Re-Arrest Patterns Amongst Individuals Screened with the LSI-R:SV in Salt Lake County Jail

Summary

Making use of instruments that assess individuals’ risk-to-reoffend is becoming standard practice in managing those under correctional supervision. Here we follow a large sample of individuals screened for their criminogenic risk and needs via the Level of Service Inventory-Revised: Screening Version (LSI-R:SV) released from Salt Lake County jail between 2017 and 2019. We find that statewide re-arrest rates follow the expected pattern wherein moderate and high risk individuals experience higher rates of re-arrests than those screened as low risk, with findings being fairly consistent when examining outcomes for males and females alone. Specifically we find that:

• Individuals classified as low risk experienced half the rate of re-arrests than those classified as high risk when the definition of a new arrest includes either a Misdemeanor or Felony offense. Those classified as low risk experienced an average re-arrest rate of around 30% compared to 60% for those classified as high risk.

• The average re-arrest rate after jail release significantly declined when the definition of a new arrest is restricted to a new Felony offense and further increases differences in re-arrest rates between those classified as low and high risk. Low risk individuals experienced a new arrest rate of 11% in comparison to 29% for those screened as high risk, a difference of nearly 60%.

• The LSI-R:SV effectively distinguished low risk females from those classified as moderate and high risk. Consistent with prior research, we find further support that the strength of the LSI-R:SV is to distinguish low risk individuals from those classified as moderate and high risk, with less accuracy distinguishing new arrests between moderate and high risk individuals for females for when the outcome is restricted to a new Felony offense only.

• While the share of females classified as high risk was higher than for males, high risk females experienced fewer instances of new arrests than high risk males. Females classified as high risk experienced an overall re-arrest rate of 55% vs 63% for high risk males.

• Current risk classifications appear appropriate for Utah’s jail population though considerations around adding a very low risk classification could be beneficial for certain criminal justice decision points. Survival analysis around each risk score revealed that current categories correlate well with actual instances of new arrests but that the low risk classification may be separated into two distinct categories.

Overall, the LSI-R:SV did well in distinguishing new arrest rates by risk classification amongst Utah’s jail population. Further analyzing and refining the instrument’s ability to accurately assess sustained criminal activity remains an important and on-going area of research.

Introduction & Background

The use of actuarial risk and needs screenings and assessments is becoming standard practice in the criminal justice setting. As part of broader criminal justice reform efforts, in 2015 Utah implemented a statewide risk and needs process that uses the LSI-R:SV to screen for an individual’s risk-to-reoffend when booked in any of its county jails. The LSI-R:SV is part of a family of risk and needs screenings and assessments. The screening version was developed to quickly identify and triage individuals who may require a more extensive risk assessment (i.e., the LSI-R). This shorter risk and needs screening version is an 8-item actuarial tool that classifies individuals into three distinct categories: low, moderate, and high risk to reoffend. It includes seven important risk areas, namely the individual’s:

• criminal history
• criminal attitudes
• criminal associates
• personal/emotional
• employment stability
• family relations, and
• substance abuse (that interfere with employment)

In the jail setting, the information obtained by the risk classification can be used for:

• determining need for a full risk and needs assessment
• guiding jail housing decisions (prevents mixing low and high risk populations), and
• assigning supervision levels if placed under community supervision.¹

While research on the efficacy of the LSI-R:SV has highlighted its strengths as effectively distinguishing low risk individuals from moderate and high risk, predict future recidivism, and reserve scarce correctional resources by reducing the time spent administrating the tool (Lowenkamp, Lovins, & Latessa, 2009), there is a continuing need to examine the instrument’s ability to distinguish new criminal behaviors by each risk classification. To address this need, this research explores re-arrests patterns amongst individuals who were screened for their criminogenic risk and needs by the LSI-R:SV that were released from Salt Lake County jail between 2017 and 2019. We further make recommendations on the recommended risk classifications by analyzing survival curves for each individual risk score.

Literature on the LSI-R:SV

Literature on how well the LSI-R:SV predicts sustained criminal behaviors amongst the jail population as a whole is exceedingly limited. When examining specific sub-populations, previous research has found the LSI-R:SV to be predictive of a variety of outcomes that are critical in managing individuals under correctional jurisdiction (Olver, Stockdale, & Wormith, 2014; Mccafferty & Scherer, 2017). For example, among studies looking at probationers, the LSI–R:SV scores accurately predicted violent recidivism and violations while under community supervision (Yessine & Bonta, 2006). Among individuals who were incarcerated, the LSI–R:SV scores predicted success in institutional misconduct (Walters & Schlauch, 2008; Walters, 2011). In regards to recidivism, based on assessed level of risk among individuals, research on the LSI-R:SV found that the arrest rates from a sample of probationers who was classified as low risk was 15%, moderate risk at 34%, and high risk at 50% (Lowenkamp, Lovins, & Latessa, 2009). When recidivism rates were defined as a return to prison, low risk individuals recidivated 13% of the time, moderate risk individuals at 29% and high risk individuals as 39% of the time (Lowenkamp, Lovins, & Latessa, 2009).

Literature on risk and needs screenings and assessments not only show support in the utilization of these tools, but there is an equal emphasis on these tools being normalized and validated to the specific population as well as providing quality training in administration (Lowenkamp, Lovins, & Latessa, 2009). To achieve this, scholars suggest that the standard cut-offs points to delineate low, moderate, and high risk individuals may need to be adjusted to accurately capture a specific population (Flores et al., 2006; Lowenkamp, Lovins, & Latessa, 2009). General areas in refining the tools predictive ability further include the relative subjective nature in the scoring which can be standardized through targeted and on-going training.

Data & Risk Classifications

Records of 25,686 unique LSI-R:SV screenings that took place between 2017 and 2018 in Salt Lake County jail were matched² to arrest records from the Bureau of Criminal Identification.³ The risk screenings and their respective classifications are considered valid for 1-year or until a significant new event occurs. For the present study, this implies that the follow-up time should not exceed 1-year while taking into account the time spent in jail prior to release. To maximize the balance between retaining a large sample of screenings while maintaining an adequate follow-up time, we exclude those that were held in jail for more than 90 days, which allows for a 6-month follow-up time while staying within the 1-year screening “expiration date”. We examine how the LSI-R:SV risk classifications correlate with actual re-arrest patterns, using a 6-month follow-up time for 2 distinct outcomes: 1) a new arrest that occurred after jail release that includes either a Misdeamenor or a Felony offense and 2) a new Felony offense only.⁴

Risk Classifications & Re-Arrest Rates

Figure 1 shows the percent classified as low, moderate and high risk to reoffend for our sample. Consistent with the literature, the majority of individuals were classified as moderate risk (~50%) with 26 and 22% classified as high and low risk respectively.

Figure 2 shows the actual 6-months re-arrest rates by risk level for when the event is defined as either a Misdemeanor or a Felony vs a Felony offense only. For the instrument to be informative, then as a general rule, we expect those that were classified as low risk to have fewer instances of re-arrests than those classified as moderate and high risk. For both outcomes, the highest re-arrest rates occur for those classified as high risk, followed by moderate and then low risk individuals, with these rates being significantly lower for a Felony offense only. Specifically, for our first definition, the highest rate of re-arrest occurred for those screened as high risk (~60%), followed by moderate risk individuals (~50%), and low risk at (~30%). For our second definition, these rates are reduced to 29% for high risk, 21% for moderate and 11% for low risk. For our first outcome, the median time between the risk screening took place and jail release was 2 days while the median time before a new arrest occurred was 43 days (min=1, max=182). Additional summary statistics is available in the Data Table below.

Figure 1: Risk Classification at Jail Booking

Figure 2: 6-months Re-arrest Rates by Risk Classification

Gender Analysis

Figure 3 shows the percent classified as low, moderate and high risk to reoffend separated by females and males. The majority of both females and males were classified as moderate risk (~50%). Females were on average, higher risk than males, and experience similar though slightly lower rates of re-arrests across the three risk classifications. This difference is most noticeable for high risk females which experienced a re-arrest rate of 55% in comparison of 63% for high risk males.

Figure 3: Risk Classification by Gender & 6-months Re-arrest Rates

Survival Analysis

As the name suggests, survival analysis, or time-to-event analysis examines the time it takes for an event to occur. In criminal justice research, this event is often an indicator of a new arrest, conviction or incarceration. The Kaplan-Meier (KM) estimate is commonly used in the survival analysis literature. A nice feature of the KM estimate is its ease of interpretation, which shows the share of individuals that “survived” until some time t. At each point along the survival curve, the probability of surviving is expressed as the number of events that occurred divided by the number of individuals that have yet to experience the event.

Survival Curves

Figures 4-9 show Kaplan-Meier survival curves (with a 95 percent confidence interval), overall, by risk level, and by selected group characteristics for when the outcome includes a Misdemeanor or a Felony offense. The steepest part of the survival curve shows at what time the highest rate of a new event occurs, or for our purpose, a new arrest. For group comparisons, in instances when two or more survival curves overlap, there is no statistical difference in survival time across groups.

As expected, the highest rates of failures or new arrests occur around jail release. This is seen by the relative steepness of the curves. The difference in survival across groups is clearly seen amongst the three different risk levels. Specifically, the curve for low risk individuals is significantly flatter than for both moderate and high risk individuals, which is consistent throughout the entire follow-up time of 6-months. Those classified as high risk depicts the steepest curve. The difference in survival between the 3 different risk levels is further seen when the sample is focused on gender (Female and Males).⁵ Furthermore, and consistent with the literature, age appears important in avoiding a new arrest. This is seen by the increased survival probability amongst those over the age of 50, starting at around day 60 after jail release.

Figure 4: Survival Curve (Overall)

Figure 5: Survival Curves by Age: 50 and Above vs Below 50

Figure 6: Survival Curves by Age Group

Figure 7: Survival Curves by Risk Classification

Figure 8: Survival Curves by Gender: Females

Figure 9: Survival Curves by Gender: Males

New Felony Arrest Only

Figures 10-15 show Kaplan-Meier survival curves (with a 95 percent confidence interval), overall, by risk level, and by selected group characteristics for when the outcome is restricted to a Felony offense only. Overall, survival after jail release follows the patterns of the previous analysis with the exception of an overall increased survival probability. One noteworthy difference is however seen amongst the female population. While the LSI-R:SV was able to distinguish survival time between moderate and high risk females in our prior analysis, those classified as moderate and high risk show similar survival until around day 50 after jail release as shown by their overlapping confidence intervals.

Figure 10: Survival Curve (Overall)

Figure 11: Survival Curves by Age: 50 and Above vs Below 50

Figure 12: Survival Curves by Age Group

Figure 13: Survival Curves by Risk Classification

Figure 14: Survival Curves by Gender: Females

Figure 15: Survival Curves by Gender: Males

Data Table

	mean	sd	min	max
New arrest: Misd. or Felony	0.47	0.50	0.00	1.00
New arrest: Felony only	0.21	0.41	0.00	1.00
LSI-R:SV	2.04	0.69	1.00	3.00
Male	0.73	0.44	0.00	1.00
Age	36.14	10.51	18.11	86.67
Age 50 or below	0.88	0.33	0.00	1.00

Table 1: Summary Statistics

Risk Score Threshold Analysis

In this section we use survival analysis to explore appropriate cut-off points for each risk classification by analyzing each risk score. While findings indicate that those classified as low risk were re-arrested at half the rate as those classified as high risk, the share of low risk individuals that ended up having a new arrest within 6-months of jail release is an area of exploration. Though the classification of low risk does not mean no risk, this risk classification is an important area for further research as it can carry weight around release and supervision decisions. As mentioned, because literature on how well the LSI-R:SV predicts sustained criminal behaviors amongst the jail population as a whole is exceedingly limited, it is challenging to make precise expectations around arrest rates amongst each risk classification. We can however use statistical analyses to see whether different risk scores correlate with different rates of survival after jail release.

Survival Curves by Risk Score

When looking at our first outcome (a new arrest due to either a Misdemeanor or a Felony offense), a significant difference in survival is seen between individuals that got a score of 0 versus those that scored 2 points (seen in Figure 16). An additional classification of very low risk versus low risk could be considered in where 0 points is classified as very low risk while 1-2 points is classified as low risk. This re-classification appear less meaningful for our second outcome (a new arrest restricted to a new Felony offense) in where the current threshold appear more appropriate (seen in Figure 17).⁶ Another noteworthy difference is survival time amongst those that scored between 3 and 5 (our moderate risk group), which show distinct survival probabilities for each separate risk score.

Figure 16: Survival Curves by Risk Score

Figure 17: Survival Curves by Risk Score: New Felony Offense Only

Figures 18 and 19 show how the actual re-arrest rates by risk level as we add an additional risk classification of very low risk that includes those that got a score of only 0.⁷ With this exploratory adjustment, the re-arrest rate for low risk individuals measures at 30% while the re-arrest rate for our hypothetical very low risk group measures at 19%. As seen in 18, while the very low risk group only include 2% of our sample, 2% is a large number of individuals, which on average has a 11 percentage point difference in re-arrest rates than those in the modified low risk group. Further refining this analysis is recommended to find the optimal thresholds for each risk classification in Utah taking into consideration the objective for such analyses.

Figure 18: Risk Classification at Jail Booking
(modified cut-off points for low risk)

Figure 19: 6-months Re-arrest Rates by Risk Classification
(modified cut-off points for low risk)

Gender Analysis

Figures 20-23 show survival curves by risk score for females and males for both of our outcomes. Based on these survival curves, there appear to be 3 groups amongst females for our first outcome (A new Misdemeanor or Felony offense). A possible re-classification for females may be to classify those who scored between 0-1 points as low risk, 2-3 points as moderate risk, and those that scored between 4-8 points as high risk. In terms of males, the very low risk category discussed in the prior section seem appropriate. For our second outcome (a new Felony offense only), overall, there is less distinction between each individual risk score and survival when the outcome is restricted to a new Felony offense only.

Figure 20: Survival Curves by Risk Score: Females

Figure 21: Survival Curves by Risk Score: Females - New Felony Offense Only

Figure 22: Survival Curves by Risk Score: Males

Figure 23: Survival Curves by Risk Score: Males - New Felony Offense Only

Moving Forward

While our current data included information on statewide arrests in Utah after jail release, information regarding whether an individual relocated to another state remains unknown, which may under count actual re-arrest rates. Additionally, our matching algorithm identified a new arrest (if one occurred) within one day after jail release. This precludes the possibility of having multiple arrests within a 24 hour period, which may further under count new arrests amongst this very high risk population. Other unknown information concerns programming that may have occurred after the risk and needs screening took place (both inside the jail or after release), which have the possibility to influence actual re-arrest rates. The findings from this study should therefore be seen as preliminary.

In moving forward, research may examine how well the tool predicts additional definitions of new criminal behaviors, including, the nature (e.g., violent vs non-violent offenses) and count of the new arrests that occurred. While findings from this study show the benefits in screening individuals for their criminogenic risk level to better understand their likelihood of new criminal behaviors, it further highlights the need for these tools to be paired with professional judgement and the circumstances surrounding each individual case.

Acknowledgements

We would like to extend our gratitude to our partner agencies at Salt Lake County Information Services and the Bureau of Criminal Identification for sharing their data and expertise throughout the creation of this study.

References

Andrews, D.A., & Bonta, J. (2006). The psychology of criminal conduct (4th ed.). Newark, NJ: LexisNexis/ Matthew Bender

Ferguson, A. M., Ogloff, J. R., & Thomson, L. (2009). Predicting recidivism by mentally disordered offenders using the LSI-R: SV. Criminal Justice and Behavior, 36(1), 5-20.

Flores, A. W., Lowenkamp, C. T., Holsinger, A. M., & Latessa, E. J. (2006). Predicting outcome with the Level of Service Inventory-Revised: The importance of implementation integrity. Journal of Criminal Justice, 34(5), 523-529.

Goel MK, Khanna P, & Kishore J. (2010). Understanding survival analysis: Kaplan-Meier estimate. Int J Ayurveda Res 2010;1:274–8. Available at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3059453/

Lowenkamp, C. T., Lovins, B., & Latessa, E. J. (2009). Validating the level of service inventory—Revised and the level of service inventory: Screening version with a sample of probationers. The Prison Journal, 89(2), 192-204.

Mccafferty, J., & Scherer, H. (2017). Beyond Recidivism: Exploring the Predictive Validity of a Correctional Risk Assessment Tool on Offender Victimization. The Prison Journal, 97(6), 674-691.

Olver, M., Stockdale, K., & Wormith, J. (2014). Thirty Years of Research on the Level of Service Scales: A Meta-Analytic Examination of Predictive Accuracy and Sources of Variability. Psychological Assessment, 26(1), 156-176.

Vose, B., Cullen, F. T., & Smith, P. (2008). The empirical status of the Level of Service Inventory. Fed. Probation, 72, 22.

Walters, G. D., & Schlauch, C. (2008). The psychological inventory of criminal thinking styles and level of service inventory-revised: Screening version as predictors of official and self-reported disciplinary infractions. Law and Human Behavior, 32(5), 454–462. https://doi.org/10.1007/s10979-007-9117-5

Walters, G. D. (2011). Predicting recidivism with the Psychological Inventory of Criminal Thinking Styles and Level of Service Inventory-Revised: Screening Version. Law and Human Behavior, 35, 211-220.

Yesberg, J. A., Scanlan, J. M., Hanby, L. J., Serin, R. C., & Polaschek, D. L. (2015). Predicting women’s recidivism: Validating a dynamic community-based ‘gender-neutral’tool. Probation Journal, 62(1), 33-48.

Yessine, A. K., Bonta, J. (2006). Tracking high-risk violent offenders: An examination of the national flagging system. Canadian Journal of Criminology and Criminal Justice, 48, 573-607.

Yukhnenko, D., Blackwood, N., & Fazel, S. (2019). Risk factors for recidivism in individuals receiving community sentences: a systematic review and meta-analysis. CNS spectrums, 1-12.

---

1. Research has shown that correctional resources are best spent on moderate to high risk individuals while those that are low risk often benefit from little to no supervision↩

2. The sample of screenings included those that had a valid State ID. The State ID is a unique and often times the preferred identifier in Utah’s criminal justice system and is generated as an individual is booked into jail. This ID follow an individual throughout their entire criminal “career”. We exclude screenings where individuals were released to places or facilities that precludes the “opportunity” to encounter a new arrest in the community in the state of Utah.↩

3. A unique screening is defined as a unique screening instance at a particular date. This implies that one individual can have multiple screenings that occurred at separate dates, which has the potential to put upward pressure on the average risk level. The matching algorithm identified the closest arrest that occurred after an individual was released from jail, if the actual offense date occurred after the screening/jail release date.↩

4. A smaller share of new arrests had an unknown or missing arrest severity.↩

5. The LSI-R:SV is considered a gender neutral instrument.↩

6. While our second outcome measure is arguably of greater public safety concern, our first outcome measure is important from both a public safety and budgetary concern as it relates to the cycle of re-incarceration.↩

7. The recommended cut-off points for each risk classification are: low risk: 0-2 points, moderate risk: 3-5 points, and high risk: 6 or more points (scale: 0-8).↩