Comparison of Decision Tree and Logistic Regression Models for Utilization in Sexual Assault Kit Processing.

To combat the influx of sexual assault kits (SAKs) that need to be tested, an exploration of data from Ohio's SAK Testing Initiative was carried out to identify variables that impact whether a SAK contains a probative DNA profile that is eligible for the Combined DNA Index System (CODIS) database. A validation study was completed to confirm the existence of variable relationships from the initial examination of data; new and modified statistical models were introduced to improve the predictive accuracy to determine if a SAK will contain at least one CODIS eligible DNA profile. Descriptive statistics from the validation data set confirmed conclusions about the effects of days between the assault and kit collection, the age of the victim, and consensual sex around the time of the kit collection for obtaining CODIS eligibility of DNA. The decision tree was selected as the best model.

Statistical Modeling of the Case Information From the Ohio Attorney General's Sexual Assault Kit Testing Initiative.

The Ohio Attorney General's Sexual Assault Kit (SAK) Testing Initiative has resulted in nearly 14,000 kits being processed since the initiation of the project in 2012. A logistic regression model was fit to the data from 2500 SAKs in order to determine the probability of obtaining at least one Combined DNA Index System (CODIS) eligible DNA profile based on a number of predictor variables. The probability of obtaining at least one CODIS eligible DNA profile from an SAK varied as a function of (i) days to kit collection following a sexual assault; (ii) years to kit submission to the laboratory for testing following kit collection; (iii) the age of the victim; and (iv) the occurrence of victim-reported consensual sex around the time of the assault and/or kit collection. These findings demonstrate the utility of the statistical modeling of data obtained from the "forklift" testing approach of sexual assault kits.

The effect of cerclage in twin gestations with short cervix: a Bayesian evaluation.

OBJECTIVE: Bayesian inference allows the revision of prior clinical estimates of treatment effectiveness based on current data. We apply it to a published dataset evaluating the effect of cerclage upon preterm delivery in twin gestations with a short cervix. STUDY DESIGN: Prior probability distributions for delivery <35 weeks gestation for the control group and the treatment (cerclage) group were constructed under assumptions ranging from treatment having no effect (prior A) to halving early deliveries (prior C). Likelihood functions were calculated based on a published meta-analysis. Posterior probability densities were derived from which risk ratios for early delivery were computed, with 95% credible intervals and the probability of cerclage benefit. RESULTS: Median posterior risk ratios (95% credible intervals) for delivery <35 weeks with cerclage are 1.51 (1.02-2.33) for prior A and 1.11 (0.72-1.77) for prior C. The probability of cerclage benefit ranged from 2.1% for prior A to 31.4% for prior C. By comparison, the conventional risk ratio (95% confidence interval) for early delivery, based on the data alone, is 2.08 (1.18-3.69). CONCLUSIONS: As might be anticipated, those with low expectation of cerclage benefit remain more convinced of the ineffectiveness (or harm) of the procedure than those with higher expectations.

Probabilistically Valid Inference of Covariation From a Single x,y Observation When Univariate Characteristics Are Known.

Participants were asked to draw inferences about correlation from single x,y observations. In Experiment 1 statistically sophisticated participants were given the univariate characteristics of distributions of x and y and asked to infer whether a single x, y observation came from a correlated or an uncorrelated population. In Experiment 2, students with a variety of statistical backgrounds assigned posterior probabilities to five possible populations based on single x, y observations, again given knowledge of the univariate statistics. In Experiment 3, statistically naïve participants were given a problem analogous to that given in Experiment 1, framed verbally. Experiment 4 replicated Experiment 3 but added an "impossible to determine" response option. Models that rely on computing sample correlations make no predictions about these investigations. From a Bayesian perspective, participants' inferences in all four experiments tended to make probabilistically valid inferences as long as the single datum was directional. The results are discussed in light of the Brunswikian notion of vicarious functioning.