Name of Tool Level of Service Inventory Revised (LSI-R)

Category General Risk Assessment (Validated)

Author / Publisher Andrews and Bonta

Year (1995)

Description

•LSI-R is a 54-item actuarial tool of the individual’s attributes and their circumstances. It is
designed to assess criminogenic risk and identify the needs of those who have offended (Watkins,
2011).
•Information is collected via a semi-structured interview, a review of case records and collateral
verification (Wilson et al., 2016).
•The tool centres on the principles of risk, need and responsivity, maintaining that those who are
at high risk of reoffending should receive higher intensity interventions, supervision and monitoring
(Watkins, 2011).
•Thirty-four items are subdivided across ten subsections. The total score is used to calculate
recidivism risk, categorised as either ‘minimum,’ ‘medium’ or ‘maximum.’ Subscale scores are used
to identify criminogenic needs (Watkins, 2011).
•In addition to recidivism, composite scores help to predict parole outcomes and the presence or
risk of institutional misconduct (Wilson et al., 2016).
•Normed on North American prison, parole and probation populations.

Age Appropriateness

16+

Assessor Qualifications

Assessors must possess advanced training, certification and experience in psychological

assessment or a related discipline, or satisfactorily complete a training course certified by the
publishers. Can be used by a large range of professionals including social work and probation
services.

Strengths

•Ability to discriminate risk across various outcome measures such as spousal abuse recidivism
(Hendricks et al., 2006).
•Provides structured professional decision-making in a way that is comprehensive and consistent
regardless of the case presented (Campbell et al., 2009).
•Both criminal history and the needs are captured with the tool. There is also an override feature
to allow for the exercising of professional judgment to be exercised (Wilson et al., 2016).

RATED page updated: July 2019

© Risk Management Authority 2019
 •Wilson and Stevenson (2017) claimed that the semi-structured interview component of the
instrument is a helpful framework for treatment and supervision, since it addresses learning,
behavioural and developmental issues.
•The LS instruments are based on ‘General Personality and Cognitive Social Learning theory,’ which
is a general theory of criminal conduct entrenched in social learning perspectives (Wormith and
Bonta, 2018).

Empirical Grounding

•The LSI-R is supported by and reflective of three primary sources of information: (1) prior literature
on recidivism, (2) professional opinions of probation officers and (3) social learning theory of
criminal behaviours (Andrews and Bonta, 1995: 1).
•The subscales reflect the main risk factors identified in the research literature (Andrews and
Bonta, 2010).
•Subject to a number of meta-analyses (Olver et al., 2014)

Inter-Rater Reliability

a) UK Research •Palmer and Hollin (2007) - inter-rater agreement levels

of 95% for females.

•Hollin and colleagues (2003) found a 90% agreement

rate in a sample of males.

b) International Research •Dahle (2006) found excellent inter-rater reliability

generating an ICC value of .93 in a sample of German
individuals who had offended.

•Lowenkamp et al. (2004) - moderate to high levels of

agreement observed across all ten subsections ranging
from 61.5% to 97.7%.

•Andrews (1982) - excellent inter-rater reliability

coefficients ranging between .80 to .99.

•Persson et al. (2017) found that the inter-rater reliability

for the LSI-R was excellent (ICC=.92).

Validation History

General Predictive Accuracy

a) UK Research •Raynor and Miles (2007) - predictive accuracy ranging

from 65.4% to 71.6%.

RATED page updated: July 2019

© Risk Management Authority 2019
 •Raynor (2007) - LSI-R presented ability to discriminate
between reconvicted individuals who received a fine and
those serving community/probationary sentences.

•Hollin and Palmer (2006) found a moderate correlation

between the LSI-R composite score and reconviction
status.

b) International Research •Duwe and Rocque (2016) administered the LSI-R to 26,
000 prisoners in Minnesota for the time period of 2003 to
2011. The results gave an AUC of 0.628, providing
moderate support for the LSI-R’s ability to assess need.

•In a study of 828 prisoners in Midwest of the United

States, the LSI-R was able to predict recidivism (Smith et
al., 2014).

•A study in Australia found that the LSI-R yielded an

acceptable level of reliability, with internal consistency
estimates in the range of 0.59 to 0.784 (Watkins, 2011).

•Campbell French and Gendreau (2009) - the LSI-R

displayed one of the largest mean effect sizes in
predicting violent recidivism (Z+ =.28).

•A study by Lowenkamp et al. (2009) found moderate

correlations between both re-arrest (r = .36) and re-
incarceration rates (r = .33) and the LSI-R composite
score.

•Manchak et al. (2008) - the LSI-R yielded an AUC value

of .73 for both general and violent recidivism.

•Dahle (2006) - the LSI-R achieved moderate accuracy in

violence prediction over a 10-year period (AUC =.65) in a
sample of Germans.

Validation History

Applicability: Females

a) UK Research •Raynor and Miles (2007) - for females in England and

Wales (n = 163) the LSI-R mean score = 21.2, % correctly
predicted = 65%.

•Palmer and Hollin (2007) found that for female

prisoners in England and Wales (n = 150) the LSI-R mean
score = 23.0. There were significant differences between
male and female scores on seven subscales, but not in

RATED page updated: July 2019

© Risk Management Authority 2019
 the overall score. Scores significantly predicted
reconviction and time to reconviction. The composite
score correctly classified 74%, with 79.7% correct
classification for those not reconvicted and 64.9% for
those who were convicted.

b) International Research •An Australian study found that the correlations between
criminal history items and recidivism rates decreased in
magnitude and significance when the LSI-R was applied
to females. The author posited that the LSI-R subscales
may not be suitable for fully assessing the criminogenic
needs of females who offend (Watkins, 2011).

•Hogg (2011) found the LSI to be gender neutral.

•Manchak et al. (2008) - the LSI-R attained excellent

predictive accuracy in relation to recidivism in a sample of
female who offended (AUC = .77).

•In a meta-analysis by Smith and colleagues (2009), it

was found that the LSI-R demonstrated a correlation of
r=.35 for recidivism in females.

•Vose et al. (2008) - the LSI-R was found to be a valid

predictor of recidivism in females, achieving a composite
score of 71.4% accuracy.

Validation History

Applicability: Ethnic Minorities

a) UK Research None at present.

b) International Research •Hsu, Caputi and Byrne (2010) - the LSI-R demonstrated
small correlations with recidivism in a sample of male and
female Australian Indigenous individuals (rs = .12 and .16
respectively). Indigenous individuals were found to score
consistently higher on every item of the LSI-R.

•Fass et al. (2008) - inconsistent validity with ethnic

minority groups. LSI-R had better predictive accuracy with
Caucasians (80.4%) and Hispanics (82.4%) than African
Americans (43.4%).

•Schlager and Simourd (2007) - few statistically

significant correlations between LSI-R composite scores
and recidivism amongst ethnic minority groups.

RATED page updated: July 2019

© Risk Management Authority 2019
 •Ostermann and Salerno (2016) applied the LSI-R to
9454 individuals in New Jersey to gauge its validity in
predicting recidivism within a year of their release from
prison. It was found that the LSI-R displayed low capacity
for distinguishing between recidivists and non-recidivists
when applied to Black males.

•A study by Watkins (2011) found that the discriminatory

power on the LSI-R were very low for those with
Aboriginal/Torres Strait Islander status in a sample of
Australian individuals

•Chenane et al. (2015) examined the predictive validity

of the LSI-R in 2778 male prisoners in the Midwest of the
United States across White, Black and Hispanic ethnic
groups. Results indicated that the LSI-R was better-suited
to predicting institutional misconduct for White prisoners
than the other two groups. It was suggested by the
authors that the tool is modified to adhere to the risks and
needs of Black and Hispanic prisoners.

•Applying the LSI-R to 95 clients within a mental health

jail diversion program, Lowder et al. (2017) determined
that the LSI-R showed weak predictive validity for African
Americans than Caucasian clients. Moreover, the risk
estimate was found to under-classify African Americans
for the moderate risk category; whilst over-classifying
them for high risk.

•Research by Lowder and colleagues (2019) suggested

that there was no racial bias in the LSI-R. Analysis focused
on 11792 probationers in Kansas (74.7% White and
25.3% Black). Risk classifications and total scores
produced similar levels of predictive accuracy between
the two groups.

•A meta-analysis of 32 articles and 12 data sets was

undertaken to examine whether the LSI-R was applicable
to Aboriginal individuals. Results indicated that all of the
Central Eight risk/need factors were predictive of general
and violent recidivism for Aboriginal individuals. Some of
the factors demonstrated significantly better predictive
validity for non-Aboriginal individuals: criminal history,
alcohol/drug and antisocial pattern (Gutierrez et al.,
2013).

Validation History

Applicability: Mental Disorders

RATED page updated: July 2019

© Risk Management Authority 2019
 a) UK Research None at present.

b) International Research •Harris, Rice and Quinsey (1993) found large weighted
correlations ranging between .43 and .53 between items
in the LSI and violent recidivism in a male psychiatric
sample. Recidivists also tended to attain significantly
higher scores on the tool than non-recidivists.

•A study assessed 193 detainees who were undergoing

a forensic psychiatric investigation in Stockholm. The
predictive validity of the LSI-R was medium, generating an
AUC of .70 (Persson et al., 2017).

Contribution to Risk Practice

•The LSI-R has the ability to create awareness of a number of static and dynamic risk factors
pertinent to the individual’s general risk of recidivism. Information obtained through the LSI-R can
inform the level and focus of monitoring and supervision strategies.
•The tool can aid on-going evaluation of an individual’s risk of reoffending and their criminogenic
needs.

Other Considerations

•Fewer validation studies conducted with other populations such as ethnic minority groups and
mentally disordered individuals.
•Requires refresher training - experience and training in the LSI-R can affect the reliability of the
instrument (Lowenkamp et al., 2009).
•The tool is a quantitative survey of risk-need factors that are supported by research, professional
opinion and social learning theory on criminal behaviour. It is not a comprehensive measure of
mitigating and aggravating risk factors related to risk practices for offending (Andrews and Bonta,
1995).
•The LSI-R should be completed using information obtained from interviews with the individual and
other collateral sources of information.
•The score of the LSI-R was found to correlate with the HCR-20V3 and the SAPROF at a considerable
rate; although the correlations between the risk or protection categories were poorer (Persson et
al., 2017).

RATED page updated: July 2019

© Risk Management Authority 2019