Can evaluators be confident in the relative-risk differences implied by Static-99R scores?

Seung C. Lee, Carleton University; Corrections Research, Public Safety Canada
R. Karl Hanson, Corrections Research, Public Safety Canada

Objective: The purpose of this study was to develop new 10-year recidivism rate norms as well as to update 5-year norms for the Static-99R risk tool for routine/complete samples. We also present the extrapolated sexual recidivism rates from these new 10-year norms for follow-up periods of 11 to 20 years. Hypotheses: We hypothesized that absolute-recidivism base rates (B02; i.e., the intercept centered on the median score of 2) would vary; however, the relative predictive accuracy (i.e., discrimination; B1) would be stable across samples. In addition, compared with the estimated sexual recidivism rates with a fixed 5-year follow-up time, the estimated rates with a fixed 10-year follow-up time would be expected to be consistently higher across the Static-99R scores. Method: The current study included 12 independent samples (N = 7,224 for the 5-year recidivism rate norms; N = 1,599 [k = 6] for the 10-year norms) classified as routine/complete samples, that is, relatively random samples from a correctional system. Logistic regression parameters (B02 and B1) across the studies were aggregated using fixed- effect meta-analyses. Results: There was statistically significant variability in the base rates (B02), whereas the between-sample variability in the relative-risk parameters (B1) was no more than would be expected by chance. As expected, the 10-year base rates were approximately 1.5 times higher than the 5-year base rates (7.20% vs. 4.58%), and the extrapolated 20-year sexual recidivism rates were approximately double the observed 5-year sexual recidivism rates. Conclusions: The current study provides empirical evidence to estimate 5- and 10-year sexual recidivism rates based on Static-99R total scores. Evaluators who are especially concerned about long-term sexual recidivism risk (e.g., civil commitment) can report the expected sexual recidivism risk based on the new 10-year norms and the extrapolated sexual recidivism rates for follow-up periods of 11 to 20 years.

sexual recidivism rates, Static-99R, norms, risk assessment

“In the last few decades, there has been substantial research indicating that structured approaches (e.g., actuarial or structured professional judgment [SPJ]) provide more accurate assessments of reoffending risk than unstructured professional judgment. Actuarial approaches combine empirically identified risk factors (e.g., prior criminal history, male victims) into an overall evaluation of risk. Total scores or risk levels are then linked to a nominal risk category, such as ‘average,’ and also to estimated recidivism probabilities” (p. 24-25).

“There is a long-standing controversy over what standards should be taken into consideration in deciding whether scientific evidence presented by experts is admitted or not in court. The 1923 Frye standard requires only that the method of forming the expert opinion has ‘general acceptance’ in the field to judge the admissibility of scientific information. Under the Frye standard, it is possible for unreliable evidence to be admitted when the method has achieved general acceptance, even if there is no evidence (and no professional consensus) to support the specific conclusion (e.g., unstructured clinical judgment of dangerousness)” (p. 25).

“Accurate actuarial recidivism risk-assessment tools require regular updates. The current study updated the 5-year recidivism rate norms and developed new 10-year norms for the Static-99R for the routine/complete samples. In comparison to previous norms, the updated 5-year norms were adjusted slightly downward for Static-99R total scores. For example, the 2016 norms estimated a 5-year rate of 5.6% for a score of 2, whereas the current update estimated a rate of 4.6%. For a score of 6 (the entry score for the well-above-average risk group), the decline was from 20.5% (the old estimate) to 17.6% (the updated estimate). These changes are unlikely to meaningfully influence decisions concerning whether individuals meet specific risk thresholds; the influence of risk factors external to Static-99R is much greater than the effect of updated norms. For example, the estimated rate for preselected high-risk samples is 11.3% for a score of 2 at 5 years (2016 norms) compared with rates for routine/complete samples of 5.6% (2016 norms) or 4.6% (current best estimates)” (p. 33).

“[T]he estimated sexual recidivism rates for a fixed 10-year follow-up period were approximately 1.5 times higher than the updated 5-year norms of the routine/complete samples (adjusted base rates [B02] were 4.58% and 7.20%, respectively, a risk ratio of 1.57). This finding supports the offense-free effect model in which the risk of sexual recidivism declines consistently and predictably when individuals remain sex offense–free in the community after release”(p. 33).

“The current study provides empirical evidence to estimate 5- and 10-year sexual recidivism rates by Static-99R score for individuals who have committed sexual crimes. Although the current norms are slightly lower than the previous 5-year norms, it is unlikely that these changes would meaningfully influence decisions concerning whether individuals meet specific risk thresholds (e.g., SVP evaluation, parole decision). The variance due to external (largely unmeasured) risk factors is much greater than the variance between the 2016 norms and those developed in this article. Nevertheless, updating the norms of risk-assessment tools with new samples provides some assurance that risk estimation remains relevant to current practice. As well, it is easier for evaluators and decision-makers to adjust to small, incremental changes than to be faced with potentially substantial changes from updates based on very different samples spaced many years apart” (p. 35).

“Evaluators who are especially concerned about a long-term recidivism risk (e.g., civil commitment) can report the expected sexual recidivism risk based on the new 10-year norms. Further, this study also provides estimated sexual recidivism rates for a follow-up period of 11–20 years based on the updated 5-year norms and the new 10-year norms. That information would be critical for decision-makers in the criminal justice system when the decision thresholds are related to lifetime risk (e.g., civil commitment, D.O. designation)” (p. 35).

“The general recommendation is to report the sexual recidivism rates for routine/complete samples. This is the most representative population of individuals who have been convicted of sexually motivated crimes. Specifically, we recommend against using the sexual recidivism rates for high-risk/high-need samples unless there is strong justification. Note, however, that the significant differences in sexual recidivism rates between routine/complete samples and high-risk/high-need samples are only in the lower and middle range of Static-99R scores (i.e., the scores between –1 and 6). The differences in risk decrease as the Static-99R scores increase, such that preselection effects make relatively little difference to the recidivism risk estimate when the Static-99R score is 6 or higher” (p. 35)

“To evaluate the predictive validity of Static-99R, more research is required to test how well the estimated recidivism probability, particularly from the new 10-year norms, corresponds with the observed recidivism probability of new samples (i.e., calibration). There are some promising statistical methods to evaluate the calibration: (a) calculating the E/O index (the ratio of the expected number of recidivists divided by an observed number of recidivists) and (b) comparing logistic regression parameters (B0) by fixed-effect meta-analysis” (p. 35).

“Further, it is necessary to evaluate if Static-99R systematically overestimates or underestimates the sexual recidivism risk for particular racial/ethnic groups (i.e., cultural bias in predictive validity), given the relatively small numbers of racial/ethnic minorities in development and normative samples (i.e., Asian, Latino, Black, Indigenous peoples). Research, however, should be cautious when interpreting unexpectedly low sexual recidivism rates for certain immigrant groups (i.e., poor calibration) because of high rates of deportation (e.g., individuals of Mexican descent in the United States). For example, after excluding the deportees from the calibration analysis, the sexual recidivism rates for individuals of Latino descent became comparable with the sexual recidivism rates for the norms of Static-99R” (p. 35).

As always, please join the discussion below if you have thoughts or comments to add!