Estimating prevalences of sensitive problems from nonsensitive data

Accurate and regularly updated prevalences are critical to guide both prevention and policy. However, despite their impact on public health, the prevalence of behavioral and mental health problems such as child maltreatment and intimate partner violence (IPV) are collected infrequently (e.g., once per decade or less). The purpose of this study is to test the viability of using nonsensitive, dynamic variables to estimate prevalences of sensitive behavioral problems. Several archival data sets of partner and child maltreatment are used. Data sets are randomly divided into development and cross-validation subsets. Sequential, backward, stepwise logistic regression is used to derive estimation equations, which are then tested in the cross-validation subset. The results indicate that estimated prevalences are close to the measured prevalences. Confidence intervals (95%) are approximately the same as those that pertain to measured prevalences (±1%-2%). In situations where it is either too expensive or too sensitive to assess a population regularly on a problem of interests, it appears as if policy makers and prevention planners can use regularly collected, nonsensitive data to estimate accurately the prevalence of unmeasured, sensitive outcomes.