In the largest study, Harer and Langan (2001) investigated the predictive validity of the US Federal Bureau of Prisons' initial risk classification instrument with women's and men's violent misconducts in prison.
In operational terms, internal security risk is measured by institutional misconduct. For a particular tool to be valid, researchers need to demonstrate that it is correctly identifying different levels of risk in prisoners (i.e., that it is identifying the likelihood that offenders will commit misconducts).
The first aim of the present study was to describe the internal security risk posed by women prisoners in the form of recorded misconducts. The second aim was to describe how that risk was distributed by security classification designation (i.e., by the final security classification outcome rather than by the security classification instrument total or item scores).
For example, prisoners who are cooperating with staff and living within the prison rules are likely to have their security classification reduced on review, whereas prisoners who are uncooperative and have misconducts are likely to remain at the same security classification or have their security classification increased.
During the study period, 152 women (17% of the total sample) committed 481 misconducts. The number of misconducts per prisoner ranged from 0 to 22: 83% had none, 10% had 1 or 2, and 7% had 3 or more.
Those with misconducts were significantly younger (M = 26.5 years, SD = 7.3 years) than women who did not have a misconduct (M = 31.3 years, SD = 9.5 years); t(234) = 6.6, p < .01, d = .56.
The proportion of misconducts committed by prisoners at each security classification varied: 29% of all misconducts were committed by women classified as a minimum security risk, 37% by low-medium women, 25% by high-medium women, and 3% by women classified as maximum security.
In order to investigate misconducts further, we collapsed the range of infractions listed in Table 2 into 3 categories.
So far the analyses tell us about the type and rate of misconducts at each security classification, and not about the proportion of women at each classification who did or did not commit misconduct.
For all misconduct categories, the proportion of women who committed misconducts at minimum security was significantly smaller than the proportion of women who committed misconducts at low-medium: assaults, [X.sup.2](1, N = 785) = 8.4, p < .01; alcohol/drug, [X.sup.2](1, N = 785) = 10.1, p < .01; disruptive/destructive, [X.sup.2(1, N = 785) = 27.6, p < .01; and any misconduct, [X.sup.2](1, N = 785) = 24.7, p < .01.
On average, women who committed misconducts were younger more likely to be in prison for a violent offence, and serving a longer sentence than those who did not.