The study of rare events data in which observations of non-event outcomes far outnumber event outcomes makes inference under these circumstances quite difficult. Ideally, for a binary dependent variable, one would like sample data to contain enough observations from both outcome categories. With rare events data, however, this is usually impossible and/or costly to achieve with random sampling. This exploratory research aims to find a set of potential predictors that could be used to quantify a person's risk for developing autism spectrum disorder. A more efficient data collection strategy will be employed that allows for a smaller sample size of more meaningful data. Then, a statistical correction to the standard logistic regression model will be applied to yield adjusted predictions that take into account the prevalence of autism cases both in the sample data and in the population of interest. |