criticism. For example, after the release of the 2014 Senate report on CIA interrogation, national polls conducted by media networks and polling organizations showed significant support for the torture of terror suspects (CBS 2014;Goldman and Craighill 2014;Gronke, Rejali and Miller 2014;Ingraham 2014;Lyte 2014;Morelli 2014;Pew 2014Pew , 2015. Blauwkamp, Rowling, and Pettit (2018) argued that these survey results were sensitive to framing. References to the threat of terrorism, especially in the context of 9/11, elevated respondent support for torture whereas explicit references to torture methods in polling questions suppressed support for torture. 4 In a meta-analysis of thirty-two surveys conducted between 2001 and 2009, Gronke et al. (2010) found that public support for torture varied depending on the torture method proposed. Respondents were most likely to endorse sleep deprivation and they were least likely to endorse waterboarding, sexual humiliation, or electric shock. 5 Mayer and Armor (2012) confirmed these findings in a series of ten surveys conducted from [2004][2005][2006][2007][2008][2009], showing that respondents were most likely to approve of sleep deprivation and least likely to approve of waterboarding. 6 Of particular relevance to the surveys that I present here are the findings of the Program on International Political Attitudes poll (PIPA 2004;Nincic and Ramos 2011) and a survey conducted by Carlsmith and Sood (2009). This PIPA poll asked participants explicitly about their objections to torture and found slightly higher support for moral arguments (77%) than for pragmatic arguments (70%). This poll also confirmed that its respondents considered sleep deprivation to be the least objectionable form of torture. Moreover, the PIPA poll highlighted two factors that tended to enhance public support for torture. Higher numbers of respondents supported torture when told that there was a strong (as opposed to modest) chance that the detainee had crucial information. More respondents also supported torture when told that the detainee had information about a possible terror attack as opposed to information about a suspected member of a terror group. In a separate survey, Carlsmith and Sood (2009) showed that respondents were more likely to endorse the torture of "guilty" detainees, suspected of direct involvement in terrorism, regardless of information about the efficacy of torture. 7 Below, I use these findings to design a survey scenario that will pose the hardest test for torture critiques: The use of sleep deprivation to torture terror suspects (as opposed to mere witnesses) said to withhold information on a pending terror attack.
Consistently across these surveys, men, Independents, and Republicans have tended to endorse torture to a greater extent than women and Democrats. For example, Lizotte (2017) found an 8% gender gap in support for the use of force, including torture. Increased threat perceptions led men but not women to be more likely to support the use of torture. Drawing on the PIPA survey, Haider-Markel and Vieux (2008) showed that Republicans were more likely than Democrats to support harsh interrogation techniques and that men were more likely to support such methods than women. For example, when asked about kicking or punching a detainee, being female or Democrat decreased the likelihood of support by 6%. Miller, Gronke, and Rejali (2014) note that the shift in U.S. public support in favor of torture around 2008 was attributable primarily to Republican voters and, to a smaller extent, to Independent voters. Mayer and Armor (2012, 443) confirm that party is the strongest predictor of attitudes about torture, with Democrats most likely to oppose torture, Republicans most likely to endorse it, and Independents occupying intermediate positions.
The following analysis also seeks to contribute to the literature on attitude change and the "backfire effect" (Aramovich, Lytle, and Skitka 2012). Several scholars have argued that providing respondents with information that contradicts their beliefs can perversely strengthen attachment to those beliefs. For example, Nyhan and Reifler (2010) showed that respondents who believed that Iraq possessed weapons of mass destruction, and who were shown evidence debunking that claim, clung to their beliefs more than respondents who were not exposed to the "debunking" information. Another study found that behavior, rather than beliefs, backfired: Debunking myths about vaccines, for example, reduced beliefs in those myths yet decreased the intention to vaccinate among respondents (Nyhan et al. 2014). Yet other studies have found no evidence of this backfire effect, or have noted it only where respondents held extreme ideological or partisan beliefs on a subject (Wood and Porter 2019;Nyhan et al. 2020). These analyses propose that corrective information can be somewhat effective at increasing belief accuracy, though this effect may decay over time as cues from elites and the media undermine the information (Nyhan 2021). The surveys reported here test whether new information on the duration of torture and the modest information derived from torture will affect levels of support for torture.
To explore whether U.S. adults consider pragmatic critiques more or less persuasive than moral critiques, I designed and fielded a pair of on-line survey experiments using national samples. Respondents were recruited via Amazon's Mechanical Turk and results were collected and tabulated by Qualtrics. 8 I administered the surveys in April 2021 to 1,031 American adults and a separate sample of 1,116 American adults.
In the first survey, I asked respondents about their support for the interrogation of a known terror suspect who refuses to share information about a terror attack. To assess the impact of pragmatic considerations, a second group read the same scenario but was also told about the modest nature of the intelligence that would be extracted from the suspect. To assess the impact of moral consideration, a third group read the initial scenario but was also informed that torture might take a month. A second survey sought to assess the impact of moral considerations using a within-subject design: I provided information about the duration of torture to all respondents in sequence and not in parallel.

Torture Provides Fragmentary and Slow Intelligence
The scholarship on torture is in its infancy. Reliable information on contemporary torture is hard to come by and bias clouds both accounts of torture and their analysis. As scholars delve deeper into the archives and strive to declassify information, a clearer picture of the reality of torture has started to emerge. Public views of torture, however, lag significantly behind the scholarly consensus, influenced in large part by media portrayals of torture as a quick and efficient, if grim, means of obtaining counterterrorism intelligence. When protagonists in top grossing films employ torture, they tend to use it instrumentally and successfully to counter a threat (Delehanty and Kearns 2020). Scholars have documented the persuasive effect that these fanciful scenarios have on public endorsement of torture and on the public belief that torture "works" (Flynn and Salek 2012;Schlag 2021;Green 2005;Mayer 2007;Kearns and Young 2020).
Two aspects of torture that have escaped public scrutiny so far bear emphasizing. The first characteristic of torture to emerge from recent analyses relates to the fragmentary nature of the intelligence that torture can provide (Hassner 2020a). Even the most confident advocates of interrogational torture note that torture cannot yield complete information about specific perpetrators or plots. For example, some of the CIA's "enhanced interrogations" provided minor hints while others yielded incorrect or misleading information (Hayden and Mukasey 2009;Rodriguez 2012, 108-111;Mitchell 2016, 191-3;U.S. Senate 2012). This undermines a public perception of torture as a "silver bullet" in the fight against global terrorism. The possibility of spurious confessions and unreliable information forces investigators to engage in cross-checking to distinguish the more reliable and relevant intelligence (the "signal") from the less reliable and relevant intelligence ("the noise") (Rejali 2007, 488 and 500;Boorman 2011;Costanzo and Gerrity 2009, 183). Recent research, drawing on archival evidence regarding historical torture campaigns, confirms that, even under conditions that facilitate torture, it yields modest intelligence that torturers treat with suspicion (Einolf 2014;Einolf 2021;Hassner 2020b).
Moreover, analyzing the intelligence obtained from torture and interlacing it with intelligence from parallel sources is a time-consuming process. Even the initial stage of extracting information from torture victims can require weeks and months (Rejali 2007, 474, 507-8;Soufan 2011, 425;McCoy 2006, 70;Rumney 2006, 488;Hassner 2020a, 13-15). Interrogational torture takes time because it is not merely pain that elicits cooperation but the psychological effects that accompany pain, such as prolonged isolation and fear of future torture (Rodriguez 2012, 103;Fried and Fried 2010, 69). Sustained imprisonment leads to disorientation and hopelessness (McCoy 2006, 42). The interrogator hopes that detainees will eventually realize the futility of withholding information and will wish to bring an end to their suffering (Lagouranis 2007, 33;Rodriguez 2012, 64, 115, and 233;McCoy 2006, 10). Even without these time-consuming measures, torture takes much longer than most Americans have been led to believe because it requires escalation and repetition, long pauses between torture sessions to elicit communication, and time to process the intelligence obtained.

Hypotheses
How might U.S. respondents react to learning about these two characteristics of torture? If opposition to torture is driven primarily by pragmatic concerns, then learning about the fact that torture rarely yields new information but rather performs a corroborative function should significantly reduce support for torture. H1: Public support for torture is lower when the public is informed that torture provides modest intelligence.
If opposition to torture is based primarily on moral grounds, then learning about the protracted nature of torture should have a strong effect on opposition to torture.
H2: Public support for torture is lower when the public is informed that torture is slow.

Perspectives on Politics
It may well be that these two types of critique, the pragmatic and the moral, work side by side and can reinforce one another.
H3: Public support for torture is lower when the public is informed that torture both provides modest intelligence and is slow.
Finally, I expect that the persuasive effect of the moral critique is driven by a public misconception regarding the standard duration of torture, influenced by misleading depictions of torture in the entertainment industry.
H4: Participants expect torture to yield intelligence within hours or days rather than weeks or months.

First Survey
Procedure 1,031 American adults participated in this survey. 9 They were randomly assigned to one of four conditions, each of which varied what the respondent read about the interrogation, the duration of torture, and the resulting intelligence.
The Control condition was designed to determine respondents' baseline support for interrogations with torture. Because my goal was to provide a hard test for the persuasiveness of different arguments against torture, I sought to create a scenario in which respondents might otherwise be expected to support torture. Consequently, this scenario downplays uncertainties surrounding interrogations with torture. Q1: Control: Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack. The suspect refuses to share information about the attack. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation?
The Control condition invited respondents to acquiesce to torture by stating as fact what is often highly questionable during interrogations. It assumes explicitly that the suspect has information, and that this information is pertinent to a specific future terror attack. It also states as fact that torture may increase the odds of cooperation. It proposes a form of torture, sleep denial, that participants in prior surveys rated as the least deplorable. My first survey included three separate treatment conditions that emphasized the modest nature of information derived from torture, that emphasized the prolonged duration of torture, and that emphasized both of these aspects. Respondents not in the Control condition read the same information as in the Control condition and were randomly assigned to also read one additional piece of information, which was either the Modest, the Time, or the Combined treatment.
Modest: He probably won't reveal new information but he could help the CIA confirm their suspicions about the attack.
Time: The interrogation might last a month.
Combined: The interrogation might last a month. He probably won't reveal new information but he could help the CIA confirm their suspicions about the attack.
After completing this first part of the survey, respondents answered three demographic questions (gender, age, political identification) as well as two questions about torture efficacy and duration. The first was a multiple-choice question about efficacy: Q2: Do you think that torture can force uncooperative terror suspects to reveal useful information?
Possible answers were "yes," "sometimes," "rarely," "never," and "I don't know." The final question in the survey was an open-ended question about torture duration. I focused this question on waterboarding, rather than sleep deprivation. Because participants in prior polls consistently ranked waterboarding as the harshest form of torture, and because participants might not consider sleep deprivation to be torture unless it was protracted, I was curious to gauge their estimate regarding the duration of this form of torture. Q3: What do you think: How long might it take to waterboard an uncooperative terror suspect before they start to reveal useful information? (Limit your answer to five words).

Results
As expected, the idealized scenario proposed in the Control condition elicited a relatively high level of support for torture. Whereas in prior surveys, respondent support for torture ranged in the 30%-50%, this scenario provoked nearly 76% of respondents to support an interrogation accompanied by sleep deprivation. Of those who read the Control condition (n=254), only 24% (n=62) did not support torture.
Men supported torture at a higher rate (81%) than women (71%). Support for torture increased with age. Those aged 45 and above supported torture at a higher rate (79%) than those aged 30-44 (77%) and at a much higher rate than respondents aged 18-29 (68%). Support was highest among self-identified Republicans (92%), compared to Independents (84%) and Democrats (67%). Yet regardless of gender, age, or political affiliation, the majority of respondents acquiesced to interrogation combined with sleep deprivation in this scenario.
The Modest treatment yielded similar support for torture. Here, 74% of respondents supported interrogation accompanied by sleep deprivation even though the detainee "probably won't reveal new information." Regardless of gender, age, or political affiliation, support for torture was similar to support under the Control condition.
The Time treatment, however, had a sizeable effect on respondents. Only 60% of those who learned that "the interrogation might last a month" endorsed interrogation with sleep deprivation. Across demographics, the Time treatment indicated a lower support for torture, even for groups that showed high support for torture in prior surveys and in my Control treatment: men (62% support compared to 81% in the Control treatment), Republicans (81% support compared to 92% in the Control treatment), and Independents (47% support compared to 84% in the Control treatment).
The Combined treatment showcased lower support for torture than the Control or Modest treatments but slightly higher support than the Time treatment alone. Among respondents exposed to the Combined treatment, 64% supported torture.
When asked Q3-"How long might it take to waterboard an uncooperative terror suspect before they start to reveal useful information?"-87% of those who offered a specific timeline in response to this question (n=701) expected that waterboarding would reveal useful information in less than one week. Only 8% (57 respondents) expected that it would take more than a week and less than a month, and only 5% (33 respondents) assessed, correctly, that it would take a month or longer. Table 1 offers an ANOVA pairwise analysis of the four treatments. Table 2 displays a linear probability model. Figure 1 shows a comparison between the four treatments along with associated 95% confidence intervals that were constructed using this linear probability model.

Discussion
My first survey confirms hypotheses 2 and 4 and disconfirms hypotheses 1 and 3. Even in response to a prompt that led most respondents to sanction torture, those respondents who learned that torture might last a month endorsed torture in far smaller numbers than those who had no information about the duration of torture. However, respondents who were informed that torture would yield modest information did not endorse torture significantly less than those respondents who had no such information. Nor did these two pieces of information reinforce one another and yield much less support for torture. Respondents who were informed that torture was both lengthy and would produce modest intelligence were less supportive of torture than the control group, but they were not, by and large, less supportive of torture than those who only learned about its duration.
In most cases, adding the Modest condition to the Time condition undermined, rather than strengthened, the opposition to torture prompted by the Time condition.

Perspectives on Politics
This suggests that the two treatments are not independent of one another but have a multiplicative effect. Perhaps combining both pieces of information, drawing on different philosophical rationales, confused the participants in this survey. Alternatively, because the Modest treatment implies that some information can be derived from torture, feeble as it may be, this suggestion may have weakened the repellant effect of the Time treatment. Rather than highlighting that torture is a weak source of intelligence, the wording of the prompt may have inadvertently emphasized that torture was at least somewhat effective. 10 A review of respondents' answers to my open-ended question about torture duration (Q3) sheds some light on why the Time treatment was so powerful and significant. The participants in my survey expected torture to yield results in hours and days, not weeks or months. Given this false belief in the brevity of torture, it is not surprising that respondents responded vehemently when I informed them that it would last a month.

Second Survey
Procedure To see whether disclosing information about torture duration indeed had a robust effect on reducing support for torture, I designed and fielded a second survey; 1,116 American adults participated in this second survey. 11 The first question in this survey was identical to the Control condition of the first survey. All participants were asked this question: Q1': Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack.
The suspect refuses to share information about the attack. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation?
The following questions varied based on respondent answers to Q1'. Those who answered "yes" received a question about the expected duration of torture. This question is a variation on Q3 from the prior survey, but instead of waterboarding it focuses on the form of torture mentioned in the Control treatment, sleep deprivation. This also ensured a larger sample of answers, given lower public rates of opposition to sleep deprivation as a form of torture.
Q2': How long do you think it would take to interrogate this suspect, using sleep deprivation, before he started to reveal useful information?
Possible answers included "several days," "several weeks," "several months," "never," and "I don't know." Given the nature of sleep deprivation, I did not include "seconds" or "hours" as possible answers. Subsequently, those who offered a positive answer to Q1' and now answered Q2' then received this question: Q3': You now learn that the interrogation might last for a month before the suspect cooperates. Would you be in favor of preventing the suspect from sleeping during the interrogation if it lasted one month?
This is a variation on the Time treatment from the first survey except that the treatment is now explicit: Survey participants are openly asked to reflect on the effect of prolonged torture duration on their deliberations.

Figure 1
Support for torture across four treatments I asked respondents who rejected torture in the initial scenario (Q1'): Q4': Which of the following pieces of information is most likely to change your mind, so that you WOULD be in favor of interrogating the terror suspect using sleep deprivation?
Possible answers included: "You find out that the suspect has precise information about the timing and location of the future attack," "you find out that the suspect is himself a terrorist," "you find out that the attack will happen very soon," "you find out that the attack will kill many people," and "none of these answers." These possible answers were displayed in random order. I asked respondents who continued to endorse torture even after they were told about its lengthy duration (Q3') the inverse question: Q5': Which of the following pieces of information is most likely to change your mind, so that you would be AGAINST interrogating the terror suspect using sleep deprivation?
Possible answers were the inverse of the answers to Q4': "You find out that the suspect has only vague information about the timing and location of the future attack," "you find out that the suspect is not a terrorist," "you find out that the attack will not happen anytime soon," "you find out that the attack will kill few people," "None of these answers." These answers were displayed in random order. I wanted to find out why respondents, who initially favored torture, no longer favored torture when they learned about its duration. In particular, I wanted to explore whether the "ticking bomb" logic determined their stance. Was prolonged torture too slow to effectively defuse an imminent threat or was slow torture too cruel? Q6': Why are you against interrogating this suspect, with sleep deprivation, for a month? Select the best reason.
Possible answers included: "A month is too long to deprive someone of sleep," "If it takes a month, it can't prevent an attack that is happening soon," "by the time he shares the information, it'll be out of date," and "none of these answers." These answers were displayed in random order. This second survey concluded with three demographic questions (gender, age, political identification).

Results
Participant responses to Q1' correspond to the results of the Control treatment (Q1) in the first survey: As before, 76% of respondents endorsed torture in this scenario and 24% did not.

HYPOTHESIS 2
In the first survey, the information about time was part of a survey experiment. Respondents who learned about the month-long duration of torture were unaware that other respondents, in parallel, were not told about torture duration. In this second survey, I was transparent in presenting information about torture duration as a novel input. Only respondents who had been willing to endorse torture in response to Q1' were told about the lengthy duration of torture. The phrasing of the question ("You now learn …") certainly primed the respondents to reconsider their prior endorsement of torture.
Nonetheless, the results are remarkable. A full 39% of participants who had initially supported sleep deprivation now changed their minds and no longer favored sleep deprivation (refer to figure 2).
Adding these novel torture opponents to the initial torture opponents suggests that new information about the month-long duration of torture shifted the camp supporting torture to the minority across all demographics (with the exception of respondents who self-identify as Republicans). If, in response to Q1', 76% of respondents endorsed torture, now, in response to Q3', 30% changed sides, leaving only 46% of all respondents in support of torture. It is worth reiterating that this is a wholesale refusal to support torture in a scenario designed to evoke acquiescence of torture: certainty about the identity of the detainee, their possession of intelligence pertaining to a future terror attack, and the use of sleep deprivation as the chosen form of torture.

HYPOTHESIS 4
The responses to Q2', regarding expectations about torture duration, also map onto the responses I received to the duration question (Q3) in the prior survey. 783 respondents, all of whom had not favored torture in Q1', selected a precise timeline in response to this question. As before, the overwhelming majority expected torture to yield results within days and few respondents expected the interrogation to last weeks or months. Table 3 compares the answers of Q3 in the first survey to Q2' in this survey.
As could be expected, there was a relationship between expectations about torture duration and refusal to endorse lengthy torture. Those who assumed in Q2' that torture should last days were particularly taken aback by the disclosure in Q3' that torture might last one month. Of those who initially endorsed torture and expected that torture to last weeks or months, 21% now opposed torture. Of those who thought that torture should last days, 46% now opposed torture.
Q4' asked respondents what it would take to sway them in favor of torture. Many respondents (40%) checked "none of these answers," implying that nothing would change their minds. Another 29% suggested that finding out "that the suspect has precise information about the timing and location of the future attack" might change 166 Perspectives on Politics their opinion, echoing the weak effects of the Modest condition in the first survey. Only 8% chose "you find out that the attack will happen very soon" as a reason to support torture.
Q5' asked respondents the opposite question: What would it take to change their opinion and oppose torture? Forty-nine percent of respondents who had sanctioned torture in Q3' chose "You find out that the suspect is not a terrorist." Another 22% claimed that they would turn against torture if they found out that "the suspect has only vague information." Only 6% chose "you find out that the attack will not happen anytime soon." Another 16% checked the "none of these answers" option.
Q6' asked those who changed their mind when they learned of its duration, and no longer favored torture, why they changed their mind; 332 respondents answered this question. Only 34% selected one of the two pragmatic answers: Torture is ineffective in preventing looming attacks (21%), and information would be outdated (14%). The majority (61%) selected the moral answer: "A month is too long to deprive someone of sleep."

Discussion
The disclosure that torture might last a month led the majority of participants in this second survey not to endorse torture, even though most had initially sanctioned torture in this specific scenario. Q4' and Q5' were designed to understand why learning about the duration of torture changed (or did not change) their endorsement of torture. Their responses to these questions should be treated with caution given their appearance at the end of a survey containing multiple loaded questions.
Given the salience of the ticking bomb metaphor, I expected that information about the proximity of an attack and the urgency of successful interrogation would be a primary factor in determining opinions about torture. Though I did not manipulate proximity and urgency in this survey, the answers to Q5' suggest that respondents were less interested in urgency. Most respondents (61%) claimed to object to prolonged torture because it was cruel, not because it was ineffective in defusing ticking bombs.

Conclusion
The participants in my surveys found pragmatic critiques of torture to be unpersuasive. Less than 28% of participants in my first survey (n=1,030) felt that torture was rarely, or never, effective at forcing uncooperative terror Change in support for interrogations with lengthy sleep deprivation Note: Confidence intervals at 95%. The column on the left depicts the proportion of respondents who endorsed torture in Q1' (n=1116). The column on the right depicts the proportion of respondents who endorsed torture once informed that torture might last for a month (Q3'). This comparison assumes that all of those who opposed torture after Q1' continued to oppose torture after Q3'. suspects to reveal useful information. Only 5% of participants did not know whether torture could yield intelligence. In contrast, 67% felt that it was effective at least part of the time. 12 In the first survey, participants exposed to prompts that emphasized the limited utility of torture did not object to torture in larger number than participants who only read the Control prompt. In contrast, participants who read prompts that emphasized the duration of torture were 14% less likely to endorse torture. In the second survey, 39% of participants who endorsed torture after reading the Control prompt changed their mind when they learned about the duration of torture. Asked why they changed their mind, they opted for a moral explanation.
Yet the scholarship on torture continues to focus the bulk of its attention on the efficacy and inefficacy of torture (Pfiffner 2014, 138-140, 145;Budiansky 2005;Duke and Van Puyvelde 2017, 312-314, 327;Costanzo and Gerrity 2009, 183;Rumney 2006, 479). Many scholars privilege the claim that "torture never works" as their primary critique of torture (Schiemann 2016;O'Mara 2015). The surveys examined earlier suggest that these pragmatic arguments are falling on deaf ears. For those audiences who find the moral critique of torture to be powerful, the pragmatic critique is, at best, irrelevant. If torture is evil, then its efficacy is immaterial. No level of efficacy or inefficacy can shift the moral balance of torture when one subscribes to a deontological critique of torture.
How might anti-torture advocates utilize my findings? As Mayer and Armor note (2012, 44), torture advocates and leaders who employ torture have relied on public ignorance regarding torture to support their policies. When these advocates discuss torture, they do so in vague terms, offering hazy reassurances regarding effectiveness but dodging details regarding methods, frequency, duration, or long-term harm. Anti-torture advocates might counter these efforts by declassifying detailed information on the gruesome nature of contemporary methods. The more detail they can offer the public on what exactly contemporary torture looks like, the more likely they are to sway public opinion.
The case of waterboarding makes for an instructive example. Though it has come to be the primary method that American audiences associate with contemporary torture, Americans don't know what waterboarding is or how it works (Hassner 2020a, 5-6). Human rights activists, NGOs, journalists, and scholars have provided conflicting accounts. Some sources depict waterboarding as full-body dunking, drowning the victim in a barrel or tub of water. Others describe it as entailing the ingestion of water, poured from a vessel into the nasal or oral cavities, perhaps entering the lungs, posing some risk of drowning. Yet other sources describe water poured only on the victim's face, perhaps on a towel or sheet of cellophane, creating the mere illusion of drowning. Which is it? Nor have anti-torture activists succeeded in declassifying information on how many U.S. detainees were subjected to waterboarding, how often each detainee underwent waterboarding, or how long these sessions lasted. For example, one CIA source claims that Abu Zubaydeh was waterboarded only once, a second CIA source claims that he was waterboarded about five times, and a third CIA source claims that he was waterboarded eighty-three times. Abu Zubaydeh told the Red Cross that he was waterboarded at least ten times (Soufan 2011, 368;Rodriguez 2012, 177, 236;Mitchell 2016, 70;Mayer 2009, 173-4).
It is easy to imagine that these conflicting accounts would provoke entirely different levels of moral outrage from American observers. Tricking a detainee once into falsely believing that they might drown is quite a different matter than forcefully drowning a restrained victim in a tub eighty-three times. As long as torture remains a hazy procedure, administered to faceless figures in unknown locations, it is easy to dismiss it as mere "enhanced interrogation." Torture opponents should not shy away from uncovering and publicizing the disturbing details of these methods. As the surveys above demonstrate, the mere disclosure of the duration of torture can undermine public endorsement. Revealing the full nature of torture, its frequency, and the physical and mental scars it leaves on victims is likely to have an even stronger effect on public opinion.
Like all survey-based research, the findings in these surveys should be treated with caution. Online surveys are of limited external validity even when administered to large numbers of participants. Respondents know that the scenarios in the survey are fictional and even slight variations in the wording or the order in which questions are posed can prime them and bias the results (Shadish, Cook, and Campbell 2002;Wallace 2013a, 116). Recall also that, because the torture scenario proposed in these surveys was designed to set a high bar for criticism of torture, it envisions a unique constellation of conditions: The use of sleep deprivation to torture terror suspects said to withhold information on a pending terror attack. It would be important to test how robust these results are to variations in the form of torture used (for example, food deprivation as opposed to sleep deprivation) or to variations in the nature of the threat (for example: kidnapping as opposed to terrorism).
The results presented here call for replication and further investigation. They suggest that American audiences can be swayed to lower their support for torture if they are presented with the right arguments. Scholars should divert at least some of their attention away from the slogan "torture does not work" and emphasize instead that torture is cruel, no matter how well it does or does not work.
168 Perspectives on Politics

Acknowledgments
The author thanks Amit Ahuja, Dan Gamarnik, Julia Macdonald, Paasha Mahdavi, Miroslav Nincic, and Geoffrey Wallace for their helpful comments on prior drafts of the paper. He thanks Anne Hassner for her help in tabulating survey responses. He is especially grateful to Max Kagan for analyzing the quantitative survey results and for producing the charts and tables.

Notes
1 Pragmatic critiques of torture on consequentialist grounds and deontological critiques are both "moral." I use "pragmatic" and "moral" as a shorthand for critiques that emphasize the consequences and efficacy of torture, on the one hand, versus those that focus on the ethical essence of torture, on the other hand. There is some overlap between these two categories. On the distinction, with a focus on torture, see Nincic and Ramos (2011) and Hassner (2020a, 29). 2 For a contemporary debate regarding pragmatism versus morality in the parallel realm of saturation bombing, see Carpenter and Montgomery (2020) and Sagan et al. (2020). 3 I thank Geoffrey P.R. Wallace for proposing this formulation of the project's significance. 4 See also Crandall et al. 2009;Porpora, Nikolaev and Hagemann 2010;Conrad et al. 2018;Wallace 2013a, b. 5 See also Jordan 2014; Kearns and Young 2020. 6 I thank an anonymous reviewer for introducing me to this survey. 7 Along similar lines, see Liberman (2013, 285-306). I thank the anonymous reviewers for drawing my attention to this source. 8 On the representativeness of MTurk samples, see Berinsky, Huber and Lenz (2012) and Huff and Tingley (2015). 9 Excluding outliers (2.6% of participants who took less than 30 seconds or more than an hour to complete the survey), the average time to completion was 118 seconds (about 17 seconds per question) and the median time to completion was 82 seconds. Excluding these outliers did not significantly affect the results. The completion rate was 90%. 10 I thank an anonymous review for this helpful phrasing. 11 Excluding outliers (5.3% of participants who took less than 30 seconds or more than an hour to complete the survey), the average time to completion was 118 seconds (about 17 seconds per question) and the median time to completion was 67 seconds. Excluding these outliers did not significantly affect the results. The completion rate was 91%. 12 This aligns with Blauwkamp, Rowling and Pettit (2018). It is worth noting that this question on efficacy from the first survey follows the treatment, so it may be influenced by the prompts in the first question. That said, the different treatments did not yield a statistically significant variation in the responses to this second question.
attack. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation?
• Yes • No Q1B: Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack. The suspect refuses to cooperate. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation? He probably won't reveal new information but he could help the CIA confirm their suspicions about the attack.
• Yes • No Q1C: Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack. The suspect refuses to share information about the attack. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation? The interrogation might last a month.
• Yes • No Q1D: Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack. The suspect refuses to cooperate. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation? The interrogation might last a month. He probably won't reveal new information but he could help the CIA confirm their suspicions about the attack. Survey 2 Q1': Imagine that the CIA has been collecting information about a future terror attack in the U.S. They have now detained a suspect who has some information about this attack. The suspect refuses to share information about the attack. If he is denied sleep during the interrogation, he is more likely to cooperate. Would you be in favor of preventing the suspect from sleeping during the interrogation?
• Yes • No [Respondents who answered "Yes" to Q1' are asked the following question} Q2': How long do you think it would take to interrogate this suspect, using sleep deprivation, before he started to reveal useful information?
• Several days • Several weeks • Several months • Never • I don't know [Respondents who answered "Yes" to Q1' are asked the following question] 172 Perspectives on Politics Q3': You now learn that the interrogation might last for a month before the suspect cooperates. Would you be in favor of preventing the suspect from sleeping during the interrogation if it lasted one month?
• Yes, I would • No, I wouldn't [Respondents who answered "Yes" to Q3' are asked the following question. Answers were presented in random order] Q4A': Which of the following pieces of information is most likely to change your mind, so that you would be AGAINST interrogating the terror suspect using sleep deprivation?
• You find out that the suspect has only vague information about the timing and location of the future attack • You find out that the suspect is not a terrorist • You find out that the attack will not happen anytime soon • You find out that the attack will kill few people • None of these answers [Respondents who answered "No" to Q3' are asked the following question. Answers were presented in random order] Q4B': Why are you against interrogating this suspect, with sleep deprivation, for a month? Select the best reason.
• A month is too long to deprive someone of sleep • If it takes a month, if can't prevent an attack that is happening soon • By the time he shares the information, it'll be out of date • None of these answers [Respondents who answered "No" to Q1' are asked the following question. Answers were presented in random order] Q4C': Which of the following pieces of information is most likely to change your mind, so that you WOULD be in favor of interrogating the terror suspect using sleep deprivation?
• You find out that the suspect has precise information about the timing and location of the future attack • You find out that the suspect is himself a terrorist • You find out that the attack will happen very soon • You find out that the attack will kill many people • None of these answers