Law and Human Behavior

:1

Assessment and Management of Risk for Intimate Partner Violence by Police Officers Using the Spousal Assault Risk Assessment Guide

Authors

  • Henrik Belfrage
    • Department of Health Sciences, Section for CriminologyMid Sweden University
    • Forensic Psychiatric Centre
  • Susanne Strand
    • Department of Health Sciences, Section for CriminologyMid Sweden University
    • Forensic Psychiatric Centre
  • Jennifer E. Storey
    • Department of PsychologySimon Fraser University
  • Andrea L. Gibas
    • Department of PsychologySimon Fraser University
  • P. Randall Kropp
    • Department of PsychologySimon Fraser University
    • British Columbia Forensic Psychiatric Services Commission
    • Department of PsychologySimon Fraser University
    • Faculty of PsychologyUniversity of Bergen
Original Article

DOI: 10.1007/s10979-011-9278-0

Cite this article as:
Belfrage, H., Strand, S., Storey, J.E. et al. Law Hum Behav (2011). doi:10.1007/s10979-011-9278-0

Abstract

Intimate partner violence (IPV) is a crime that is present in all countries, seriously impacts victims, and demands a great deal of time and resources from the criminal justice system. The current study examined the use of the Spousal Assault Risk Assessment Guide, 2nd ed. (SARA; Kropp, Hart, Webster, & Eaves, 1995), a structured professional judgment risk assessment and management tool for IPV, by police officers in Sweden over a follow-up of 18 months. SARA risk assessments had significant predictive validity with respect to risk management recommendations made by police, as well as with recidivism as indexed by subsequent contacts with police. Risk management mediated the association between risk assessment and recidivism: High levels of intervention were associated with decreased recidivism in high risk cases, but with increased recidivism in low risk cases. The findings support the potential utility of police-based risk assessment and management of IPV, and in particular the belief that appropriately structured risk assessment and management decisions can prevent violence.

Keywords

Intimate partner violenceRisk assessmentRisk managementSARAPolice

Intimate partner violence (IPV), particularly the abuse of women perpetrated by men, is a societal problem of pandemic proportions. A recent World Health Organization study (Garcia-Moreno, Jansen, Ellsberg, Heise, & Watts, 2006) examined prevalence studies of intimate partner violence conducted in 10 countries, with a total sample of approximately 24,000 women. The results of this study showed widespread physical or sexual abuse by intimate partners; the lifetime prevalence of various forms of IPV ranged between 15 and 71%. A review of population-based studies in 35 countries suggested a lifetime prevalence of physical IPV between 10 and 52% (Heise & Garcia-Moreno, 2002). Recent research has estimated the lifetime prevalence of physical IPV to be between 25 and 34% in the United States (Tjaden & Thoennes, 2000; Thompson et al., 2006) and 25–35% in Canada (Clark & DuMont, 2003). Recidivism is also common and estimated to be between 40 and 80% (e.g., Shepard, 1992). In addition to being prevalent, IPV has acute and chronic consequences including impaired social functioning, psychological trauma, physical injury, and death (e.g., Bonomi et al., 2006; Campbell, 2002; Campbell, Glass, Sharps, Laughon, & Bloom, 2007; Kernic, Wolf, & Holt, 2000; Plichta, 2004; Tolman & Rosen, 2001).

The frontline management of IPV most often falls to the police. The police response to IPV has shifted its focus from reactive, responding to IPV after it has occurred, to proactive, attempting to prevent IPV (Belfrage, 2008; Belfrage & Strand, 2008; Hilton, Harris, & Rice, 2010; Kropp, 2004, 2008). Police now are responsible for assessing and managing the risks posed by IPV perpetrators, the vulnerabilities of IPV victims, and liaison and coordination with community services.

Increasingly, police are using specific risk assessment and management tools to assist them in their efforts to prevent IPV. For example, the Ontario Domestic Assault Risk Assessment Guide, or ODARA, is a 13-item actuarial risk assessment instrument constructed by comparing groups of recidivistic and non-recidivistic adult males who had been convicted of physical assault against female intimate partners with whom they cohabited (Hilton, Harris, Rice, Lang, Cormier, & Lines, 2004; see also Hilton et al., 2010). The ODARA has been piloted in several sites in Ontario, Canada. Another example is the Brief Spousal Assault Form for the Evaluation of Risk, or B-SAFER (Kropp, Hart, & Belfrage, 2005, 2010). The B-SAFER is a set of structured professional judgment (SPJ) guidelines designed for use by police and other criminal justice professionals that comprises 10 perpetrator risk factors and 5 victim vulnerability factors. The B-SAFER is used by police agencies in several Canadian provinces, and also has been translated for use by police agencies in several other countries, including the Czech Republic, Italy, Norway, Portugal, Slovakia, and Sweden (e.g., Kropp, Hart, & Belfrage, 2004, 2008; see also Baldry & Winkel, 2007). Other risk assessment instruments used by various police departments include the Spousal Assault Risk Assessment Guide, or SARA (Kropp, Hart, Webster, & Eaves, 1994, 1995, 1999), from which the B-SAFER was derived; the Danger Assessment or DA (Campbell, 1995); and a number of agency-specific tools.

Despite their widespread use, there have been very few published studies evaluating the utility of IPV risk assessment instruments in law enforcement contexts. In one study, Belfrage and Strand (2008) examined a series of 698 IPV cases from Sweden in which police had used the B-SAFER. They found that decisions regarding IPV risk level were strongly associated with the number of perpetrator risk factors present in a given case. Although victim vulnerability factors were rated as present less often, they were also strongly associated with decisions regarding violence risk level. In another study, Trujillo and Ross (2008) examined a sample of 501 IPV risk assessments completed by police officers in Victoria, Australia, using an agency-specific tool, the Family Violence Risk Assessment and Management Report. The results indicated that decisions about level of risk and intervention were influenced primarily by a small number of risk factors, including escalation in the reported incidents and the level of fear expressed by the victim. We were not able to locate any published studies that examined the predictive validity of IPV risk assessments conducted by police with respect to recidivism.

In the present study, we evaluated the utility of the SARA as used by police in Sweden using a true prospective design. Our primary interest was the extent to which police officers’ ratings of risk for IPV made using the SARA were associated with their recommendations regarding intervention and with subsequent IPV recidivism. Based on the principles of the Risk Needs Responsivity or RNR model of correctional intervention (Andrews & Bonta, 2006; Andrews, Bonta, & Wormith, 2006) and the prevention model underlying the SPJ approach to violence risk assessment (Douglas & Kropp, 2002; Hart, 1998, 2001), we made the following predictions:
  1. 1.

    Risk assessment will predict risk management. Specifically, police officers’ ratings of risk made using the SARA will be positively associated with the level of intervention they recommend.

     
  2. 2.

    Risk assessment will predict recidivism. Specifically, police officers’ ratings of risk made using the SARA will be positively associated with subsequent IPV.

     
  3. 3.

    Risk management will mediate the association between risk assessment and recidivism. Specifically, the association between police officers’ ratings of risk made using the SARA and subsequent IPV will depend, at least in part, on the level of intervention recommended, such that a high level of intervention will be most effective in reducing recidivism in high risk cases and least effective (or even counter-productive) in low risk cases.

     
Our predictions are unusual in the sense that research to date has tended to ignore the potential impact of risk management on recidivism, whereas we hypothesize explicitly that risk assessment guides risk management, which in turn influences recidivism. The difference between these two causal models is illustrated in Fig. 1. Panel (a) of Fig. 1 ignores risk management. It posits an unmediated association between risk assessment, X, and recidivism, Y. According to the diagram, whatever causal factors are identified by risk assessment have a direct influence on recidivism, Path c. Risk management does not need to be included as a separate factor in this diagram, as its influence on recidivism is incorporated in Path c. In contrast, panel (b) includes risk management, M, as a separate causal factor, one that mediates the association between risk assessment and recidivism. Risk management is considered a mediator rather than a moderator in panel (b) because it is plausible that, as indicated by Path a, risk management occurs after and is determined in part by risk assessment; and also, as indicated by Path b, recidivism occurs after and is determined in part by risk management. The extent to which the association between risk assessment and recidivism is mediated by risk management can be evaluated statistically by comparing Path c in panel (a) to Path c′ in panel (b). For further discussion of conceptual and statistical issues related to mediation models, see Baron and Kenny (1986), MacKinnon and Dwyer (1993), MacKinnon, Fairchild, and Fritz (2007), and Preacher and Hayes (2004).
https://static-content.springer.com/image/art%3A10.1007%2Fs10979-011-9278-0/MediaObjects/10979_2011_9278_Fig1_HTML.gif
Fig. 1

Illustration of (a) unmediated and (b) mediated models of the association between risk assessment and recidivism

Method

Overview

The research was conducted in three counties in Sweden: Kalmar, population about 35,000; Växjö, population about 80,000; and Blekinge, population about 150,000. Sworn officers of the Swedish national police in these three counties were trained in the use of the SARA by HB or PRK. Police then used the SARA in all IPV cases to assess risk and develop case management plans. After attending IPV calls and completing an investigation, police officers completed the SARA and recommended intervention strategies. The officers then reviewed this information with a supervisor prior to filing it in records.

For research purposes, we retrieved official records and completed an audit of all spousal assault cases in the three counties responded to by police over a period of 18 months, starting in 2000. We coded the following information from police records for each case: the SARA risk assessments and management plans filed after the initial investigation, whether there were any further complaints of IPV made to police involving the same perpetrator during the 18 months following the initial response by police, and new SARA risk assessments and management plans filed as a result of any further complaints of IPV.

Note that this is a true prospective design. Following each contact with police, risk assessments were conducted prior to recommendations regarding risk management, which in turn were made prior (and therefore blind) to actual recidivism. Thus, there is no way that risk assessment and management decisions could have been contaminated by knowledge of outcome. Of course, risk assessment and management decisions made at the second or subsequent contact with police were not made blind to the past, but were still made prior and therefore blind to future recidivism.

The research was done in cooperation with and received ethical approval from the Swedish national police.

Cases

From the entire sample of cases responded to by police during the study period, we excluded a small number in which the accused—that is, the alleged or suspected perpetrator—was a dependent minor (younger than age 18 and still living at home) or female. This resulted in a final sample of 429 cases, all of which involved male-to-female IPV, defined as any actual, attempted or threatened physical or sexual violence against a past or current intimate partner. For the sake of brevity, we will refer to the men who were investigated for complaints related to IPV as perpetrators and the women who were their current or former intimate partners as victims.

At the time of the first police response, the mean age of the perpetrators was 39 years, with a range of 17–92 years. Most perpetrators were born in Sweden and of Swedish heritage; 27% were born abroad or had parents who were born abroad. The age and cultural background of victims was virtually identical. In terms of relationship status, about half of the perpetrators (51%) had relationships with their victims that were intact, and the other half (49%) had relationships that had dissolved. In most cases (57%), the relationship between perpetrators and victims had resulted in one or more children.

As a result of the first police contacts during the study period, all of the perpetrators were charged with criminal offenses. Most (66%) were charged with a single offense and the remainder with multiple offenses. The most common offense type was assault (at least one charge in 66% of cases), followed by unlawful threat (at least one charge in 21% of cases); the remaining charges were for a wide range of offenses, including harassment and breach of peace.

Procedure

Risk Assessment

As noted previously, the SARA is a set of SPJ guidelines for assessing risk for IPV (Kropp et al., 1994, 1995, 1999). The guidelines were based on a systematic review of the scientific literature as well as consideration of existing standards of practice, ethical codes, and legal principles. The SARA comprises 20 standard risk factors in four domains: General Criminality, Psychosocial Adjustment, Spousal Assault History (i.e., past incidents of IPV), and Index Offense (i.e., the current or most recent incident of IPV). These SARA risk factors are presented in Table 1. The presence of each risk factor is coded on a 3-point scale, No/Absent, Possibly/Partially present, and Yes/Present. Presence ratings may be omitted if no valid information is available. The relevance of each item with respect to risk and risk management is coded on a 2-point scale, No versus Yes. For research purposes, it is common to translate the presence ratings for individual risk factors into numerical total scores (Omit or No/Absent = 0, Possibly/Partially present = 1, and Yes/Present = 2) and then sum the scores to create an index reflecting overall risk. Based on the presence and relevance of risk factors, evaluators make summary risk ratings (i.e., an overall judgment of risk level) on a 3-point scale, Low risk, Moderate risk, or High risk. According to a recent review by Kropp and Gibas (2010), SARA numerical total scores and summary risk ratings have good to excellent interrater reliability, good convergent and discriminant validity with respect to measures of risk for general violence and IPV, and good predictive validity with respect to future IPV.
Table 1

Risk factors in the SARA and frequency of presence ratings at first contact with police

SARA risk factor

Frequency (%)

Omit

N

P

Y

Criminal history factors

1. Past assault of family members

15

54

7

25

2. Past assault of strangers or acquaintances

21

56

5

19

3. Past violation of conditional release or community supervision

17

80

1

3

Psychosocial adjustment factors

4. Recent relationship problems

5

14

16

65

5. Recent employment problems

15

58

7

21

6. Victim of and/or witness to family violence as a child or adolescent

36

60

1

3

7. Recent substance abuse/dependence

13

42

10

35

8. Recent suicidal or homicidal ideation/intent

25

63

3

10

9. Recent psychotic and/or manic symptoms

26

58

3

13

10. Personality disorder with anger, impulsivity, or behavioral instability

22

39

12

26

Spousal assault history factors

11. Past physical assault

10

35

8

47

12. Past sexual assault/sexual jealousy

22

68

4

7

13. Past use of weapons or credible threats of death

18

59

15

8

14. Recent escalation in frequency or severity of assault

14

48

17

21

15. Past violation of “no contact” orders

6

90

1

2

16. Extreme minimization or denial of spousal assault

20

53

5

22

17. Attitudes that support or condone spousal assault

25

62

4

10

Index offense factors

18. Index offense: Severe and/or sexual assault

1

39

41

19

19. Index offense: Use of weapons or credible threats of death

3

53

22

22

20. Index offense: Violation of “no contact” order

2

94

2

2

Note: N = 429. N = No/Absent; P = Partially/Possibly present; Y = Yes/Present

In the present study, police officers made presence ratings for risk factors and summary risk ratings using the SARA based on all information collected in the course of their investigations. The SARA ratings were reviewed with a supervisor before being filed. As it was not possible to have a second independent police investigation at or close to the same time as the original investigation, we did not evaluate interrater or test–retest reliability. We did, however, examine the association between SARA ratings made after the first and second contact with police in the subsample of 93 recidivistic perpetrators. We expected some change in risk assessments. Specifically, we anticipated that SARA numerical total scores and summary risk ratings made after the second contact with police should be similar to but slightly higher than those made after the first contact. We indexed stability of SARA risk assessments—here, a combination of interrater and test–retest reliability—using ICC1 for single ratings (mixed effects model, absolute agreement method).

Risk Management

The police officers were required to document their recommendations regarding risk management following the completion of each SARA risk assessment. Most of these recommendations involved protective actions or safety planning for victims. Like their risk assessments, police officers’ risk management recommendations were reviewed by supervisors. To assist the police officers, a menu of risk management strategies available to the Swedish police was attached to each SARA coding form. Although we were not able to determine whether, when, and how recommended strategies were implemented, feedback from police indicated that most recommendations were implemented in some manner.

Table 2 presents the 14 management strategies examined in this study, as well as the frequency with which they were recommended. The most commonly recommended strategies were to have contact with the prosecutor (84%), search the crime registry for details regarding past offenses (69%), discuss security or safety plans with the victim (40%), and initiate a no contact order (30%). In 12 percent of the cases, no specific management strategies were recommended. The mean number of recommended management strategies in each case was 3.11 (SD = 1.80, range = 0–10, Mdn = 3).
Table 2

Management strategies: frequency of recommendation by police officers and correlations (r) with SARA numerical total scores and summary risk ratings

Management strategy

Frequency of recommendation by police officers (%)

Correlation (r) with SARA

Numerical total score

Summary risk rating

1. Contact a prosecutor

84

.13

.09

2. Search the crime registry

69

.02

.04

3. Discuss security (i.e., safety plan) with victim

40

.20

.26

4. Initiate a no contact order with victim

30

.39

.35

5. Contact a victim support agency

20

.15

.04

6. Establish a victim support person

19

.37

.39

7. Initiate other protective actions

19

.02

.05

8. Contact a victim shelter (safe house)

11

.17

.16

9. Install a home alarm system

7

.19

.23

10. Improve security at victim’s home

4

.10

.15

11. Conduct further risk assessment

4

.20

.20

12. Contact police supervisor on duty

2

.07

.07

13. Establish a police contact person for victim

1

.10

.11

14. Protect identity of victim

1

.00

.03

Note: N = 429

Recidivism

Subsequent to the initial response by police, 93 of the perpetrators (21%) had further contacts with police for investigation of incidents related to IPV (as defined previously) during the 18-month follow-up. Among these recidivists, the number of new contacts with police ranged from 1 to 20, although only a minority of men (32 of 93, or 34%) had multiple further contacts with police.

Results

Distribution of SARA Ratings

Table 1 presents data concerning the frequency with which individual SARA risk factors were rated present. Two general patterns are apparent in the table. First, the frequency with which ratings for individual risk factors were omitted was quite high, Mdn = 16%. Feedback from police officers suggested they were reluctant to rate risk factors No/Absent even when available information indicated risk factors were not present. Second, the frequency with which risk factors were rated present varied considerably, from a low of 2 or 3% for risk factors related to violations of “no contact” orders and conditional release to a high of 65% for the risk factor reflecting recent problems in the intimate relationship between perpetrator and victim, Mdn = 19%. After converting presence ratings, the numerical total scores ranged from 0 to 32 out of a possible 40 with a mean of 9.41 (SD = 5.99) and a median of 8.

Despite the relatively frequent omission of presence ratings for individual risk factors, police officers made summary risk ratings in every case. Of the 429 perpetrators, 201 (47%) were rated Low risk, 169 (39%) were rated Moderate risk, and 59 (14%) were rated High risk. The correlation between SARA numerical total scores and summary risk ratings was r = .64, p < .001.

As might be expected, perpetrators in the current sample were relatively low risk compared to samples of men already serving community or institutional sentences for IPV-related offenses. For example, Kropp and Hart (2000) summarized SARA ratings for 1,671 adult male probationers; they found that 22% of probationers were rated Low risk, 49% were rated Moderate risk, and 28% were rated High risk, and the mean numerical total score was 12.98 (SD = 6.46.)

Stability of SARA Ratings

Among the 93 perpetrators who had multiple contacts with police, the mean numerical total score after the first contact was 11.48 (SD = 6.08). After the second contact, it was 13.04 (SD = 6.28). This increase in risk was statistically significant, t(92) = 3.69, p < .001. The stability between the first and second contacts was high, ICC1 = .76, and similar to the interrater reliability of numerical total scores reported by Kropp and Hart (2000). As police officers typically do not have extensive training or experience in assessment of mental disorder, we also evaluated the reliability of the three individual risk factors that reflect symptoms of mental disorder. The reliability of ratings across the first and second contacts was moderate, ICC1 = .56, .68, and .59 for Risk Factors 8, 9, and 10, respectively.

Looking at summary risk ratings, the number of perpetrators rated Low risk, Moderate risk, and High risk at the first contact was 33, 46, and 14, respectively; and at the second contact it was 19, 55, and 19, respectively. This increase in risk was statistically significant, McNemar’s χ2 (1, N = 93) = 9.77, p = .021. Despite the increase in risk, the stability of summary risk ratings between the first and second contacts was fair, ICC1 = .45, and similar to though slightly lower than the interrater reliability of summary risk ratings reported by Kropp and Hart (2000). (Note that for ordered categorical ratings, ICC1 is equivalent to weighted κ, or κW.)

Prediction #1: Risk Assessment Will Predict Risk Management

As predicted, SARA numerical total scores and summary risk ratings were significantly correlated with the number of management strategies recommended in each case, both r = .40, p < .001. Thus, police officers tended to recommend more management strategies in higher risk cases than in lower risk cases.

To examine this association in a different way, we calculated the number of risk management strategies recommended as a function of SARA summary risk ratings. Among the 201 perpetrators rated Low risk, the number of management strategies recommended ranged from 0 to 8, with Mdn = 2 and M = 2.41 (SD = 1.46). Among the 169 perpetrators rated Moderate risk, the number of management strategies recommended ranged from 0 to 8, with Mdn = 3 and M = 3.48 (SD = 1.68). Finally, among the 59 perpetrator rated High risk, the number of management strategies recommended ranged from 1 to 10, with Mdn = 4 and M = 4.44 (SD = 1.10). ANOVA indicated that the difference between risk groups was statistically significant, F(2, 429) = 45.51, p < .001. On average, every 1-step increase in risk was associated with an extra recommendation for risk management.

Table 2 presents the correlations between the recommendations for individual management strategies and SARA ratings (numerical total scores and summary risk ratings). Most strategies were positively and significantly correlated with ratings of risk. The two recommended management strategies most strongly associated with ratings of risk were Initiate a no contact order with victim and Establish a victim support person.

Prediction 2: Risk Assessment Will Predict Recidivism

Recidivism was associated with higher numerical scores on the SARA. The association can be indexed in several ways. First, the point-biserial correlation between recidivism (coded 0 = no, 1 = yes) and numerical total scores was rpb = .178, p < .001. Second, the mean numerical total score of recidivists was 11.45 (SD = 6.10), compared to 8.86 (SD = 5.84) for non-recidivists, t(92) = 3.79, p < .001, Cohen’s d = .43. Third, ROC analyses yielded AUC = .63 (SE = .032). Overall, these analyses indicated that the unmediated association between recidivism and numerical total scores (i.e., the total effect, or Path c in Fig. 1) was small-to-moderate in magnitude.

Recidivism was less consistently associated with SARA summary risk ratings. First, the point-biserial correlation recidivism and summary risk ratings was rpb = .092, p = .056. Second, the number (percentage) of recidivists among perpetrators rated Low risk, Moderate risk, and High risk was 14 of 59 (24%), 45 of 169 (27%), and 33 of 201 (16%), respectively, χ2 (2, N = 429) = 5.89, p = .053. Third, ROC analyses yielded AUC = .57 (SE = .033). Overall, these analyses indicated that the unmediated association between recidivism and summary risk ratings (i.e., the total effect, or Path c in Fig. 1) was small in magnitude. Due to their weak association with recidivism, we did not analyze summary risk ratings further.

Prediction #3: Risk Management Will Mediate the Association Between Risk Assessment and Recidivism

To explore possible mediation, we conducted logistic regression analysis using SARA numerical total scores, number of recommended management strategies, and their interaction to predict recidivism. The results are summarized in Table 3. The overall model was statistically significant, the main effects for numerical total scores and number of management strategies recommended were significant, and there was a significant interaction between numerical total scores and number of management strategies recommended. Next, we constructed a number of contingency tables dichotomizing numerical total scores and the number of management strategies recommended, and the pattern was consistent: As predicted, risk management recommendations were associated with decreased recidivism in high risk perpetrators but with increased recidivism in low risk perpetrators. To illustrate, Table 4 presents the results when both variables were dichotomized comparing the top quartile (Q1) to the other quartiles (Q2–Q4).
Table 3

Logistic regression: predicting recidivism with SARA numerical total scores and number of management strategies recommended

Predictors

B

SE

p

eB

SARA numerical total score

.15

.05

.001

1.17

Number of management strategies recommended

.32

.15

.032

1.38

Interaction

−.02

.01

.030

.98

Note: χ2 (3, N = 429) = 18.28, p < .001, Nagelkerke R2 = .07

Table 4

Recidivism as a SARA numerical total scores and number of management strategies recommended (Q1 versus Q2–Q4)

Numerical total scores

Number of management strategies recommended

Recidivism (%)

Low risk (≤10)

Low (≤2)

15/131 (12%)

High (≥3)

29/139 (21%)

High risk (≥11)

Low (≤2)

14/38 (37%)

High (≥3)

34/121 (28%)

Finally, we directly evaluated the mediation model using the Sobel test (Sobel, 1982) according to the procedures for logistic regression outlined by MacKinnon and Dwyer (1993) and Preacher and Hayes (2004). Referring to Fig. 1, this involves quantifying the coefficient of the indirect effect of the mediator in the mediated model—that is, the product of coefficients for Paths a and b in Panel (b)—and its standard error. The coefficient of the indirect effect divided by its standard error is distributed as z. The observed value of the Sobel test was 2.99, SE = 0.009, p = .003, indicating that the mediation effect was significantly greater than 0. The results of Aroian (1944/1947) and Goodman (1960) tests, popular alternatives to the Sobel test, yielded findings that were identical, within rounding error.

Discussion

Police-Based Assessment of Violence Risk

As first responders to reports of IPV, police officers face the difficult task of managing both perpetrators and victims. Their duty to protect victims and the public requires that they conduct risk assessments in either a formal or informal manner (Kropp, 2008). But there has been little research on the implementation and effectiveness of police-based risk assessment. Consistent with some previous research in Sweden (Belfrage & Strand, 2008), Australia (Trujillo & Ross, 2008), and Canada (Hilton et al., 2010), the present study yielded several findings consistent with the view that police officers can conduct meaningful IPV risk assessments.

First, after relatively brief training in the use of the SARA, the police officers participating in this study had little difficulty understanding and applying the SPJ risk assessment approach. They were able to code most of the SARA risk factors in most cases from available information, and to make meaningful summary risk ratings in all cases. The frequency with which presence ratings for individual risk factors was relatively high, however, and suggests that there may be some benefit in simplifying the definition of risk factors and clarifying the administration procedures. (For further discussion, see Belfrage, 2008.)

Second, the SARA ratings made by police officers were consistent with the nature of the sample. Specifically, the ratings suggested the current sample comprised perpetrators who posed a moderate or low-to-moderate risk for IPV, at least compared to previous studies of convicted IPV perpetrators (Kropp & Hart, 2000; see also Williams & Houghton, 2004).

Third, the stability of SARA risk ratings across contacts with police was moderate to good. This was true even for risk factors related to mental disorder. Also, SARA ratings were change-sensitive, increasing as a function of repeated contact with police.

Fourth, the association between SARA ratings and number of management strategies recommended suggests that police officers’ risk assessments influenced their decisions regarding risk management. This is consistent with previously published research on police decision making in IPV cases (Belfrage & Strand, 2008; Trujillo & Ross, 2008).

Finally, SARA ratings were associated with recidivism. Although the magnitude of the association was somewhat lower than that found in previous studies of the SARA (see Kropp & Gibas, 2010), this was due in part to the mediating effect of risk management. This is a critical point, one that we discuss further below.

Risk Assessment, Risk Management, and Recidivism

Despite frequent discussion of the need for research to expand its focus from risk assessment to include risk management and prevention (Andrews & Dowden, 2006; Douglas & Kropp, 2002; Hart, 1998, 2001; Heilbrun, 1997), few studies have examined the complex associations among risk assessment, risk management, and recidivism. The predictive validity of risk assessment procedures cannot be judged solely on the basis of their simple bivariate associations with recidivism; risk management must be taken into account. Analysis of simple bivariate associations assumes that risk management either was not implemented at all or was universally (in-) effective. Both assumptions are untenable. If risk management strategies are indeed implemented and differentially effective, then simple bivariate associations are likely to substantially underestimate the validity of risk assessment—particularly when risk management is based on risk assessment (Douglas & Kropp, 2002; Hart, 1998, 2001).

In this study, we found that risk assessment predicted risk management: higher risk ratings were associated with more management recommendations. Risk assessments also predicted recidivism: higher risk ratings were associated with more contacts with police during the follow-up. Taken together, these findings are consistent with the RNR model of correctional intervention (e.g., Andrews & Bonta, 2006; Andrews et al., 2006) and the violence prevention model underlying the SPJ approach (e.g., Douglas & Kropp, 2002; Hart, 1998, 2001). Furthermore, they can be taken as support for the validity of the SARA as a risk assessment procedure.

But more important was the finding that risk management mediated the association between risk assessment and recidivism. Specifically, high risk perpetrators with many management recommendations were less likely to have subsequent contacts with police than were high risk perpetrators with few management recommendations, whereas low risk perpetrators with many management strategies were more likely to have subsequent contacts with police than were low risk perpetrators with few management strategies. This finding is also consistent with the RNR and violence prevention models.

Some readers may find it curious that our findings indicated that high levels of intervention with low risk perpetrators may actually have increased recidivism. But similar findings have been observed many times before in the correctional intervention literature. Although the precise causal mechanism is unclear, it appears that too much intervention may interfere with the natural—and often effective—coping strategies of low risk offenders. Thus, low risk offenders may require relatively little formal case management. As Andrews and Dowden (2006) put it, “If it ain’t broke, don’t fix it” (p. 88).

Study Design Issues

A major strength of this study was that it that it took place in the field. The risk assessment and management decisions were made by police officers in situ, allowing us to evaluate their real-world effectiveness, as opposed to their efficacy in controlled settings. Another strength was the true prospective design, which helped to clarify the interpretation of findings. But the study also had limitations that need to be addressed in subsequent research. First, our reliance on subsequent police contacts as a measure of recidivism meant that we only recorded recidivism in cases where police were notified of an IPV incident. The notifications come from primarily from victims themselves, as well as from witnesses. On one hand, police contacts exclude cases where IPV occurred but was not reported to police (i.e., false negative errors). On the other hand, it may include cases in which IPV actually did not occur (i.e., false positive errors). Although we think that police contact is a more sensitive index of actual IPV recidivism than is arrest, charge, or conviction, police contact may also be less sensitive than victim self-report. Second, we did not monitor cases directly and so were able to document recommended strategies, but not which strategies were actually implemented and, if so, at what time and in what order. Although feedback from the patrol officers and their supervisors involved in this study suggested that their management recommendations were often followed, a more detailed analysis of case management may have provided a more clear picture regarding the effect of intervention on violence prevention (for a discussion of this complex issue with respect to stalking, see (Storey & Hart, in press). Third, we were not able to gather detailed information about other events or occurrences in the lives of perpetrators, including arrests, charges, and convictions for non-IPV offenses. If some perpetrators had police contact or even spent time in custody related to other (alleged) offenses, the absence of IPV recidivism in those cases may have resulted in an under-estimate of the predictive power of risk assessment and management decisions.

Conclusion

The current findings support the use of the SARA by police officers for assessing risk for IPV and also for making decisions about risk management (e.g., prioritizing cases, allocating resources). Importantly, this is the first study to illustrate that the SARA is an effective tool not only for risk assessment and risk management, but also for violence prevention. Given the scope and severity of IPV, further research on police-based assessment and management of risk should be a priority.

Acknowledgments

The authors thank the Swedish national police, and in particular the Rikspolisstyrelsen and Kenneth Krantz, and support and assistance. We also acknowledge the contributions of Victoria Jeffries and Kim Reeves to previous analyses of the data.

Copyright information

© American Psychology-Law Society/Division 41 of the American Psychological Association 2011