Incremental and Predictive Validity - Appendix B 1

Appendix B: “Been In a Fight” Item Removed from Baseline/Follow-Up SRO

Descriptive Statistics and Sex/Racial Differences in APSD Scores and Violent Offending
The rate of prior offending in the sample was 18.5% (n = 62). At the 6-month follow-up, the mean number of violent offenses was 0.12 (SD = 0.47, range 0 – 5). Males had a significantly higher number of violent offenses (Mean rank = 176.50) compared to females (Mean rank = 160.25), U = 12442, p = .002, r = .17. The number of violent offenses did not significantly differ as a function of whether the youth was South Asian, Asian, Caucasian, or a member of another ethnic minority group (p = .980, η2H = .01).

Association between the APSD, Common Criminogenic Factors, and Violent Offending
APSD total and subscale scores were significantly positively associated with violent offending (rs = .15 to .27, p < .001 to .007). Youth with higher levels of parental monitoring were significantly less likely to engage in violent offending (rs = -.26, p < .001). In addition, youth with prior offending, delinquent peer affiliation, poor school achievement, and substance use were significantly more likely to engage in violent offending (rs = .16 – .22, p < .001 to .003).[1]
Predictive Validity of the APSD
In the current study, the base rate of any offending at the 6-month follow-up was 8.7% (n = 29). For violent offending, AUCs were .78 (95% CI [.69, .86], rpb = .27, p < .001), .68 (95% CI [.58, .79], rpb = .21, p = .001), .72 (95% CI [.63, .81], rpb = .21, p < .001), and .65 (95% CI [.54, .76], rpb = .16, p = .007) for the APSD total, IMP, NAR, and CU scales, respectively. As shown in Table B1, more liberal APSD cut-off scores of 6, 9, and 12 generated low specificity and PPP, but high values for sensitivity and NPP. In contrast, more conservative APSD cut-off scores of 15 and 17 generated low sensitivity and NPP, but high values for specificity and PPP.
--Insert Table B1 about here--
Do APSD Total Scores Add Incrementally to the Prediction of Violent Offending?
Prior to conducting analyses, self-reported violent offenses were examined to determine if overdispersion was present (i.e., a high proportion of zero responses causing the variance to be greater than the mean of each distribution). Overdispersion was not present in the data (z = - 0.03, p = .513), thus violent offending was examined using Poisson regression models. As shown in Table B2, adding psychopathic features to the violent offending model did not provide incremental predictive utility over the demographic control variables and criminogenic factors in the model.
--Insert Table B2 about here--
Similarly, when subscale scores were entered in block two in place of APSD total scores, neither IMP, Exp (B) = 1.09, p = .521, NAR, Exp (B) = 1.04, p = .664, nor CU subscale scores, Exp (B) = 1.24, p = .068, added incremental utility above and beyond the block one variables,
χ2 (13) = 67.09, p < .001, Δ χ2 (3) = 0.24, p = .108.
Does the Predictive Utility of the APSD Vary Across Sex and Ethnicity?
Moderation analyses were conducted controlling for demographic and criminogenic factors found to be significantly associated with violent offending in the incremental validity analyses. There were no significant differences in the ability of the APSD total or subscale scores to predict violent offending across male and female to predict violent offending, Exp (B) = 0.83 to 1.32, p = .266 to .816. In addition, ethnic background did not significantly moderate the relationship between APSD total or subscale scores and violent offending, Exp (B) = 0.69 to 1.23, p = .395 to .959.

Table B1. Predictive Accuracy of APSD Total Scores at Various Cut-Off Points

Percentile Cut-Off / APSD
Cut-Score / Sensitivity / Specificity / PPP / NPP
25th / 6 / 96.56% / 20.20% / 10.41% / 98.39%
50th / 9 / 93.10% / 46.71% / 14.29% / 98.61%
75th / 12 / 65.52% / 74.43% / 19.59% / 95.78%
90th / 15 / 34.49% / 91.83% / 28.57% / 93.67%
95th / 17 / 13.79% / 94.44% / 19.05% / 92.04%


Note. APSD = Antisocial Process Screening Device. PPP = Positive Predictive Power. NPP = Negative Predictive Power.

Table B2. Incremental Predictive Validity of the APSD over Demographic Control Variables and Common Criminogenic Factors

b (SE) / Z / p / Exp (B), 95% CI
Step 1
Age / -0.27 (0.56) / -0.49 / .627 / 0.76 [0.25, 2.28]
Male / -0.00 (0.02) / -0.10 / .918 / 1.00 [0.97, 1.03]
Asian / 0.31 (0.51) / 0.61 / .543 / 1.36 [0.50, 3.70]
Caucasian / 0.22 (0.48) / 0.45 / .654 / 1.24 [0.48, 3.19]
Other Ethnic Minority / -1.36 (0.78) / -1.74 / .082 / 0.26 [0.06, 1.19]
Prior Offending / 0.00 (0.45) / 0.00 / .996 / 1.00 [0.41, 2.43]
Delinquent Peers / 2.37 (0.65) / 3.62 / < .001 / 10.65 [2.95, 38.39]
School Failure / 0.97 (0.43) / 2.23 / .026 / 2.63 [1.13, 6.16]
Substance Use / 2.41 (0.85) / 2.83 / .005 / 11.14 [2.10, 59.06]
Parental Monitoring / -0.20 (0.13) / -1.54 / .123 / 0.82 [0.64, 1.06]
Model / χ2(10) = 66.85, p < .001
Step 2
Age / -0.25 (0.58) / -0.43 / .667 / 0.79 [0.25, 2.43]
Male / -0.00 (0.02) / -0.08 / .936 / 1.00 [0.97, 1.03]
Asian / 0.39 (0.51) / 0.78 / .439 / 1.48 [0.55, 4.00]
Caucasian / 0.16 (0.50) / 0.32 / .750 / 1.17 [0.44, 3.11]
Other Ethnic Minority / -1.36 (0.76) / -1.78 / .076 / 0.26 [0.06, 1.15]
Prior Offending / -0.09 (0.44) / 0.20 / .844 / 0.92 [0.38, 2.18]
Delinquent Peers / 2.38 (0.67) / 3.53 / < .001 / 10.78 [2.88, 40.36]
School Failure / 1.11 (0.44) / 2.52 / .012 / 3.02 [1.28, 7.15]
Substance Use / 1.71 (0.96) / 1.78 / .075 / 5.53 [0.84, 36.27]
Parental Monitoring / -0.12 (0.14) / -0.85 / .395 / 0.89 [0.68, 1.16]
APSD / 0.09 (0.06) / 1.65 / .100 / 1.09 [0.98, 1.21]
Model / χ2(11) = 69.43, p < .001, Dχ2(1) = 2.58, p = .108

Note. APSD = Antisocial Process Screening Device. CI = Confidence intervals. * p < .05, ** p < .01, *** p < .001 (two tailed). All significant Exp (B) values did not contain one based on 95% CIs.

[1] With respect to demographic control variables, violent offending was significantly associated with sex (rs = .17, p = .002), but not with age (rs = -.08, p = .127) nor ethnicity (rs = -.01, p = .879).