Understanding Sex-Offense Risk Assessment

Go where you need to go

If you only need help understanding a report in front of you, Sections 1–4 are the best place to start. The later sections explain particular tools, research findings, and questions to ask when the assessment is being used in a decision.

Depending on the setting, you may be interviewed, records may be reviewed, one or more instruments may be scored, treatment providers or supervision officers may add information, and a final report may combine several different kinds of conclusions.

Different settings also use assessment for different purposes. Depending on the case, an assessment may inform sentencing, treatment planning, supervision intensity, release planning, institutional decisions, civil proceedings, or another specific decision. The purpose matters because a tool designed for one question should not automatically be treated as answering every other one.

If the assessment is part of a federal criminal case, the Federal Sex-Crime Process Guide explains where sentencing and related evaluations fit in the larger federal process.

You might see all of the following in one document:

• Historical or static risk score: a score based on historical factors;
• Dynamic or change-sensitive assessment: an assessment of factors intended to change over time;
• Risk category or relative-risk level: a group classification or comparison;
• Reference-group recidivism estimate: an observed or estimated rate tied to a comparison population;
• Treatment targets or needs: issues identified for treatment or intervention;
• Supervision recommendations: recommendations about case management or supervision;
• Professional judgment or an override: a conclusion or adjustment beyond the instrument output;
• Other case-specific comments: additional information the evaluator considers relevant.

The practical problem is that these pieces can look like one unified scientific conclusion even when they came from different methods and answer different questions.

That distinction matters because a recommendation can be stricter, more lenient, or simply different from the instrument output. The difference may reflect another assessment, a treatment issue, an agency rule, case-specific information, or an explicit professional override.

A fictional anatomy of a risk-assessment result

Imagine that a report contains the following kinds of entries. These are deliberately illustrative rather than real scoring instructions:

• Instrument: Static-99R
• Raw score: [score]
• Risk level: [risk category]
• Relative risk: [comparison with a reference group]
• Estimated five-year rate: [percentage for the applicable norm group]
• Professional conclusion: [treatment, supervision, or case recommendation]

Those lines are related, but they are not interchangeable.

A real report may include only some of these layers. For example, it may give a score and category without an absolute percentage, or it may discuss treatment needs without reporting a separate actuarial estimate. The point of this example is to show how different kinds of information fit together when they do appear.

A five-step interpretation workflow

Relative risk vs. absolute risk

Relative risk answers a comparison question: higher or lower compared with whom? Absolute riskis an observed or estimated event rate over a defined period, such as a five-year rate in a particular reference group.

A person can be higher than a low-risk comparison group while the absolute event rate remains modest. A dramatic-sounding relative difference does not automatically mean a high individual probability.

Group prediction vs. individual certainty

Risk instruments estimate patterns across groups and place an individual within those empirical patterns. They do not observe the future. A group rate is evidence about a reference group, not an individual destiny.

Outcome definition and follow-up period

Rearrest, charge, reconviction, reincarceration, self-report, and detected offending are not interchangeable. Official outcomes can miss undetected conduct; self-report has different limitations.

A recidivism rate is incomplete unless it tells you the outcome definition, population, follow-up period, and starting point.

A five-year rearrest rate beginning at supervision start is not the same quantity as a ten-year reconviction rate beginning at release. Comparing them as if they were the same can create false precision.

Validation population and population fit

Every validation study has a population: a jurisdiction, setting, offense mix, sex composition, age range, entry point, and follow-up design. A tool validated in one population is not automatically calibrated for another.

Before relying on a percentage, ask which reference group generated it and whether that group resembles the person and setting at issue.

Static vs. dynamic factors

Static factors are historical facts that do not change because time has passed or treatment has occurred: for example, parts of a person's prior offense or supervision history. Static tools are mainly about baseline group risk.

Dynamic factors are intended to capture risk-relevant characteristics that can change. Some change over months or years; others may shift much more quickly.

A dynamic score is not a promise that change has been measured perfectly. It is an attempt to add current, change-sensitive information to historical baseline information.

Actuarial, structured professional judgment, and unstructured judgment

Actuarial tools use specified empirical items and scoring rules to place people into relative risk groups or categories.

Structured professional judgment (SPJ) also uses a defined framework, but leaves more room for professional synthesis of case information.

Unstructured judgment is professional intuition without a comparable standardized empirical structure.

Meta-analytic evidence in the SOLAR evidence matrix supports empirically derived actuarial approaches over unstructured professional intuition on average, with SPJ performing differently from both and generally falling between them.

Base rates

A base rate is how often the outcome occurs in the population being studied before a particular score is considered. When the outcome is uncommon, precise individual prediction becomes harder.

Even a tool that sorts people better than chance will still make errors when applied to a low-frequency outcome.

AUC: ranking, not probability

AUC is a discrimination statistic. In plain English: if you randomly select one person who later had the measured recidivism outcome and one who did not, the AUC estimates how often the tool ranks the person with the later outcome as higher risk.

Calibration

Calibration asks a different question: do the predicted or reference-group percentages line up with the observed rates in the population where the tool is being used?

A tool can rank people reasonably well and still overpredict or underpredict absolute rates in another setting.

The workbook flags this issue for Static-99R/Static-2002R and CPORT. Static meta-analytic research found more stability in relative predictive accuracy than in absolute rates across samples. A Spanish CPORT validation found observed sexual recidivism substantially below developer expectations, illustrating why population-specific norms and reference groups matter.

Readers who want to go deeper into SOLAR's research sources, definitions, and evidence base can continue to Research & Data Resources.

Static-99R

Static-2002R

STABLE-2007

STABLE-2007 is designed to assess relatively stable but changeable risk and need factors relevant to sexual recidivism. Ratings draw on structured interview, file, treatment, and supervision information.

The output is used for risk/need formulation and treatment or supervision planning rather than as a stand-alone long-term probability.

Where might you encounter it? Most often in treatment or community-supervision settings where professionals are trying to understand changeable risk and need factors rather than relying only on historical baseline information.

A prospective Canadian study in the workbook followed 768 community-supervised adult men and found that STABLE measures predicted sexual, violent, and any recidivism. The workbook treats this as evidence that dynamic information can add clinically and practically relevant information beyond static history.

ACUTE-2007

ACUTE-2007 is aimed at shorter-term, rapidly changing concerns during community supervision. It is meant to be reassessed repeatedly and interpreted in its supervision context.

Where might you encounter it? In ongoing community supervision or monitoring where short-term changes may matter to immediate case management.

It does not establish a person's long-term actuarial risk and does not diagnose dangerousness.

CPORT

CASIC

CASIC is a structured proxy/index related to evidence of sexual interest in children. It was developed in part to operationalize a CPORT-related factor when direct admission or other evidence is unavailable.

Where might you encounter it? Within CPORT-related CSEM assessment work where an evaluator needs a structured way to code the relevant sexual-interest factor.

The workbook describes CASIC as relying on historical or behavioral correlates, not as a stand-alone recidivism instrument.

PCRA and PCRA-R

The federal Post Conviction Risk Assessment family is used for general recidivism risk, criminogenic needs, supervision planning, and allocation of intervention resources in federal post-conviction supervision.

It combines criminal-history information with dynamic needs and officer-assessment fields.

Where might you encounter it? In federal post-conviction supervision, where probation officers use a general risk-and-needs framework to support supervision planning. For the practical rules and decisions that may follow, see SOLAR's Supervision Conditions Survival Guide.

That scope distinction matters. PCRA can include factors that correlate with sexual recidivism, but its primary validated purpose is general federal risk and needs.

Correlation with a specialized outcome does not transform it into a specialized sexual-risk probability calculator.

SOTIPS

The Sex Offender Treatment Intervention and Progress Scale is a structured, change-sensitive assessment used in treatment and supervision contexts. It measures dynamic, treatment-relevant factors over time rather than asking only what happened in the past.

Where might you encounter it? In community treatment or supervision where practitioners are monitoring treatment-relevant needs and change over time.

The primary validation study in the workbook involved 759 adult men under correctional supervision in Vermont community treatment. SOTIPS ratings predicted sexual, violent, and any recidivism and return to prison.

Reductions in SOTIPS scores were associated with lower recidivism, and combining SOTIPS with Static-99R improved prediction in that validation study.

VRS-SO

The Violence Risk Scale–Sexual Offense version combines static and dynamic information. It is designed to assess baseline risk, treatment targets, and treatment-related change and is more intensive than a quick screening instrument.

Where might you encounter it? In more intensive treatment or evaluation settings where baseline risk and treatment-related change are both being assessed.

The workbook's foundational validation involved 321 adult men and found prediction of sexual and nonsexual violent recidivism over an average follow-up of about ten years.

Later multisite work developed updated risk categories and five- and ten-year recidivism estimates using pretreatment risk and change information.

SPJ uses defined risk factors and a structured process, while leaving room for professional synthesis. That makes it meaningfully different from an unstructured impression such as “this person feels dangerous” or “my experience tells me the score is wrong.”

The meta-analytic evidence in the workbook found empirically derived actuarial approaches more accurate than unstructured professional judgment across sexual, violent, and any recidivism outcomes.

SPJ performance was intermediate between actuarial and unstructured judgment in that synthesis.

The practical lesson is not “scores only.” Individualized professional information can matter.

The lesson is that a professional opinion does not become individualized science merely because a professional expresses it. A defensible assessment should connect judgment to a structured method, relevant case facts, and an empirical baseline.

The federal CSEM risk-tool study examined a validation cohort of 5,768 male federal CSEM supervisees using a fixed 60-month follow-up beginning at supervision start or initial PCRA assessment.

• Outcome: rearrest for any new sexual offense.
• Observed five-year rate: 4.5% (262 of 5,768).
• Contact sex-crime rearrest: fewer than 1%.
• PCRA discrimination: AUC .61 (95% CI .58–.64).
• CPORT discrimination: AUC .62 (95% CI .58–.65).

Those facts teach several different things at once. The outcome was rearrest, not all offending. The population was male, federal, and CSEM-specific. The follow-up was fixed at five years. The sexual-rearrest base rate was low. And both AUCs showed modest ranking ability rather than individualized certainty.

Use the five-step workflow first

Identify → Translate → Compare → Check → Question.

That short workflow gets you oriented. The fuller checklist below is for the details that matter when a score affects treatment, supervision, sentencing, litigation, reporting, or policy.

If the disagreement involves a government action or legal process, SOLAR's Your Rights at Every Stage guide can help identify the separate legal questions. A disagreement about scoring or interpretation, by itself, does not establish a rights violation.

• “Low risk” does not mean zero risk.
• “High relative risk” does not automatically mean a high absolute probability.
• AUC .70 does not mean a 70% chance of recidivism.
• A group recidivism rate is not an individual destiny.
• A static score is not moral severity.
• CASIC is not a pedophilia diagnosis.
• CPORT is not proof someone will offend.
• PCRA is not a specialized sexual-risk probability calculator.
• Dynamic improvement does not guarantee no future offending.
• A professional recommendation is not automatically the same thing as the instrument result.
• Unstructured intuition is not individualized science simply because a professional expresses it.