Classification review is carried out to decide and establish the confinement facility and the plan of individual treatment of sentenced inmates, and thus to help their successful return to society through proper living under confinement. In Korea, there are two types of classification index, one is Security Treatment Level Classification Index and the other is Correctional Recidivism Prediction Index (CO-REPI). The Security Treatment Level Classification Index was developed in order to determine the level of treatment for inmates, including confinement facility and safe guard level, education, and living condition, based on the inmates’ risk of escape or absconding from the facility, etc. This index is used to attain the goals such as to separate high risk inmates from others, and provide a differential treatment to those who are predicted to be rehabilitated relatively easily or to have greater possibility of improvement (Kim Dong-hyun, 2017). On the other hand, the CO-REPI was developed to distinguish different risk levels of inmates based on the tested elements of prediction about their readmission to the confinement facility and to use the results in the classification review for new inmates, reclassification/reassessment, and application for release on parole and has been enforced since March 1, 2012 (Ministry of Justice, internal source, 2011)

The objectives of this study are to check the validity of the current classification indexes, Security Treatment Level Classification Index and CO-REPI, and to diagnose the issues, if any, to improve them. For such objectives, we first examined the indexes’ background of development and purpose of use, method of development and component items, as well as the classification index used in England and the USA at the correctional stage. To verify the statistic validity of the indexes, we reviewed the predictability of total score or individual score in each section and the contribution of each item to prediction of recidivism, and thus identified inadequate items. Further, we made a recommendation to improve the classification indexes related to determination of confinement security level or parole review, through consultation and discussion with experts and the TFs of relevant authorities, and by collecting opinions of the frontline officers at the correctional facility classification division.

The theoretical model adopted in the study is the Risk, Need, Responsivity model ("RNR model"), which explains the risk and risk factors that prisoners present, focusing on the treatment of criminals (Andrews, Bonta, 1998). Here, risk refers to the level of recidivism of admitted inmates, and the risk factors refer to the factors related to recidivism more concretely. Risk assessment is a process that determines the level of recidivism risk by using the risk factors, and if a proper tool that is acknowledged internationally is used, the level of recidivism can be assessed and be assigned to specific levels. Such assignment allows the classification officers to decide the level of management and control over and the intensity of treatment programs (such as the period of the program) for subject inmates. Risk can be assessed as either static or dynamic factors. Generally, static factors which represent the consequences of an action shows higher prediction about recidivism than dynamic factors which are related to emotion or perception. Needs are distinguished between the needs that are and that are not likely to trigger a crime. Although the needs that are likely to trigger a crime is the totality of risk level of a criminal, and of dynamic attribute, the change of needs is not always related to recidivism. Risk can be interpreted as the total sum of risk factors because a crime is an consequential act from the aggregation of all risk-triggers of crime and the possibility of commit a crime is the risk. Responsivity is an influential element which either the person in charge of a treatment program or the inmates who are subject to the treatment program have.

The current classification indexes have been criticized as the items are mostly quantitative and composed of static factors, and therefore they pose a difficulty to assess the change of risk of inmates (Bonta, 1999), as well as to reflect the psychological assessors’ opinion who interview, observe, and uate the inmates. Moreover, the re-uation of the two indexes are mostly composed of scores, without including the dynamic factors which measure the changes, if any, caused by their living in correctional facility. Other issues include that the re-uation index of the two indexes is consisted of almost the same content, that there are still some remnants of the progressive treatment system in the re-uation index, and that the system is operated heavily based on the result of administrative report, rather than the risk of inmates themselves. Some even question if classification review requires two indexes, Security Treatment Level and CO-REPI, at all. The most serious issue in using those indexes is that never has been a validity test conducted up until now, its tenth anniversary of being applied in the field.

From this context, the current study has obtained cooperation of the Correctional Service and, for verifying the validity of the current classification indexes, applied the CO-REPI to the cases of the inmates, among those who were admitted to confinement facility and then released between 2012 and 2021, who were not re-admitted to the facility or did not commit a crime again on or before July 31, 2021, and thus extracted the initial admission and re-admission information material of those released inmates. The number of sample inmates that met the first data extraction criteria were 96,077. Of those, the second random sampling was conducted based on the type of release (expiration of prison term, or parole), sex, age, history of admission, type of crime, and term of imprisonment, according to the Korea Correctional Service Statistics 2020. The number of sample inmates were 9,929, of which repeat offenders were 1,592 (16.0%).

The study reviewed whether the two indexes properly determine the risk (recidivism) level of prisoners, and has found out that they both predict recidivism, properly. To verify the validity of each question item, we further analyzed the relationship between the variables by putting all items simultaneously, as well as analyzed each item by putting each item, individually. Under the premises that cross-validation is satisfied, the researchers reviewed the results of two analyses, and interpreted that ① after controlling the relationship between variables, the items that predict recidivism meaningfully are highly valuable, ② the items which predict recidivism in a static direction when the relationship between variables is not controlled but lose prediction when the relationship is controlled are somewhat valuable but the assessment method should be improved, and ③ the items that have troubles in predicted direction and prediction itself even when they are put in individually should be reconsidered of its usage as they seem to have no relevance to recidivism prediction.

As for the Security Treatment Level classification index, S3 or S4 level appeared to have higher prediction than S1. Especially, S4 showed twenty-two times higher in recidivism risk(odds) than S1, and therefore its treatment level was confirmed to be valid. Of the sixteen items that compose the Security Treatment Level classification index, the items that predicted recidivism meaningfully even after the relationship between variables were controlled include motivation of crime, seriousness of crime, recidivism period score, initial age score when charged by the police, the total period score of being confined in the facility, and punishment score in the past confinement. In the meantime, the variables that constitute risk assessment for dynamic factors and improvement assessment, as well as escape or violation failed to predict recidivism meaningfully. The items which predicted recidivism in a static direction when the relationship between variables are not controlled but loses prediction after the relationship were controlled include pretrial attitude score, number of criminal committment, score of committment of less than serious or serious crimes, REPI dynamic assessment score, and probation effectiveness score. The items that showed some issues in prediction direction and prediction itself when putting individually to the assessment include usage of weapon score, total imprisonment period score, possibility of improvement score, and attempt score to escape or suicide. However, as for re-uation index, only the REPI question items, of those used in the current indexes, were found to predict recidivism, and all the other items, such as correction treatment performance score, score of imprisonment period as sentence execution, correction performance score, score of possibility to cause troubles, score related to punishment, and consideration score for individual treatment, failed to predict recidivism in a meaningful way. Therefore, the team has concluded that it is necessary to re-review the Security Treatment re-uation index, each item’s construct idea and assessment method and to improve and reinforce the reliability and validation of those items.

Based on the final ratings, in the CO-REPI it has been found that R2 is three times, R3 is eight times, R4 is twenty times, R5 is forty times higher in risk of recidivism than R1. As for the total score of CO-REPI, the initial uation score has higher prediction than re-uation score. In initial uation, when the standard deviation was raised by one point (that is, the total score is raised by 6.368 points), the recidivism risk was increased by 2.44 times. Of the twenty-three variables that constitute the CO-REPI, those which predicted recidivism meaningfully even after the relationship between variables were controlled include sex, name of crime, relationship to the victim, punishment of the current and previous, if any, crime, frequency of the same or similar crime committed, age at the time when the initial sentence was confirmed, age when the subject was released concerning previous crime, experience of punishment during the school years, recidivism environment after being released, tendency to wrongdoing, and tendency to renouncement score. The variables which predicted recidivism in a static direction when the relationship between variables were not controlled, but lost prediction or revealed prediction direction after the relationship were controlled include the total imprisonment period of previous crime, number of probation cancelled in effect, history of receiving mental hospital treatment, attitude toward authority, and possibility score of recidivism by specific factors. Lastly, the variables which showed some issues in prediction direction and prediction itself when put individually include the mental state at the time of committing crime, recidivism period, number of times living with partners, occupation at the time of committing crime, economic and habitation states, respectively, before being admitted to the confinement facility, and education score. As for the re-uation index, just as the Security Treatment Level index, only the security treatment scores predicted recidivism meaningfully when the relationship between variables were controlled. Correction performance, punishment related items, and failure to pay fine or assignment of additional fine were meaningful only when put individually, and the correctional treatment performance score did not predict recidivism even when put in individually.

Furthermore, this study conducted the survival analysis to see if the Security Treatment Level classification index and CO-REPI would predict the recidivism period. Survival analysis considers the time that a target inmate takes until he or she recommits a crime, and shows whether each rating, item, and variable would affect that time. The rating and total score of the two indexes predict the time until recidivism reasonably. In both indexes, compared to S1 and R1, all the other ratings appeared to have higher aggregated risk toward recidivism (meaning, the crime committing interval becomes shorter) as time passed.

With all these results putting together, this study could conclude that in both indexes, static factors predict recidivism better than dynamic factors, and especially factors related to previous commitment of crime have highest prediction in all. As for the CO-REPI, socio-demographic factors predict recidivism most poorly, and thus the application of socio-demographic characteristics to recidivism prediction, except sexes, should be reconsidered.

Because it is necessary to interpret the results of testing the statistic validity of the indexes, this study also reviewed the issues that the frontline workers have in the course of their enforcement of the classification index and the areas that appeared to require improvement based on the validity test results, by means of consultation with the field workers at the Correctional Service and external experts. First, concerning the issues that the frontline workers have in their using and operating the Security Treatment Level classification index, they have pointed out that history of previous crime is given a lot of weight, that the content validation of dynamic factors is low, that standardized re-uation index is being used and used inadequately for the objectives, and that the security level and treatment are used in a mixed condition. As for the items that require improvement, from the result of testing the validity some concern has been raised out that the serious and less than serious crime committment item might produce an unstable statistic result because the sampling of this research does not include the high-scorers whose total imprisonment period is at least ten years, and the item’s statistic result might be only produced because of prisoners who have higher risk of recidivism despite shorter period of imprisonment sentence depending on criminality and type of crime. Nevertheless, this study argues that although the relevant item does not predict recidivism well, they need be maintained as they are useful items for the treatment of inmates. Further, this study maintains that the items concerning possibility of improvement and troublesome action require subjective uation, but because the enforcers are afraid of the risk of subjectivity, an objective uation standard has been prepared regularly after the enforcement of classification indexes. Since the dynamic factors of the Security Treatment Level classification index are uated not by the same standard but by a method mixed in time line, the possibility should be considered that recidivism was not predicted in a meaningful direction due to contamination of measuring tools. Also, this study has suggested that since it is possible that uating the dynamic factors at the stage of new inmates uation has resulted in unstable statics verification, it would be more reasonable that the dynamic factors be moved to re-uation index and at the stage of new inmates’ uation only static factors be uated. However, in order to measure the degree of change, a standard point is required and thus dynamic factors should be included, too, at the stage of new inmates’ uation. This study has also suggested a dual-system in which we reduce the dynamic factors and organize them as item to assess the current risk and security level/degree of confinement, while the possibility of rehabilitation or personal change should be left to CO-REPI for measurement. The study comments that in considering the security level, escape or violation classification area does not have important variables or sample cases and thus would not have produced a meaningful result, and suggests to move the relevant items to the assessment standards of other items. It also points out that in case of ‘score of probation in effect’, if a case occurs even after the security treatment level is determined, the treatment level is adjusted to be a lower level, and therefore elimination of the item should be also considered.

The opinions about the CO-REPI were widely collected through the interviews with the frontline officers of Correctional Service and external experts, and their view on the issues concerning the use and operation of the index, as well as the survey conducted over the staff in charge of the improvement of the CO-REPI at the Classification Center of the Korea Correctional Service, Seoul District. In this, it has been pointed out that the use and operation of the CO-REPI has some other issues, such as unreasonable prediction for the future, human right violation and other discriminative elements, in addition to the same issues as the Security Treatment Level classification index (unfit use to the objectives, and too much weight on the criminal history). To summarize the opinions of the frontline staff about the prediction of recidivism, the verification results which did not predict recidivism even when each item related to socio-demographic factors, such as the number of times living with partners, occupation at the time of committing crime, economic and habitation states, respectively, before being admitted to the confinement facility, and education, was put separately, might not reflect the reality adequately because the item are graded based on the inmates’ voluntary reporting response, and are not able to reflect the change of time especially in living with partner- score or education score. Also, according to their view, the recidivism prediction is not meaningful because of the low content validation questions. As for the phenomenon that predict recidivism negatively under the condition that the items of recidivism possibility by specific factor are inserted into all questions, the interviewees and experts believe that because those are the questions to uate the tendency of specific crime of high recidivism rate based on criminal history and criminal records of inmates, their verification ability would be unstable for specific factors to recommit a crime.

In addition, the frontline workers and experts have pointed out, persistently, that the two current classification indexes do not fully investigate the dynamic factors. Since the dynamic factors of the current classification indexes are found to be insufficient to predict recidivism, this study urges that the concept of the dynamic factors to be measured through the classification indexes be established more reasonably, and the uation method be refined further. Another problem is the low face validity, meaning that as for the items which show unstable explanations, such as possibility of troublesome action or of improvement in the Security Treatment index, recidivism risk by specific factors in the correction psychology test, and attitude toward authority, the titles of the items themselves are generally abstract, their application scope is too wide, and their measurement targets are not easily apparent. It is worth reconsidering if the use and employment of such items up to the present is theoretically reasonable.

To suggest a method to improve, this study attempts, first, to apply the RNR principle to the Korean classification indexes. Static risk factors, such as inmates’ criminal history and the age at the time of his or her first commitment of crime, and criminal punishment record during one’s school years cannot be changed. Nevertheless, dynamic risk factors which are one’s own internal and psychological factors can be changed in content, and therefore they sometimes become the target area of the correctional treatment programs. It is likely that the majority of the items which are classified, in the current classification index, as dynamic factors (such as recidivism environment after being released, risk of recidivism by specific factors, risk of wrongdoing at the security treatment level, possibility of improvement, correctional treatment performance of the re-uation index, correction result, and possibility to commit trouble action) cannot be regarded as a dynamic factor strictly, in terms of their construct idea itself, and they rather assess the risk factors which cannot be changed, or non-crime trigger factors or achievement, than assess crime-trigger factors which can be changed. Therefore, it would be necessary to reorganize the current dynamic factors in order to use the dynamic factors for their proper objectives.

As a result of reviewing the RNR principle and verifying the statistic validity of the indexes, the research team of this study, having gathered and put all the opinions and views of the frontline workers and external experts together, would like to make the following suggestion for improvement of the indexes. First, the classification indexes should be started from a one completely integrated index, regardless of the specific purpose, and then its assessment area be either reduced or expanded depending on the purpose. In the case that the two indexes are integrated. the problem of overlapping areas, which has been criticized continuously, can be solved. Also, if the questions are composed mostly of factors of dynamic nature, it will be more efficient as there is no need to organize new inmates uation index and re-uation index, separately. At each assessment, we also suggest that, an uation table should be prepared and the assessing officers specify in the table the type of uation, such as new inmates uation or re-uation, uation in parole decision, and assessment of release, etc. and manage the index values, consecutively.

Crime trigger factors used in the classification indexes are summarized as Big 8 factors, and can be categorized into the key crime trigger factor (anti-social awareness, anti-social peers, anti-social action history, and anti-social personality type) and the potential crime trigger factor (family/marriage, school/workplace, hobby, and drug abuse)(Andrews, Bonta, Wormith, 2004). The integrated classification index that this study suggests can be consisted of, primarily, the key crime trigger factor, the potential crime trigger factor: Socio-demographic characteristics and living pattern) and the potential crime trigger factor: Psychological attitude. First, the key crime trigger factors uate anti-social action history, a static factor, that measures the number of crime commitment, age at the time of first committing a crime, and the seriousness of the crime, all of which are relevant to criminal history, as well as anti-social awareness (attitude toward crime, law-abiding attitude, and attitude toward penalty, etc.), anti-social personality (hostility, impulsiveness, aggressiveness, emotional apathy, and psychopathic tendency, etc.), and anti-social peers (social isolation, friends/acquaintances with criminal history), which are dynamic factors, at the same time. This study further suggests that under the premises that the security treatment standard (security level) can be changed depending on inmates, the confinement level of inmates be determined based on the key crime trigger factors that are measured. If a relevant authority tries to separate the security level and treatment level, the security level should be determined based on the anti-social action history of the key trigger factor, while anti-social awareness, personality, and peers could be adopted to decide the treatment level.

As for the CO-REPI, this study suggests to form three areas by organizing both the key and the potential crime trigger factors as uation items. Considering that the treatment performance should be reflected in the uation of parole decision, it is necessary that we increase the dynamic factors. The first area should use the same key trigger factors that are used in distinguishing the security treatment level, and the second area should measure socio-demographic characteristics and living pattern, such as family/marriage, employment and education, leisure time, alcohol/drug abuse/mental disorder. The third area is related to the psychological state of inmates. Here, the psychological state could be composed of four sectors which best predict the action of an offender, regardless of the type of crime he or she has committed, (cognitive distortion, emotional problem, interpersonal relationship issues, lack of problem-solving/ self-management technique), and then the question items would be either organized to reflect the inmate’s state at the time of uation or constructed by using the existing psychological scale. The second and the third potential crime trigger factors are consisted of static elements in recidivism prediction and relevant to crime induction, although the degree of induction is lower than that the key trigger factors. With the existing index, it was impossible to measure the change of state at each time of uation because the questions were mostly concerning inmate’s socio-demographic characteristics before being admitted to confinement facility. On the other hand, if the questions are organized newly with the upgraded index, the changes depending on the time of uation can be measured and reflected to assessment.

In conclusion, the key argument this study would like to make is to separate the security treatment levels of inmates based on the key crime trigger factors (Security Treatment Level classification index), to include the potential crime trigger factors in uation as well, and thus to apply any changes discovered in the process to the risk of recidivism assessment of target inmates (Co-REPI). A further study would be required to organize and improve the questions, standard method to rating and score system, matrix for calculating the ratings, and method to adjust the ratings. A user manual that provides technical support and details in relation to the use and adoption of the classification indexes should be completed, gradually, as well through follow-up researches.