Estimating the prevalence of substance abuse with social indicators

Governments are increasingly interested in estimating the prevalence of substance abuse with social indicators, largely because of the high cost of estimating prevalence with surveys of random samples of the population. With both the individual and county as the unit, we regress measures of the use of alcohol, marijuana, and other drugs on social indicators that fall into three categories: demographics, measures of social disorganization, and measures more directly related to the use of substances. The measures of explained variance are fairly low, but even more troubling is that the effects of several social indicators are in the "wrong" direction. Reliance on social indicator data to supplant survey estimates of the prevalence of substance abuse requires further validation, attention to sources of bias in the indicator data, and replication of the models over time.