Unique Research Opportunity

This research began with an invitation to assess a training program titled Judging in a Democratic Society. This initiative combined the efforts of American and British lawyers to provide instruction in democratic norms and the rule of law to Iraqi judges following the military intervention in Iraq. The invitation offered a valuable opportunity to observe a rule-of-law indoctrination exercise in action: there are few in-depth studies of how local actors adapt to forced efforts by outside agents to impose law reform initiatives into war-torn and unstable settings (Reference Klug, Dezalay and GarthKlug 2002). After extensive search, we have been unable to locate any empirical studies of Iraqi judges since the period of the recent occupation by Coalition forces.

The training sessions for Iraqi judges were part of the U.S. State Department's plan to impose legal order in the aftermath of the invasion and ensuing occupation of Iraq. During this time, joined arguments for democracy and rule of law steadily gained priority over the weakening and eventual disappearance of weapons of mass destruction justifications for the war. President George W. Reference BushBush (2003) emphasized the linkage of democracy and rule-of-law themes in his November 2003 speech to the National Endowment for Democracy, insisting that “successful societies protect freedom with the consistent and impartial rule of law” (Reference BushBush 2003: n.p.). Two years later, in his National Strategy for Victory in Iraq, President Bush reported success along his roadmap to judicial independence. He observed that “hundreds of judges have been trained” and concluded that “as Iraq's political institutions mature, its judicial system has become an independent branch, better able to promote the rule of law” (Reference BushBush 2005:16–7). Our case study is focused on the normative frameworks of these judges.

Of course, the Bush administration joining of democracy and rule-of-law themes is not new, and what A. V. Dicey by the end of the nineteenth century called “the rule of law” is now widely regarded as an essential foundation for modern democracy (Reference WalkerWalker 1988). An independent judiciary (e.g., embracing the equality norm of “similar cases, similar treatment”) is the symbolic institutional match for electoral democracy (e.g., honoring the equal enfranchisement norm of “one person, one vote”). The Bush administration was faced with the challenge of how to jointly and symbolically impose these ideological pillars of the rule of law.

Rehabilitating the Iraqi judiciary became an important answer. In contrast with the Iraqi military, the Iraqi legal system and the judiciary appointed during the Hussein regime were not subjected to de-Baathification. Instead, to presumably enhance judicial independence from financial corruption, the Coalition Provisional Authority (CPA) increased the salaries of judges from approximately 100 to 1,000 dollars per month. Yet the judges still required what the American bar calls “continuing legal education.” That is, the law and the judges still required reform and re-education. Since the semblance of the latter could be mobilized more quickly than the substance of the former, the idea quickly emerged of what in less professionally conscious circumstances might have been called a “judges' school.”

This research opportunity to observe Iraqi judges followed a post-invasion introductory meeting initiated by American and British lawyers in August 2003 in Baghdad between the leadership of the Iraqi bar and representatives of Coalition countries. Tragically, this meeting ended in the chaos of the August 19 suspected al Qaeda–led bombing of the U.N. headquarters that killed Sergio de Mello, the U.N. High Representative to Iraq, among scores of others. The Coalition team was forced to leave immediately by car for Jordan. The bombing added impetus to a British-based international legal association's proposal to assemble groups of Iraqi judges with American, British, and other international judges and lawyers at a legal training institute located in a more secure setting. A legal institute devoted to such purposes was already in operation in a central European capital, training judges and lawyers from the post-Communist states of eastern and central Europe aspiring to join NATO and the EU.

With the support of the State Department, the institute invited successive groups of Iraqi judges to attend two-week “Judging in a Democratic Society” programs, taught pro bono by U.S. judges and lawyers. Instruction focused on the role of judges in a democratic society, the importance of judicial independence, court and case management, sources of law (with an emphasis on international human rights and humanitarian law), judicial ethics, media and the courts, community outreach, and judicial leadership. We were invited to attend and assess these two-week courses with an evaluation methodology of our choice.

Of course, our invitation was not independent of the goals of the legal institute and its sponsors, who wished to be positively appraised for their contribution to the advancement of democracy and the rule of law in Iraq. The State Department and the organized legal profession expected that an independent and relatively powerful judicial branch could help legitimate the forced regime change and to overcome the violence in Iraq, while the institute hoped that we could positively evaluate their contribution to meeting this need. We use a critical legal perspective to mitigate this potential bias.

A Critical Legal Perspective on Torture and the Rule of Law in Iraq

An immediate problem confronting the CPA was that Iraq follows an inquisitorial civil law rather than an adversarial common-law tradition. Nonetheless, emulating the American legal system, the CPA decided early in 2003 to establish a new central criminal court and courthouse in central Baghdad, modifying Iraq's existing system of provincial courts with a model based on the specialization of the U.S. federal courts in trying significant cases of corruption, violence, and national security. The problem was that at a minimum, as Dezalay and Garth observe, “the results of particular exports of state expertises will depend on the extent to which there are structural homologies in the respective state fields of the importers and exporters” (2002a:14).

Yet Dezalay and Garth and a critical legal perspective further warn that even this prerequisite is too narrow, because the basic indeterminacy of law does not provide the necessary foundations of certain and settled principles that a direct application of this model of law in Iraq would require. Legal indeterminacy is a particularly vexing issue in an occupied Iraq that is plagued by torture and terrorism, if only because the indeterminacy problem is especially prominent in laws regarding violence (Reference CoverCover 1986) and torture (Reference WaldronWaldron 2005).

Studying the Rule of Torture Law in Iraq

Our challenge was to understand the normative judgments of Iraqi judges during the occupation of their country. We used the factorial survey method pioneered by Reference Rossi and MertonRossi (1979) and developed with associates (Reference Rossi and BerkRossi & Berk 1985; Reference JassoJasso 2006), which has been called “the methodological gold standard” for the study of social norms in decisionmaking situations (see Reference Sorenson and TaylorSorenson & Taylor 2005:79; Reference SeronSeron et al. 2006:931; also Reference HerzogHerzog 2003). This method has been widely applied to a variety of beliefs and normative judgments (Reference Rossi and SampsonRossi, Sampson, et al. 1974; Reference Sampson and RossiSampson & Rossi 1975; Reference Jasso and RossiJasso & Rossi 1977; Reference Alves and RossiAlves & Rossi 1978; Reference Alves, Rossi and NockAlves 1982; Reference Nock and RossiNock & Rossi 1978; Reference Rossi, Anderson, Rossi and NockRossi & Anderson 1982; Reference Bose and RossiBose & Rossi 1983; Reference JassoJasso 1988, Reference Jasso1990; Reference Jasso and OppJasso & Opp 1997; Reference Jasso and WebsterJasso & Webster 1999; Reference HechterHechter et al. 1999; Reference Jasso and Lewis-BeckJasso 2003), including judgments about crime and punishment (Reference Berk and RossiBerk & Rossi 1997; Reference Jacoby and CullenJacoby & Cullen 1998; Reference JassoJasso 1998; Reference Rossi and SimpsonRossi, Simpson, et al. 1985). In our application, each Iraqi judge was asked to respond to hypothetical cases or vignettes by assigning “just” prison sentences to prison guards accused of torturing prisoners who were suspected terrorists. Our design is thus in the tradition of Reference Berk and RossiBerk and Rossi (1997) and Reference Jacoby and CullenJacoby and Cullen (1998).

The factorial survey method joins features of surveys and experiments. Vignettes, or scenarios—in this instance describing torture cases—are followed by questions that elicit normative judgments—in this study, about sentence length. The vignettes contain common elements or variables whose categories or values are randomly varied in an experimental fashion. The random assignment allows variables to be examined independently of one another. For example, factorial studies of sentencing norms in criminal cases allow the researcher to assess the independent effects of characteristics of the defendant (e.g., a Coalition soldier) and the victim (e.g., an al Qaeda prisoner) upon judges' punishment (e.g., sentence length) of particular behaviors (e.g., torture). Factorial designs further allow the researcher to examine how characteristics of the respondents (e.g., the fears and protective concerns of Iraqi judges) affect their judgments.

“In other words,” Sorenson and Taylor surmise, “factorial designs allow researchers to examine what goes into the judgments of thoughtful respondents” (2005:79). They note that this cannot be done with direct yes-no questions, even when the questions pose extreme situations, because of the range of factors respondents may bring to bear in responding to even apparently simple situations, and the social desirability bias involved in simplified responses. The factorial survey method overcomes these limitations by systematically widening the range of circumstances or conditions that in practice complicate normative situations and judgments. Reference SeronSeron et al. (2006:931) further note that in punishment contexts a factorial survey design presents both rule-breaking and mitigating evidence, and in this way it is designed to mimic the reasoning that judges actually apply. They speak of this methodology as removing the “veil” that so often obscures official decisionmaking.

In our research, the factorial survey method can be thought of as allowing an estimation of “the equations in the heads” of the judges as they assign “just sentences” to the hypothetical cases. These data and the method allow an exploration of the issue of judicial independence in terms of the influence of legal and extralegal offender and victim characteristics as components of these judgments, the interjudge variability in judgments and their components' characteristics, and the added estimation of characteristics of judges as determinants of their judgments and their components.

We estimate both judge-specific (Tables 2, 3, and 5) and pooled (Table 4) OLS equations in the analyses that follow. The pooled results are corrected for heteroskedasticity due to clustering of cases within judges. The judge-specific equations yield significantly better-fitting models and diminish potential problems resulting from distributional variation in sentencing outcomes. We follow the classic approach in presenting OLS regression estimates that allow us to interpret our results in terms of years of sentence length, but we also estimate median regressions and log-transform sentence models to confirm that our results are robust with regard to distributional variation. The latter results are available on request from authors.

Table 1. Descriptions and Means of Selected Variables in the Analysis

Table 2. Means of Regression Coefficients from Judge-Specific Just-Sentence Equations, November and April Groups

Notes: The regression coefficients are drawn from the judge-specific equations, which include the full set of vignette characteristics; the omitted reference categories are Terrorist Level – Prisoner not a terrorist; Gender – Prison Guard: Male, Prisoner: Male; Command Responsibility – Prison Guard not given command to torture; Group Affiliation – Prison Guard: Coalition Member, Prisoner: Iraqi; Marital Status – Prisoner Guard: Single, Prisoner: Single. The April Group 2 equations also include the nationality/military affiliation variables (described in the text and summarized in Table 3).

Table 3. Summary of Estimated Effects of the Guard-Prisoner Nationality Combination in the Judge-Specific Just-Sentence Equations, April Group 2

Notes: The effect for each guard-prisoner nationality combination represents the prison years added to the sentence. It is obtained in each judge-specific equation by subtracting the smallest coefficient from the associated coefficient (where the coefficient for the left-out category is set to zero in the case of all positive coefficients).

Table 4. Estimates of OLS Pooled Specifications of Just-Sentence Equations, April 2005, Group 2

^a Reference Categories: Terrorist Level – Prisoner not a terrorist; Gender – Prison Guard: Male, Prisoner: Male; Command Responsibility – Prison Guard not given command to torture; Group Affiliation – Prison Guard: Coalition Member, Prisoner: Iraqi; Marital Status – Prison Guard: Single, Prisoner: Single. Standard errors (in parentheses) corrected for heteroskedasticity due to clustering by judge.

^* p<0.05;

^** p<0.01;

^*** p<0.001.

Table 5. Sources of Cross-Individual Variation: Based on OLS Estimates of the Respondent-Specific Just-Sentence Equations

Note: The dependent variables in the three specifications are based on the estimates of the judge-specific equations of the perceived determinants of a just prison sentence; means and other summary characteristics of these estimates are reported in Table 3, and the underlying model (Model 3) is summarized in Table 4. The dependent variable in Specification 1 measures the total effect in just-sentence years of the group affiliation category. It is equivalent to the range between the lowest effect and the largest effect, with the lowest effect set to zero. The dependent variable in Specification 2 was constructed using the variable coal/alq, which measures the effect in just-sentence years of the coal/alq combination. It is equivalent to the range between the lowest coefficient and the coefficient of coal/alq. The dependent variable in Specification 3 was constructed from the estimated coefficients for the group affiliation variable in the respondent-specific equations of the perceived determinants of a just prison sentence reported in Table 3. The standard errors appear in the parentheses.

⁺ p<0.1, one-tailed test

^* p<0.05, one-tailed test

As we noted at the outset, this method cannot represent extant sentencing practices in Iraq, because U.S. soldiers are immune from prosecution in Iraqi courts. Nonetheless, it is relevant to note that the cases we consider could be—and may ultimately be—tried in Iraqi courts. There are a number of recent cases whose fact situations recently have pressed the boundaries of U.S. immunity (“U.S. Wants Iraq to Try American Suspected of Aiding Insurgency,”USA Today, 8 Feb. 2006; Reference LewisLewis 2006). Future cases are expected to test this boundary in new and numerous ways (see also Reference MossMoss 2006b). There is abundant international precedent for transfers and prosecutions of American soldiers and citizens to domestic courts. The principle of comity in international law obliges nations to allow the extradition of its own citizens who have committed crimes in host countries. For example, Italian prosecutors recently have sought to prosecute 25 U.S. CIA agents for their involvement in the kidnapping of a militant Egyptian cleric from Milan in 2003 (Reference FisherFisher 2007).

Yet it is also important to note that such cases are likely atypical, much as Reference SeronSeron and colleagues (2006:927) similarly note with regard to highly publicized domestic cases of police misconduct or brutality. It is in such circumstances, where our knowledge is so limited and atypical, that a factorial survey design can be especially useful. In addition to broadening our social scientific understanding of legal decisionmaking in a little-studied but important kind of international setting, this research can speak to important policy concerns by widening our knowledge of the normative standards of key stakeholders—in this case, members of the Iraqi judiciary.

The Unfolding Design

Methods and Data Collection

As indicated above, the factorial surveys with the Iraqi judges were conducted with one group of 43 judges in November 2004 and two groups of 20 and 19 judges each in April 2005. These groups respectively produced 2,115; 996; and 947 imposed sentences. We first examine mean differences in sentences across the three groups of judges described. Next we consider differences within and between normative judgment equations estimated from the factorial surveys conducted in November and April, before and after the election of the Transitional National Assembly. Finally, we focus in greater detail on the second group of April judges who sentenced cases that included information about the nationality/military group affiliations—including Coalition forces and suspected al Qaeda members—of both the guards accused of torturing and of the imprisoned suspected terrorists. After examining coefficients reflecting substantial sentencing differences in the normative-judgment equations for cases involving salient combinations of guard and terrorist characteristics, we examine the cross-level interaction of judicial fear and concern about protection with salient combinations of the prisoner-guard case characteristics. This last step fully incorporates the multilevel capacities of the factorial survey method.

The Factorial Survey Results

Table 1 presents mean values for four alternative codings of the sentences the judges assigned in the hypothetical cases. The third coding scheme uses plausible life span assumptions in assigning 25 years for a life sentence and 50 years for a death sentence. The initial results in Table 1 indicate that the several coding schemes yield similar results, and we selected the third coding scheme for the remainder of our analysis.

A key early finding in Table 1 is that in contrast with our first hypothesis, the mean sentences—in excess of four years in prison—are not lenient. Thus even though mean sentence length modestly but significantly declined between November 2004 and April 2005 (focusing on the first April group, which participated in the same design as the November group, and regardless of the sentence coding), the overall average sentences were substantial. After the elections and the accompanying upsurge in violence, the mean severity of sentences for the torture of suspected terrorists decreased, with the third sentence coding indicating a decline from 4.51 to 4.01 years. However, these are still long sentences. The Bush administration memo and the presence of the American judges in the institute program apparently did not lead the Iraqi judges to leniently sentence most convicted torturers.

Even more striking are the sharply increased mean sentencing scores by the second group of Iraqi judges in April 2005, again across all codings. These increases are on the order of 50 percent, for example, from 4.01 to 5.95 years. Recall that the only change in the hypothetical cases presented to the second group of judges in April is the inclusion of added information on the nationality/military group affiliations—including Coalition forces and al Qaeda members—of the prison guards and prisoners. When these affiliations of the respective combatant parties are indicated, the Iraqi judges on average become much more severe in their selections of appropriate sentences and therefore would seem to be more committed to strict enforcement of the international rule of law about torture. This abrupt upturn in sentence severity is consistent with our second hypothesis, that torture convictions involving Coalition forces and al Qaeda members represent “trouble cases” for these Iraqi judges.

Table 1 also includes several plausible judge-level sources of variation in sentencing, including the judges' ages, scores on a fear scale, death rates in the judges' jurisdictions, the leadership of their assigned “breakout” discussion group in the program, the importance they attach to the elections in Iraq, and a measure of the importance judges assign to police protection. Incorporating a critical legal perspective on punishment, we included the fear and protection measures as likely sources of severity in sentencing. The judicial fear scale is the sum of the importance the judges assign to the threats posed by assassinations, kidnappings, rapes, and property crimes in recent months in Iraq. These rankings yield a reliable 20-point fear scale (alpha=0.6). The protection scale is based on the judges' responses to a single item asking the importance they attach to the issue of police protection in Iraq. The reported means in Table 1 indicate that the Iraqi judges are both highly fearful and concerned about police protection. In April Groups 1 and 2, the judges score an average of 17 and 17.8 on the 20-point fear scale and between 3.8 and 4.5 on the five-point police protection measure.

Tables 2 and 3 provide an overview of the main effects of the case characteristics on sentencing in the three different groupings of judges (November/April 1 and 2) introduced in Table 1. We ran separate regressions for each of the judges. Tables 2 and 3 summarize the main effects by presenting the average regression coefficients for case characteristics in the judge-specific just-sentence equations for each of the three groupings; the main effects of the nationality/military group affiliation case characteristics newly introduced with the Iraqi April Group 2 judges are presented separately and in greater detail in Table 3.

In Table 2, as in Table 1, distinctive results emerge in April Group 2, and it cannot be overemphasized that assignment to the two April groups was random. Thus although there are apparent differences in the coefficients from the equations for just sentences across the first two groups in Table 2, the coefficients in the first two columns are more similar to one another than they are to the coefficients in the third column representing April Group 2. For example, while guards accused of torturing prisoners who are identified generically as suspected terrorists are indicated in the first two columns as receiving nearly identical lenient sentences compared to undifferentiated suspects (mean b=−1.539 and −1.424), this difference grows and reverses direction in the last column (mean b=2.239). These similarities are not as great across the first two columns in the mean effects for suspects identified more specifically as low- and high-level terrorists, but the effects in the third column with April Group 2 judges are nonetheless again more dissimilar. There are similarities in the first columns and reversals for most of the characteristics in Table 2. These findings of sharp differences when Coalition and al Qaeda affiliations are introduced into the vignettes suggest further support for the second hypothesis, that this information transforms these cases into “trouble cases.”

A final comparison of effects across the three groups is of interest in Table 2 before we focus on the specific nationality/military group affiliations in the second April group in Table 3. This comparison involves the variable indicating whether the accused guard and suspected terrorist are from the same or different ethnic groups. When the guard and prisoner ethnicities differ, the coefficients are negative in all three groups (mean b=−0.825, −1.072, −1.171). The increase in the negative coefficients in April suggests that, if anything, over this time period the judges on average increased their willingness to tolerate cross-group torture. We also explored the interaction of the ethnicity difference variable and the nationality/military group characteristics (to be discussed below), but the results were not statistically significant.

More generally, the data above indicate substantial similarities in the Iraqi judges' imposition of sentences across the November and the first April group—the groups whose vignette design omitted information about Coalition and al Qaeda involvement as guards and prisoners. Yet when these nationality/military affiliations are introduced into the cases for the second April group, the structure of the decisionmaking involving many of the case characteristics changes markedly, often even changing direction. The vignettes have become trouble cases. This difference suggests the importance of the guard-prisoner affiliations introduced into the experimental design in response to the critical legal perspective in April Group 2 and now explored in greater detail in Table 3.

Table 3 considers the relationship between the combinations of nationality/military group affiliations and sentence severity in the second April group of Iraqi judges. It is relevant to note here that Reference Tittle, Bridges and MyersTittle (1994) actually anticipates the significance of an occupying army such as the Coalition forces. In doing this, Tittle warns that “the idea that societal institutions encompass the emotions of the dominant groups or categories is not as simple as it might seem, and spelling out the exact empirical expressions of identification and fear in the social control process is complicated” (1994:43). This is the kind of elaboration in our design that the recognition of trouble cases and a critical legal perspective demands.

Table 3 provides the mean, standard deviation, minimum, and maximum of the regression coefficients for the guard-prisoner combinations of nationality/military affiliations estimated in the judge-specific normative-judgment equations. The third column expresses these coefficients as actual estimated years of just imprisonment by setting the minimum coefficient to zero.

One of the combinations we have highlighted for comparison stands out in terms of the standard deviations of the effects: the largest standard deviation (14.9 years) involves Coalition guards who tortured al Qaeda prisoners. The smallest standard deviation (2.84 years) involves Coalition guards who tortured Iraqi prisoners. The sentencing range for the latter group is about 12 years, while the range for the former group quadruples to 48 years. The sensitivity of the judges to Coalition force involvement in torturing al Qaeda prisoners is underscored when we contrast the standard deviations of those in the Iraqi–Iraqi and Iraqi–al Qaeda combinations, which are the second and third smallest in the table. This is clear evidence of a heightened variability in the judicial response to the torture of al Qaeda prisoners by Coalition guards. It is seemingly the indeterminacy highlighted by the critical legal perspective in the meaning of torture, coupled with the “troublesome” power dynamics of the nationality/military affiliations, that produces this large variability in sentence severity.

Examination of the results of the individual judge-specific equations (not shown, but available on request) that form the foundation for Table 3 further reveals that even though each judge sentenced only 50 hypothetical cases, the Coalition–al Qaeda combination coefficients are significant (at the one-tailed 0.10 level) in 14 of 19 instances. It is of further note that in nine of these instances the coefficients are negative, reflecting sentence leniency, while in five of the instances the coefficients are positive, reflecting sentence severity. This is a further indication that in response to the indeterminacy of torture law, these Iraqi judges sentenced the Coalition–al Qaeda combination in a highly variable and polarized fashion. Nearly half of these judges sentenced this guard-prisoner combination to less than a year in prison, while more than a third sentenced these offenders to more than five years in prison. Thus there is both selective leniency and severity in the sentencing of these trouble cases. There is support here for our third hypothesis of disparity, but it is more complicated than expected, with notable severity as well as leniency, and with results directed very specifically at the combination of Coalition guards and al Qaeda prisoners.

We now explore in further detail the second April 2005 group of Iraqi judges—the group who assessed cases including the nationality/military affiliations of guards and suspects. We carry out two types of analyses designed to pinpoint the effects of judge characteristics. First, we estimate a set of three pooled specifications that include all the ratings for the 19 judges and in which the regressors systematically add judge characteristics to the case characteristics. Second, we estimate a second set of three specifications in which the dependent variables are based on estimates from the judge-specific equations and the regressors are the judge characteristics.

Results of the first set of specifications are reported in Table 4. Specification 1 includes as regressors only the case characteristics. Specification 2 adds six judge characteristics: judge's age (linear and quadratic terms), assignment to discussion group 1, death rate in the judge's city of practice, and the judge's scores on the election, fear, and police protection attitudinal items. Specification 3 adds two interactions, between the coefficient associated with the Coalition-on-al Qaeda guard-prisoner combination and the fear and police protection measures, respectively. The estimates obtained in these three specifications are corrected for heteroskedasticity due to clustering by judge.

The results that are of particular interest in Table 4 involve the fear and protectiveness of the judges and the combined nationality/military group affiliations of the guards and prisoners, as at least partially anticipated by our fourth hypothesis drawn from a critical legal perspective. The omitted comparison group for the models summarized in Table 4 is the scenario in which Coalition guards are accused of torturing suspected Iraqi terrorists. Across the columns of Table 4, we see that in contrast to the omitted comparison grouping, the most severe sentences are imposed on Coalition guards accused of torturing al Qaeda suspected terrorists (b=4.9). Yet these coefficients also have relatively large standard errors and are therefore not statistically significant. On the other hand, the effect for a Coalition guard torturing a Coalition suspect (b=3.2) is significant. This finding is suggestive of the judges' inclination to severely sentence Coalition offenders. However, while the possibility of the latter cases occurring was included in this research because it is theoretically possible, this occurrence is presumably rare and therefore not particularly informative for purposes of this analysis.

We turn back to the Coalition–al Qaeda combination that is of greater interest, and in particular to the likely meaning of the large standard errors of these regression coefficients. Fear and protection are the judge-level sources of variation that a critical legal perspective deems likely to influence the sentencing process. The last columns of Table 4 bring such sources of judicial heterogeneity into the analysis by extending the previous regression equations to include the judges' characteristics introduced earlier. The fear and protection measures speak directly to concerns of the critical perspective that judges respond punitively to threats perceived in their environments. The measures of fear and protection are now centered to reduce consequences of multicollinearity in the creation of interaction terms. Recall that the descriptive statistics reported in Table 1 for these measures indicated that the Iraqi judges in this research are both highly fearful and very concerned about police protection, presumably in response to perceived threats.

As our fourth hypothesis and a critical legal perspective predicts, in the second column of Table 4 we see that the judges' severity in sentencing is significantly increased by scores on the fear scale (b=1.531, p<0.05) as well as by the importance they attach to the issue of police protection (b=5.671, p<0.05). Even more important for our explanatory interests, the final column in Table 4 tests the interacting influence of the judges' fears and protectiveness with the nationality/military group affiliation that is our particular focus of interest: the Coalition guards accused of torturing al Qaeda prisoners. The two interaction terms tested in the last column of Table 4 are both statistically significant: both fear (b=2.530, p<0.05) and the importance attached to police protection (b=4.892, p<0.05) substantially and significantly increase the severity of the sentences given to the combination of Coalition guards accused of torturing suspected al Qaeda terrorists.

The implication of our findings is that judges who most fear crime and most want police protection are normatively most strongly disposed to severely punish Coalition guards for torturing al Qaeda–suspected terrorists. Note further that this finding can help explain the variance displayed in the Iraqi judges' sentencing of this combined grouping. While all the judges in both program sessions rank high on fear and protection, those who are less fearful and protective are the judges most likely to sentence this nationality/military combination more leniently. This offers insight into the observed judicial variance in sentencing severity. That is, the fear and protection measures help explain why some Iraqi judges sentence this Coalition guard–al Qaeda suspect combination more severely, while other judges in this group sentence this combination more leniently.

The roles of judicial fears and protective concerns are expressed visually in the bar graph presented in Figure 1. This figure uses the final equation in Table 4 to estimate the average prison sentences given to Coalition guards for torturing al Qaeda prisoners by the second April group of Iraqi judges who scored lower and higher on the fear and protection scales. Recall that the judges overwhelmingly score high on these measures, so that even judges scoring lower are still fearful and yearning for police protection in Iraq. However, it is the highest-scoring judges (i.e., scoring 20 on the 20-point fear scale and five on the five-point protection scale) who sentence Coalition forces especially severely for torturing al Qaeda prisoners. Judges scoring high on the fear scale and at the mean on protection give on average more than 21-year sentences to these offenders, while judges scoring high on the protection scale and at the mean on fear give on average more than 25-year sentences to these offenders. By contrast, judges who score lower on fear or protection and average on protection and fear respectively give sentences of five and six years in length. However, the most extreme polarization is between judges who score high on both the fear and protection scales and judges who score lower on both these scales. The highest-scoring judges on the combination of fear and protection give Coalition torturers of al Qaeda prisoners average sentences of more than 34 years, while judges scoring lower on both these scales give average sentences to Coalition torturers of al Qaeda prisoners of just over four months. These estimates reveal extreme disparity in the sentencing of Coalition forces for torturing al Qaeda that is strongly linked to the fears and protective concerns of Iraqi judges.

Figure 1. Estimate of Sentences in Years by Levels of Judicial Fear and Concern for Police Protection

Our final set of analyses in Table 5 is designed to further confirm the striking effects of judicial fears and protective concerns on sentencing norms by using three judge-level regression equations. These equations provide a further test of the robustness of the cross-level, judge-case interactions just reported. The dependent variables in the three specifications are the differences between judge-specific slope coefficients. The first of these dependent variables is the difference between the highest and lowest judge-specific slope coefficients. The second dependent variable is the difference between the judge-specific slope coefficients for a Coalition guard who tortures an al Qaeda terrorist and the lowest of the remaining judge-specific slope coefficients. The third of these outcomes is the difference between the judge-specific slope coefficients for the sentence given to a Coalition guard who tortures an al Qaeda terrorist and a Coalition guard who tortures an Iraqi terrorist. The last of these equations is most directly relevant to our theoretical concerns, while the first two equations represent formats that speak more generally to the effects of the judges' reactions to the broader range of nationality/military combinations of guards torturing terrorists.

Note that because of the smaller number of judges (i.e., relative to cases) available at this aggregated level of analysis, the chances of observing statistically significant results are much reduced. Nonetheless, five of the six coefficients representing the impact of the judicial level fear and protection variables on the sentencing of guard-suspect combinations yield statistically significant effects at the 0.05 level in one-tailed tests. All three fear coefficients are statistically significant, and two of the three protection coefficients are also statistically significant. The judge-level models reinforce our confidence in the findings from the case-level models.

The Normative Judgments of Iraqi Judges

Supreme Court Chief Justice John Roberts remarked in his confirmation hearing that “Judges wear black robes because it doesn't matter who they are …. That's not going to shape their decisions” (cited in Reference RosenRosen 2006:27). This view, of course, is completely at odds with critical traditions of legal thought. Yet this kind of faith in judicial independence has led many Americans to believe that a global institutional diffusion of the democratic rule of law can solve many world problems.

We found little support for the idealized rule-of-law conception of judicial independence in war-torn Iraq, where the open texture of law interacted with characteristics of judges to produce highly variable outcomes in hypothetical sentencing decisions. Drawing on a critical legal perspective, we configured a factorial survey experiment to include consideration of the indeterminacy of torture law, the hierarchical power of Coalition prison guards, and the fear associated with their torture of imprisoned al Qaeda suspects.

The results of our experiment were mixed with regard to our initial four hypotheses. Overall, there was little support for our first hypothesis of a pervasive “climate of leniency” in the punishment of torture. Relatively long sentences were imposed, with average sentences from four to six years. Nonetheless, there was clear support for our second hypothesis of distinct trouble cases that produced wide variation in sentence length. There was also partial support for our third hypothesis, that the sentence disparities would involve lenient punishment for Coalition soldiers convicted of torture. Yet this disparity was more focused than expected—namely, on Coalition guards specifically convicted for torturing al Qaeda prisoners. Furthermore, the disparities went in the direction of both leniency and severity. Finally, there was support for our fourth hypothesis of the driving force of fear of crime and concern about police protection as sources of disparities, but in directions opposite to our expectations. Thus especially intense fear and protective concern led to severe sentencing of the torture of al Qaeda prisoners by Coalition soldiers, while less intense fear was associated with more-lenient sentencing of these Coalition soldiers.

The variability in the sentencing of the hypothetical cases can be understood partly as a response to the indeterminacy of law as reflected in the Bush administration advisory memos on torture, which initially and then less consistently included the assertion of a special presidential power for “harsh interrogation” of al Qaeda prisoners. Variance in the sentencing of Coalition guards who tortured al Qaeda suspects may also have been responsive to uncertainty about the classification of the latter as enemy combatants rather than as prisoners of war, in another contested area of legal indeterminacy. A critical legal assumption of indeterminacy anticipates fluctuation between severity and leniency, which occurs at the cost of certainty and consistency in legal decisionmaking.

Our factorial survey analysis of simulated sentencing decisions by Iraqi judges also supports the critical legal view that judicial behavior often involves a variable and selective response to fear. It is difficult to imagine a more fear-provoking set of structural and contextual forces than those that confronted the pre- and post-Saddam Iraq judiciary. Our data confirmed that these Iraqi judges feared the criminal correlates of their circumstances and yearned for police protection from them.

The hypothetical cases we posed to these judges clearly evoked conflicts and concerns involving sentencing offenders convicted of torturing suspected terrorists representing differing nationality/military combinations of combating parties. As noted, the judges responded with a polarized variability that led some to impose significantly more severe sentences on Coalition guards convicted of torturing al Qaeda prisoners, while others imposed more lenient sentences on these same guards convicted of torturing these same prisoners. The polarized variation in sentencing severity by itself is confirmation of important effects anticipated by a critical legal perspective on the indeterminacy of law. Our findings indicate that while nearly half of the Iraqi judges in our final experimental group sentenced Coalition guards to less than a year in prison for torturing al Qaeda prisoners, more than a third severely sentenced Coalition guards to more than five years in prison in the same kind of cases. Indeed, Iraqi judges who scored high on both fear and concern about police protection assigned sentences of more than 35 years to Coalition torturers of al Qaeda prisoners, while judges scoring lower on both of these scales gave sentences in these same cases of only about four months. We surmise that this extreme variability is at least partially a response to indeterminacy in torture law and the license it gave Coalition forces to harshly interrogate al Qaeda prisoners.

It bears note that the Iraqi judges did not completely abandon conventional sentencing norms. Their sentencing decisions sustained a semblance of formal rational commitment to the rule of law, for example, by responding to the degree of injury and harm indicated by days in the hospital resulting from torture. It is noteworthy that in all three columns of Table 4 presented above, days in the hospital had a persistent impact on sentence severity.

Still, at both the case and judge levels of analysis, we found quantitative empirical evidence that indeterminacy of the law, the hierarchical position of the occupying forces, and variation in fear of crime and a yearning for police protection were sources of selectively severe and lenient sentencing of cases involving Coalition torture of al Qaeda suspects. Greater fear of crime and concern for protection led to more severe sentencing of Coalition forces for torturing al Qaeda suspects, while reduced fears and protective concerns led to lenient sentences. This connotes a greater tolerance for the torture of suspected al Qaeda terrorists and their treatment as unlawful enemy combatants.

Media speculation in earlier phases of the Iraq war suggested that the Coalition torture of al Qaeda suspects created a vengeful motivation for their entry and involvement in the Iraq insurgency. It is also as if the Iraqi judges who severely sentenced Coalition torture of al Qaeda suspects shared the view of a New York Times editorial, namely, that “torture is a terrible way to do the very thing that the administration uses to excuse it—getting accurate information. Centuries of experience show that people will tell their tormentors what they want to hear, whether it's confessing to witchcraft in Salem, … or concocting stories about Iraq and al Qaeda” (“Illegal, Immoral and Pointless,” Unsigned editorial, New York Times, 10 Dec. 2005, p. A28). Coalition torture of the high-ranking al Qaeda leader Ibn al-Shaykh Libi is a frequently cited source of the now recanted and widely discounted claim that prewar Iraqis trained al Qaeda members (Reference JehlJehl 2005:A1). Furthermore, the Iraqi judges' severe sentencing of Coalition torturers of al Qaeda suspects may have been a reasoned response to their fear that this torture incited growing al Qaeda involvement in Iraq. The fear of al Qaeda may have overcome for these judges concerns about the power of the Coalition forces. The severe sentencing of these Coalition guards for torturing al Qaeda suspected terrorists was also consistent with according these suspects the protection mandated by the Geneva Conventions for lawful prisoners of war.

In a novel sense, the Iraqi judges who sentenced Coalition guards severely for torturing al Qaeda prisoners exercised an unanticipated judicial autonomy from the occupying powers. The attempt to try 25 CIA operatives in Italy for abducting and transporting a suspected terrorist in Egypt illustrates a highly publicized effort to encourage this kind of autonomy from U.S. influence in international settings (Reference FisherFisher 2007:A7). The Italian prosecutor in this case has lamented tolerance for torture tactics that defy international law by noting that “we don't need to give our enemies any reason for recruiting other members for their organizations” (Reference FisherFisher 2007:A7). This position can be understood as a plea for judicial autonomy.

On the other hand, the Iraqi judges who sentenced these same cases leniently may have acceded to the power of the Coalition leadership and its apparent view that the way to drive al Qaeda from Iraq was to torture them into leaving and informing on their peers. Even if these judges did not fully condone the torture of al Qaeda, they may have been more accepting of its use by Coalition forces and therefore less inclined to punish these torturers severely. Or they may simply have been more inclined to accommodate the powerful Coalition occupiers in hopes of professional career advancement (see also Reference Dezalay and GarthDezalay & Garth 2002a). We drew from a critical legal perspective to predict a pervasive leniency in the sentencing of torture cases by Iraqi judges, but our data require a more nuanced and contingent explanation that takes into account alternative kinds of reasoning. In the case of the military occupation of Iraq, it was the less-fearful judges who acceded to the U.S. accommodation of torture tactics.

There was little or no evidence in our findings that the January 2005 election in Iraq influenced sentencing decisions. Neither the occurrence of this election, between the November 2004 and April 2005 courses, nor the importance assigned in the surveys by individual judges to elections, notably influenced sentencing outcomes. Our design also provided a generic consideration of interethnic conflict including a variable indicating whether the torturer and suspected terrorist were from the same or different ethnic groups. The effect of ethnic variation between the torturer and the suspect was actually in the unexpected direction of lenience, although this effect was never statistically significant in Table 4. The ethnicity variable was also nonsignificant in its influence when we interacted it with the Iraqi–Iraqi prisoner–guard combination. However, the sectarian turn in the Iraq war may later have overtaken this aspect of our findings from an earlier phase of the occupation. As we have noted, it is also quite possible that the Coalition torture of al Qaeda suspects contributed to the growth in sectarian violence in Iraq, leading some judges to sentence these suspects severely.

Public attention in Iraq was concentrated during the last days of 2006 on the execution of Saddam Hussein by hanging. There is a theme that ties the Hussein execution to our research: a constitutional override provided the legal indeterminacy that allowed the Iraqi government to hastily conduct its frenzied execution. This execution is at the extreme end of a continuum of severity in the punishment of torture and larger crimes and may in itself become a lasting symbol of retributive justice in Iraq. Yet this unique and extreme case is uninformative about the judicial punishment of torture more broadly. Our study brought the use of torture by occupying Coalition forces into an analysis that revealed both punishment severity and leniency. The selectivity of the lenience we observed is perhaps the most notable finding in this volatile context: the judges who were less fearful of growing violence leniently sentenced Coalition guards for torturing al Qaeda suspects. The implication is that these less-fearful judges were freed by the indeterminacy of law to accommodate the imposed power of a forced democracy with sentences that revealed a tolerance for U.S. torture policies.

The introduction of democracy with rule-of-law ideology in Iraq is has emerged as a negative case in the worldwide diffusion of American institutions (see also Reference Dezalay and GarthDezalay & Garth 2002b). Further research will be needed to widen the scope conditions beyond our study for a more generically framed critical legal perspective. The most obvious scope conditions of the current case study involve the circumstances of occupied Iraq. It is important to recognize that although Iraq is an extreme case of social and political pressures imposing their influence through the indeterminacy of law, threats to formal rational predictability are common in the rule of law. Using a critical legal perspective framed in more European and Weberian terms, Reference SavelsbergSavelsberg (1992) demonstrates this point in a contemporary American case study. Savelsberg shows that determinate sentencing reforms, operationalized in the United States through the implementation of federal and state sentencing guidelines, have resulted in charging practices that counter guideline goals and reintroduce new forms of substantive rationality that again render sentencing law indeterminate. Legal indeterminacy is a ubiquitous, malleable, and frequently flawed instrument for furthering agendas of democratic reform.