The potential future violent conduct of capital offenders is a frequent juror consideration at capital sentencing. In the post-Furman v Georgia (1972) era of death penalty determination, “whether there is probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society” (see Jurek v Texas, 1976) is available as a statutory or nonstatutory aggravating factor in many jurisdictions. In two states, Texas and Oregon, the jury must answer this issue in the affirmative to sentence a defendant to death. The obverse, that the defendant is likely to have a positive and nonviolent adjustment to prison, can be raised as a mitigating factor at capital sentencing (see Skipper v South Carolina, 1986). There are also indications that the sentencing verdicts of capital jurors are influenced by their appraisal of the prospect of future violence from capital defendants, whether or not this issue is raised by the prosecution or the defense (Blume & Garvey, 2001; Cunningham & Reidy, 1998b; Geimer & Amsterdam, 1988; Sandys, 1991; Costanzo & Costanzo, 1992, 1994).

The quest for reliable determinations of violence risk at capital sentencing

Is the confidence of the court well-placed?

The U.S. Supreme Court has ruled that both unaided juries and mental health experts can reliably identify capital offenders who will be violent in the future (see Jurek v Texas, 1976; Barefoot v Estelle, 1983). This confidence appears to arise, in part, from expectations that these determinations can be reliably made through reliance on common sense (Cunningham, 2006). Employment of “common sense” judgments focusing on factors such as the heinousness of the capital offense, past conduct in the community, and antisocial personality traits has intuitive appeal. Research, though, on the rates and correlates of prison violence, investigations of predictive accuracy, identification of error-inducing conceptual strategies, and emerging correctional data cast doubt on such conventional wisdom (see American Psychological Association, 2005; Cunningham, 2006; Cunningham & Reidy, 1999; Dawes, Faust, & Meehl, 1989; Edens et al., 2005; Shaw, 1978).

Predictive accuracy of capital juries

Data that would illuminate the predictive accuracy of capital juries in their determinations of “whether there is a probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society” are illusive. This is principally a function of a lack of information regarding how individual jurors or respective capital juries interpreted this special issue in terms of either the operational definition of violence (alternatives addressed subsequently) or the context of prediction (community vs. prison). Regarding predictive contexts, only prison is relevant or measurable. Current capital life inmates face multi-decade prison confinement or life-without-parole,Footnote 1 and many of the relevant death-sentenced inmates have been executed or remain on death row. As a result, there is not a cohort of inmates who were sentenced to death under the special issue, gained relief from a death sentence, served a multi-decade term (if parole eligible) and subsequently paroled (but see Cunningham & Reidy, 1998b, for parole recidivism rates for capital offenders sentenced pre-Furman).

The data are limited that illuminate the predictive accuracy of jurors who may have contemplated the singular practical context of prison in their application of the special issue. Marquart, Ekland-Olson, and Sorensen (1989) reported on 92 capital offenders in Texas who had gained relief after having been sentenced to death under the special issue. Averaging 6.3 years in the general prison population when their disciplinary records were retrospectively reviewed, only 10% of these inmates had been cited for serious institutional violence and just one percent had killed another inmate.

Predictive accuracy of mental health experts

The ambiguity in behavioral dispositions or predictive contexts assumed by mental health experts testifying at capital sentencing renders a determination of their predictive accuracy correspondingly illusive. For reasons outlined above, only the prison behavior of the capital offenders about whom the experts opined is available for analysis. Such an analysis is reported by Edens et al. (2005), who retrospectively reviewed the disciplinary files of 155 capital defendants sentenced to death in Texas following expert testimony asserting their “future dangerousness.” If contemplating a prison assault requiring more than first aid treatment, these expert predictions and subsequent agreements by the respective juries were unrealized 95% of the time.Footnote 2 This error rate was consistent whether the defendant had remained on death row during an era that allowed inmate contact with other inmates, or had gained relief from the death sentence and spent decades in the general prison population. Despite the poor predictive performance of these risk assertions by mental health experts and their associated methodology of unstructured subjective judgments (see American Psychological Association, 2005; Cooper & Werner, 1990; Cunningham, 2006; Cunningham & Reidy, 1999; Edens et al., 2005), there is evidence that such testimony strongly influences capital jury determinations (Krauss & Sales, 2001; Krauss & Lee, 2003; but see also Montgomery, Ciccone, Garvey, & Eisenberg, 2005).

The anchoring role of base rates of prison violence

Recommendations for improving the accuracy of violence risk assessments at capital sentencing have specified that these estimates should be anchored to context-relevant violence base rate data that are conservatively particularized with empirically demonstrated factors (American Psychological Association, 2005; Cunningham, 2006; Cunningham & Reidy, 1998a, b, 2002; Cunningham & Goldstein, 2003). Base rate data regarding the prison violence of capital offenders have been reported (e.g., Cunningham, Reidy, & Sorensen, 2005; Edens et al., 2005; Marquart & Sorensen, 1989; Marquart et al., 1989; Reidy, Cunningham, & Sorensen, 2001) and reviewed (e.g., Cunningham, 2006; Cunningham & Reidy, 1998b, 2002; Marquart, Ekland-Olson, & Sorensen, 1994). These base rate data are summarized in Table 1.

Table 1 Assaultive rule violations of capital inmates and comparison inmates
Full size table

The scale and specificity of broader base rate data regarding prison violence of relevance to capital defendants have substantially expanded in recent years. Large scale studies have reported prison violence rates as a function of sex (Harer & Langan, 2001), sentence length and type (Cunningham & Sorensen, 2006a, b; Cunningham, Sorensen, & Reidy, 2005); offense of conviction (Sorensen & Cunningham, 2006 under review), and custody level (e.g., Cunningham & Sorensen, 2006a, in press; Texas Department of Criminal Justice, 2006).

To summarize these base rates of prison misconduct, the majority of capital offenders are never sanctioned for serious violence when confined in the general prison population (see Cunningham, 2006; Cunningham & Reidy, 1998b, 2002; Marquart et al., 1994).Footnote 3 Most studies regarding convicted murderers and life-without-parole inmates have reported that they are not a disproportionate source of prison violence (Cunningham, Reidy, & Sorensen, 2005; Cunningham & Sorensen, 2006b; Sorensen & Cunningham, under review-a; but see Wolfson, 1982; Sorensen & Cunningham, in press). The frequency and prevalence rates of prison violence among capital offenders and other inmates fall dramatically as the severity of the assault increases. These general trends are observed in various correctional jurisdictions, pointing to their broad generalization and capital application. Shared characteristics in architecture, procedures, and demographics among correctional jurisdictions in the United States provide a conceptual rationale for broad generalization of these base rate data (see Cunningham & Reidy, 2002).

Whether females exhibit different base rates of assaultive misconduct in prison appears to be related to the severity of the assault, though the findings are mixed. Some studies (e.g., Baskin, Sommers, & Steadman, 1991) have reported that females have much lower rates of prison assaults. Others (e.g., Harer & Langan, 2001; Sorensen & Cunningham, in press) have reported similar rates of fighting and minor assaults for females in prison, but much lower rates of serious assaults and no prison homicides. Still others have reported equivalent rates of assault of all severities for male and female inmates (Cunningham & Sorensen, 2006a,b).

Individualizing base rate data

The availability of reliable base rate data that can serve as anchors in capital violence risk assessments, however, is only an essential first stage. There remains the challenge of quantifying the specific influence of various factors on the anchoring base rates. A number of potential particularizing factors have not been demonstrated to be helpful in this regard. Diagnosis and testing appear to offer little contribution at present to capital risk particularization. There is inadequate data to support a conclusion that intuitive risk indicators such as Antisocial Personality Disorder or PCL-R psychopathy are reliably predictive of serious violence in American prisons (Cunningham & Reidy, 1998a; Cunningham, Sorensen, & Reidy, 2005; Edens et al., 2005; Edens, Petrila, & Buffington-Vollum, 2001). Objective personality assessment such as the Minnesota Multiphasic Personality Inventory (MMPI, MMPI-2) and the Personality Assessment Inventory (PAI; Morey, 1991) have fared no better (see Cunningham & Reidy, 1998b, 2002; Cunningham, Sorensen, & Reidy, 2005; Edens et al., 2005). The performance of violence risk assessment measures such as the HCR-20 (Webster, Douglas, Eaves, & Hart, 1997), Level of Service Inventory—Revised (LSI-R), and VRAG (Quinsey, Harris, Rice, & Cormier, 1998) in predicting serious violence in American prison have been similarly disappointing (Cunningham, Sorensen, & Reidy, 2005; Edens et al., 2005).

Another approach to particularizing risk of violence in prison has been to utilize logistic regression to identify pre-confinement demographic or history factors that are correlated with assaultive misconduct, and the relative influence or “odds ratios” associated with these factors. These can then be combined to create scales or models to specify increasing categories of risk. Exemplars of this approach include scales generated from 6,390 convicted murderers in Texas prisons (Sorensen & Pilgrim, 2000); 2,595 inmates in a high security prison in Missouri (Cunningham, Sorensen, & Reidy, 2005); and 13,341 newly admitted inmates in the Florida Department of Corrections (Cunningham & Sorensen, 2006a).

These studies have not only significantly expanded the available data on base rates of prison violence, but also quantified the influence of correlates of prison misconduct and violence (see also Grendreau, Goggin, & Law, 1997). Such correlates include age (inversely), community stability (e.g., educational attainment), prior prison history, and drug conviction (inversely). Among murderers, a concurrent robbery or burglary and prison gang membership have also been correlated with assaultive prison misconduct. Other research, though, has provided mixed support for some of these correlates. Cooper and Werner (1990) did not find educational achievement to be related to violent misconduct in prison, and Cao, Zhao, and Van Dine (1997) found that prior incarceration did not have “an appreciable effect on the severe rule violations” (p. 109).

Race of the inmate has been reported as a predictive factor for violent prison misconduct (e.g., DeLisi et al., 2004; Harer & Steffensmeier, 1996), but is an unconstitutional consideration at capital sentencing (Saldano v Texas, 2000). Accordingly, this characteristic has been excluded from the predictive models for prison violence in a number of studies (e.g., Cunningham, Reidy, & Sorensen, 2005; Cunningham & Sorensen, 2006a,b; Sorensen & Pilgrim, 2000).

Operational definitions of “criminal acts of violence”

Central to the jury’s determination of “whether there is a probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society” is their definition of the qualifying acts. Little guidance has been provided by federal appellate Courts, who have declined to operationally define the associated terms (see James v. Collins, 1993). Further confusing the determination is an unfortunate adoption of terminology by the Court (Estelle v Smith, 1981; Barefoot v Estelle, 1983) and some scholars (e.g., Claussen-Schulz, Pearce, & Schopp, 2004) that treats a “future dangerousness” condensation of this issue as synonymous with probability of criminal acts. Dangerousness, however, is a state or descriptor that arguably applies to virtually all capital offenders, if not almost all violent felons, and thus does little to individualize the application of the death penalty. As Cunningham (2006) posed the question, under what circumstances would a convicted capital murderer not be considered dangerous at sentencing as compared to non-criminal community members? Further, while probability of acts is amenable to both scientific measurement and preventative interventions, “dangerousness” is not. Accordingly, it appears closest to the issue as affirmed in Jurek, most amenable to individualized determination, and most accessible to scientific measurement to conceptualize the question in terms of acts of varying severity in particular contexts rather than as an over-arching state of dangerousness (see Cunningham, 2006; Cunningham & Reidy, 1998b, 2002). For reasons discussed previously, a prison appears to be the only relevant and measurable context for assessments of the “probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society.”

With a focus on future behavior, an argument can be made that “continuing threat to society” envisions acts of sufficient severity that a sanction of death is a proportionate preventative intervention (Cunningham & Reidy, 1998b; Edens et al., 2005). Further, if defined too broadly (e.g., belligerence, mutual fighting) the incidence may be so ubiquitous that the issue again loses any particularizing or individualizing value in the application of the death penalty (see Cunningham, 2006). That said the operational definitions of aggressive prison misconduct or violence utilized by researchers have varied widely. Some authors have restricted their study of prison violence to aggressive acts resulting in serious injuries, while others have included physical altercations of any sort. To compensate for low base rates of serious violence in prison and to increase predictive power, researchers have often broadened the definition even further to include acts with violence potential, such as verbal threats, possession of a weapon, or escape-related misconduct – regardless of the outcome (for an expanded explanation of this methodological issue, see Cunningham & Sorensen, 2006a,b; Edens et al., 2005). Obviously, disaggregating aggressive misconduct by severity is most informative of both the associated rates and the respective predictive value of correlates. Such severity-specific rates also help differentiate probability from mere possibility.

The current study

Utilizing a sample of recently incarcerated capital murder offenders, the current study provides base rate data specific to the initial phase of confinement of these prisoners that is disaggregated by the severity of the violence. The study further examines how the operational definition of prison “violence” influences the calculation of its base rate, the identification of correlates, and the ability to construct an actuarial risk assessment instrument. This is the first reported investigation of an actuarial violence risk scale specific to capital offenders.

Methodology

Participants

The current study is based on 136 male inmates indicted for a capital murder committed in the year 2000, entering TDCJ after February 1, 2001, under a sentence other than death. These capital offenders were drawn from a larger sample of inmates who had been convicted of murders occurring during the calendar year 2000 (see Sorensen & Pilgrim, 2006). Females and death row inmates were excluded from the current sample of capital offenders for a number of reasons. Both female and death-row inmates are held, respectively, under conditions of confinement that are distinct from other capital offenders. Such marked differences in conditions of confinement render their assault rates difficult to interpret. Also problematic, previous analyses demonstrated that death sentence and sex variables were not consistent predictors (see Cunningham, Sorensen, & Reidy, 2005; Sorensen & Cunningham, in press). Further, the female and death-sentenced inmates failed to commit any assaults resulting in serious injuries, making it impossible to estimate the influence of their group membership on that particular form of violent outcome.

Detailed breakdowns of violent incidents and the extent of harm incurred were not routinely recorded by TDCJ until in February 2001. Accordingly, capturing the period of initial confinement with violent misconduct data disaggregated by severity required restricting the sample utilized in the current study to inmates entering TDCJ after February 1, 2001. Inmate files that included demographic, offense, conviction, and institutional information were provided by the Texas Department of Criminal Justice and hand-searched. While the observation period ranged from 6 to 40 months for the sampled capital murderers, the average time “at-risk” was 2.37 years. Information related to the demographics, offense, conviction, and criminal history of this cohort of 136 incarcerated capital murderers is presented in Table 2.

Table 2 Age, criminal history, and offenses committed by the Year 2000 Cohort (N = 136)
Full size table
Table 3 Disciplinary violations involving actual or potential violence among the year 2000 Cohort
Full size table
Table 4 Outcome measures used to assess inmates’ violence risk
Full size table

Measures

As noted earlier, a variety of operational definitions have been used in studies of violent or assaultive misconduct in prison (Cunningham & Sorensen, 2006; Delisi & Munoz, 2003; Edens et al., 2005). The broadest measure includes a range of rule violations that have the potential to result in a violent outcome. Table 3 provides an overview of the types of violent or potentially violent rule infractions occurring among the cohort of capital murderers incarcerated in TDCJ that could be used in determining base rates, identifying correlates, and constructing the actuarial assessment instrument. These included misconduct impacting on institutional security (i.e., weapon possession, riot, and escape); aggression against other inmates (i.e., threaten to harm, fighting, extortion, and assault); and aggression against officers (i.e., threaten to harm and assault). The violence-related disciplinary offenses were aggregated by severity in Table 4. These formed the outcome measures of all potentially violent acts, assaultive violations, and assaults with serious injuries. The category of “All potentially violent acts” reflects the totality of the violence-related misconduct reported in Table 3. “Assaultive violations” includes all assaults, regardless of severity. “Assaults with serious injuries” extracts from this broader category any assaults that resulted in injury requiring more than first aid treatment.

Results

Base rates

The frequency or incidence of specific violence-related misconduct is reflected in the first column of Table 3. The percentage of inmates involved in, or prevalence of, particular types of violations during February 2001 through May 2004 is indicated in the second column. The third column standardizes the incidence of offending as a yearly rate per 1,000 inmates. Fighting was both the most frequent (99 per 1000 inmates annually) and most prevalent (19.1%) of these infractions, followed by assault on staff with an annual frequency of 84 per 1,000 inmates and a prevalence of 9.6%.

Two observations are readily apparent from inspection of these infractions aggregated by severity in Table 4. First, rule violations involving potential violence had a significant annual rate (357 per 1,000 inmates) and were committed by a substantial minority of capital inmates. Just over 36% of the cohort was involved in some form of rule violation with violence potential during their initial period of incarceration. The inmates involved in potentially violent rule violations committed an average of just over two infractions during the observation period of 2.37 years. Second, both the prevalence of involvement and frequency of occurrence decrease as the severity of the violence increases.

A further breakdown of the assaultive rule violations in Table 3 and Table 4 was undertaken to provide additional detail about the nature and severity of these offenses. The annual frequency rate of assaults on officers in Table 3 (.084) is quite high relative to other studies: Florida 2003 staff assault = .002 (Sorensen & Cunningham, under review); Florida 2003 combined inmate and staff assaults in close custody = .055 (Cunningham & Sorensen, in press); Missouri 1991–2002 combined inmate and staff assaults among life-without-parole inmates = .096 (Cunningham, Reidy, & Sorensen, 2005); and Texas 1974–1988 staff assaults among capital life inmates = .0156 (Marquart et al., 1989). It should be noted, though, that only 7 of the 27 assaults involved weapons. One inmate committed two of these acts, while another was responsible for five of the assaults on staff with a weapon. Upon further inspection, it became clear that the staff assaults by this latter offender were mainly nuisance offenses. Disciplinary offense indicators for three of the five included the descriptor “CH,” which refers to chunking, defined by TDCJ as “the throwing of some substance…” Further, this one inmate was responsible for more ‘than 34 total disciplinary offenses and more than one-fourth of the total combined staff and inmate assaults (11 of 41), though none resulted in serious injury. This illuminates the extent to which one troublesome inmate can influence the overall frequency and yearly rate of rule violations, particularly in a modest-sized sample. Because of the influence of this “outlier” inmate, the number of acts and the yearly frequency rates of various types and severities of violence in Tables 3 and 4 were recalculated excluding this inmate and are reflected in parenthesis. The outlier does not influence any of the remaining analyses because they are dichotomous measures of prevalence of occurrence.

Of the 9 assaults listed as resulting in serious injuries in Table 4, all were committed against inmates. None of the 27 assaults on staff resulted in serious injuries. This finding is consistent with a study of convicted murderers in TDCJ, where the annual frequency rate of aggravated assaults on staff was .0011 (Sorensen & Pilgrim, 2000). Only one of the attacks on an inmate resulting in serious injury involved a weapon. This offense, as well as most of the other inmate-on-inmate “assaults” included the descriptor “FT,” which indicates that the inmates were mutual combatants. Although it cannot be determined from the electronic database, it is possible that the injured loser was also cited for fighting, with the inmate who was more successful in inflicting injuries cited with an assault. These findings suggest the need for further disaggregation, as well as more detailed data collection and offense specification, when examining violent misconduct in prison.

Application of actuarial methods

The next step in the analysis involved an attempt to utilize easily identifiable characteristics at entrance to prison to construct a scale that could be used to successfully predict prison violence. A scale, referred to herein by the acronym of its journal publication outlet- JCLC, previously developed to predict prison violence among 6,390 murderers incarcerated in the Texas prison system during the 1990s (Sorensen & Pilgrim, 2000), was used as the starting point. The specific criteria utilized in that scale and their respective weightings are detailed in Table 5. One variable included in the original JCLC scale, security group membership (i.e. prison gang affiliation), was unavailable because TDCJ no longer allows researchers to record this information. Accordingly, it was deleted from the adaptation of the original JCLC scale as depicted in Table 5. Though the absence of this variable is unfortunate for the replication of the original scale, whether a particular defendant will join an inmate gang is often unknown at the time of a capital murder trial, and as such, the variable has been found to be of little assistance in real-world assessments of risk of prison violence.

Table 5 Original JCLC scale weightings and the new reduced Burgess scale
Full size table

In calculating the JCLC scale, the weightings from a logistic regression model were summed and added to the expected base rate of .164, the overall score from the original scale indicating the probability that a capital murderer sentenced to a 40-year life sentence would commit an aggravated assault on or kill an inmate or correctional staff member. These weights are reflected in the second column of Table 5. Consistent with the procedure used in the construction of the original scale, overall scores were split into six risk levels based on .08 probability increments of the scale scores. None of the inmates in the current study attained a score that would place them in the highest risk category.

Table 6 presents the prevalence of violent rule infractions among the current cohort by risk levels of the JCLC scale. The prevalence of violent misconduct, disaggregated by severity, is reflected by the first row of each level. The second row for each of the five risk levels provides the raw numbers used in calculating the prevalence of violence, (number of inmates who committed violent acts/total number of inmates)×100, at each risk level. With the exception of the missing information on prison security threat group membership, the analysis of data presented in Table 7 serves as a validation of the original instrument. The model AUCs suggest that the JCLC scale was moderately successful at predicting potential violence or actual assaults among the year 2000 cohort. The scale was least successful at predicting assaults in general (AUC = .612, n.s.), but much more successful at predicting assaults that resulted in serious injuries (AUC = .755, p < .05). While it is impossible to know whether the scale would have performed better if the missing information on security threat group membership had been included, an item analysis revealed that some of the remaining indicators from the JCLC scale did not assist in predicting violence among the year 2000 cohort. The two variables that were not found to be related to any of the violent outcomes in this study were the concurrent killing of multiple victims and additional contemporaneous murder attempts or aggravated assaults occurring during the course of the capital murder. Further, the age categories 26 to 30 years old and 31 to 35 years old were not significantly different from the reference category (21 to 25 year olds), or related in an opposite from expected direction, in predicting violent outcomes.

Table 6 Prevalence of violent rule infractions among the year 2000 Cohort by levels of risk based on the JCLC scale (N =136)
Full size table
Table 7 Prevalence of violent rule infractions among the year 2000 Cohort by levels of risk based on the reduced Burgess Scale1 (N = 136)
Full size table

Development of a reduced Burgess scale

Because some of the items included in the original scale did not assist in predicting violent outcomes, a parsimonious version of that scale was constructed herein. Those items that were not significantly related to outcome were excluded, while the remaining items that significantly predicted violent outcomes were retained. It was also decided that since the original weightings of these items in the JCLC scale were sample specific, herein each item would be weighted equally. This adaptation of the original JCLC scale to a reduced Burgess scale is demonstrated in the third column of Table 5. A simple additive Burgess scale has been demonstrated to be nearly as successful at predicting behavioral outcome as more complex forms of scaling (Gottfredson & Snyder, 2005; Cunningham & Sorensen, 2006). Further, a Burgess scale has advantages in its ease of application and interpretation, and is especially appropriate when being applied across samples. A number of other factors found to be significant predictors in previous studies (see Cunningham & Sorensen, 2006; Cunningham, Sorensen, & Reidy, 2005; Sorensen & Pilgrim, 2000) were also analyzed in an attempt to improve upon the simplified scale’s (termed: Risk Assessment Scale for Prison—Capital [RASP-Cap]) predictive power, but to no avail. These included additional crime-specific distinctions (type of capital murder; weapon used; other aggravating case features), inmate personal characteristics (I.Q., marital status), and criminal history (prior violent offense record, behavior in jail, prior number of convictions, criminal justice status at the time of the offense).

The resulting RASP-Cap scale includes four indicators found to be significantly related to violent outcomes from the original JCLC scale. One item relates to the circumstances of the capital murder. Those inmates who committed capital murders involving a contemporaneous robbery or burglary were scored +1. The second item relates to the criminal history of the inmate. Those with prior records of prison incarceration were given +1. The last two items were related to the age of the offender at entrance to prison. Those age 35 or below were given +1 and those under the age of 21 were given an additional +1, so that those over age 35 scored a 0 on age, while those 21 to 35 scored +1 and those under 21 scored +2. The simplified Burgess scale resulted in an overall score of 0 through 4, although once again none of the cases attained the highest risk score of 4 (i.e., under 21 years of age, with a prior record of incarceration, and committing capital murder during the course of a robbery or burglary). The lowest two risk categories were combined, as only a small number (n = 6) of inmates scored zero (i.e., over 35 years of age, with no prior prison record, and committing a capital murder that was not in the course of a contemporaneous robbery or burglary). No one in the two lowest scoring groups (i.e., scoring 0 or 1) was cited for an institutional assault. The prevalence of violent acts for risk levels corresponding to these scores is presented in Table 7.

As noted by the model AUCs in Table 7, the simplified Burgess Scale is moderately successful in its predictive ability for all of the violent outcomes. Those in the highest level of risk were inmates under the age of 35 with a prior record of prison incarceration whose capital murder involved a contemporaneous robbery or burglary. Inmates with those characteristics were much more likely than other inmates to commit violent rule infractions. In fact, more than half of these higher risk inmates committed a potentially violent act during their first 28 months, on average, of being incarcerated. During the same time period, one-quarter committed an assault and more than one-tenth committed an assault resulting in serious injuries. Six of 7 inmates who committed assaults resulting in serious bodily injury were categorized as higher risk inmates by the scale. In comparison, lowest risk inmates, those scoring either 0 or 1 on the Burgess scale, committed only three potentially violent acts, but no assaultive rule violations during the observation period. Those in the medium risk level were close to, but slightly below, the average prevalence for the entire cohort in their commission of violent rule infractions.

As shown above, the prevalence and frequency of violence in the prison are influenced by its operational definition. The correlates of prison violence, however, may or may not be influenced by its operational definition. Nonetheless, a limited number of correlates were found to be useful predictors of prison violence, however defined, in an actuarial assessment.

Discussion

This study examined the rates and correlates of officially cited violent misconduct among 136 capital murder offenders during the initial phase (M = 2.37 years) of their life sentences in the Texas Department of Criminal Justice. The characteristics of the sample were unsurprising. Consistent with the well-established relationship between youthfulness and violent criminal offending, almost two-thirds of these offenders were less than 26 years old and 86% were less than 31 years old. These capital offenders had “lighter” criminal histories than other TDCJ inmates. Only 26% had served a prior prison sentence, compared to 37.2% of the inmates in the general prison population of Texas prisons (TDCJ, March 2001). The proportion of life-sentenced capital inmates with prior prison histories changed little in the last two decades. Marquart et al. (1989) reported that 31% of the capital life inmates in their post-Jurek sample had served a prior prison term. For most of these capital offenders, then, their criminal histories were characterized by a rapid escalation to capital murder in early adulthood without prior recourse to prison confinement. A substantial minority (22%) had committed multiple contemporaneous murders.

An important feature of this study was the disaggregating of potential and actual prison violence by specific misconduct, and by severity (i.e. potentially violent acts, assaults, assaults with serious injury). This convention embraces recent recommendations (see Edens et al., 2005), and also best illuminates base rate data. Again consistent with other research on prison violence (Harer, 1992; Sorensen & Cunningham, under review), as the seriousness of the violent misconduct increased, the frequency rate dropped sharply and many fewer inmates were involved. To demonstrate these declining prevalence rates, during the study period 37% were involved in potentially violent misconduct, 19% were in a fight, 14% perpetrated an assault, 5% assaulted with serious injury, and none committed a prison homicide. These rates, of course, are based on incidents that resulted in disciplinary reports. It is acknowledged that unreported and “unsolved” inmate-on-inmate violence occurs in prison at an unknown rate. It is hypothesized that such unreported/unsolved violence would have greater impact on rates on the less severe portion of the violent misconduct continuum. In other words, it is more likely that threats and fights go unreported than assaults with serious injury or homicides.

Determining whether these offenders have a disproportionate incidence of violent misconduct in prison relative to other inmates is problematic. Because violence is more likely to occur early in a prison sentence (see Cunningham, Sorensen, & Reidy, 2005; Reidy et al., 2001; Flanagan, 1981), comparisons with inmates who are deeper in their sentences are likely to be misleading. A comparison of a sample of general population inmates who were in their initial phase of confinement in TDCJ during this same interval was not undertaken. However, data from the initial period of TDCJ incarceration of life-sentenced capital inmates, non-capital murderers, and manslaughter offenders found that capital offenders comparatively had higher rates of assaults whether with or without serious injury (Sorensen & Cunningham, in press).

As an additional cautionary note in the interpretation of these base rate data, the rate of violent misconduct observed during the initial years of prison incarceration is unlikely to continue at an unabated pace. Rather, after rising during the initial period of confinement, rates of prison disciplinary infractions begin to decline (see Cunningham, Sorensen, & Reidy, 2005; Flanagan, 1981; Sorensen & Pilgrim, 2000; Sorensen & Wrinkle, 1996). This is likely a product of the combined effects of adjustment to confinement, aging, early failure by inmates most at risk, and the associated individualized security measures applied to these problem inmates. Stated more succinctly, it would be a grave error to assume that the 20-year prevalence rate would be ten times the two-year prevalence rate.

In a continuing effort to distill correlates of institutional violence that are available prior to sentencing, this study examined factors that had previously been specified by Sorensen and Pilgrim (2000) from a study of over 6,000 convicted murderers sentenced in the 1990’s. Of these factors, prison gang membership is no longer coded by TDCJ for external researchers and thus was unavailable for analysis. Two other factors, multiple murder victims and contemporaneous attempted murder or assault, were not predictive with the current sample. Interestingly, though some trend was observed for a continuous age-related effect, age was significant as a predictor only at the extremes of the distribution: less than 21 years old and older than 35 years of age. A stronger continuous effect of increasing age has been reported in studies of much larger samples of high security inmates (Cunningham, Sorensen, & Reidy, 2005), convicted murderers (Sorensen & Pilgrim, 2000), and general population prisoners (Cunningham & Sorensen, 2006a). The limited number of inmates in the various age-related cells may be responsible for the more limited precision demonstrated by the current data. Having a high school diploma or equivalent has been associated with reduced incidence of assaultive prison misconduct in other studies (Harer & Langan, 2001; Cunningham & Sorensen, 2006a; Cunningham, Sorensen, & Reidy, 2005), but this factor was not reliably specified for the current sample and thus its affect could not be analyzed.

In spite of not being inclusive of some promising predictive factors, application of a reduced Burgess scale based on the limited set of pre-confinement factors available was modestly successful. These variables were age, prior prison commitment, and contemporaneous robbery or burglary. The predictive power was similar across a range of misconduct severity, ranging from AUC = .715 to .766. The same model was similarly effective, then, in predicting inmate misconduct of various severities. Importantly, the model was most successful in predicting assaults of greatest interest to criminal justice researchers, correctional professionals, and capital juries: those resulting in serious injury.

Before becoming overly enamored with the predictive potential of this scale, however, it is important to note the relatively low prevalence or base rate of serious injury assaults at all risk levels of the scale. Though inmates scoring in the higher risk level were far more likely to perpetrate a serious assault than those at lower risk levels (11.5% vs. 0.0%–1.8% prevalence rate), almost nine out of ten higher risk inmates did not perpetrate a serious assault during an observation period averaging 2.37 years. In this sense, the scale may have greatest utility in identifying inmates who are particularly unlikely to perpetrate assaults in prison. Inmates in the lowest risk category had no assaults of any severity, and only 1.8% of inmates at risk level 3 perpetrated serious assaults. This application of varying improbability is consistent with broader prison risk assessment findings regarding capital offenders (Cunningham, 2006). For this reason, the labels associated with the respective risk levels have been identified by number; rather than low, medium, and high risk.

This study of life-sentenced capital murderers was restricted to offenders who were allowed to plead to capital life terms or were sentenced to capital life terms by juries. A contributing factor for offenders who were offered pleas bargains may have been a judgment by prosecutors that the offenders in question were not disproportionate institutional threats. Similarly, life rather than death sentences at trial may have reflected the determination of their juries that there was not a “probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society.” Accordingly, an argument could be made that the prison outcomes of a “capital life” sample are not informative of capital offenders who are not offered pleas or whose probability of committing future “criminal acts of violence” remains to be determined at trial by their juries. In support of a conclusion that the data from this study do not generalize to broader population of capital inmates pre-sentencing, Sorensen and Cunningham (in press) reported that TDCJ death row inmates exhibited higher rates of assault on staff than life-sentenced capital offenders during their initial periods of confinement. This finding, however, is less straightforward in its implications than might appear. One interpretation is that offenders sentenced to death under the special issue are more disposed to violence in prison. Another is that the Spartan and highly restrictive conditions that currently characterize the Texas death row unit may serve to induce rather than inhibit inmate misconduct against staff (see Lyon & Cunningham, 2006; Toch, 2001).

Other studies provide support for an argument that these findings regarding life-sentenced capital inmates generalize to a pre-trial capitally-charged population. For example, Marquart et al. (1989) did not find significant differences in the elements of capital murder, weapon choice, relationship to victim, prior criminal histories, or other discernible characteristics of those cases where the capital juries answered “yes” as opposed to “no” to the special issue regarding future violence. Further, the inmates who had gained relief from their death sentences exhibited similar, though modestly lower rates of assault when in the general prison population than inmates who had been sentenced to capital life terms at trial. In another study, death-sentenced inmates who were “mainstreamed” in the general prison population in the Missouri Department of Corrections displayed assault rates that were quite similar to inmates serving life-without-parole sentences in the same facility (Cunningham, Reidy, & Sorensen, 2005). Admittedly, these Missouri death-sentenced inmates were not sentenced under a mandatory consideration of the special issue.

There is an obvious “Catch 22” problem in obtaining directly comparable base rate data and associated actuarial models with a capital population sentenced under the special issue. Data from life-sentenced capital offenders are not directly comparable because these offenders had not been identified at trial as having a probability of committing “criminal acts of violence that would constitute a continuing threat to society.” Data from death-sentenced inmates are not directly comparable because these condemned inmates are held under super-maximum conditions that are distinctly different from those that these inmates would have encountered had they been sentenced to capital life terms and placed in a general prison population.

Generalization of the RASP-Cap beyond Texas capital offenders is supported by the predictive potential of these same factors with other inmate samples, including convicted murderers (Sorensen & Pilgrim, 2000; Sorensen & Cunningham, in press), high security inmates (Cunningham, Sorensen, & Reidy, 2005), and newly admitted general population inmates (Cunningham & Sorensen, 2006a). The scale, however, has not yet been validated with other capital offender samples. Accordingly, it should be considered a research instrument at present.