RTI uses cookies to offer you the best experience online. By clicking “accept” on this website, you opt in and you agree to the use of cookies. If you would like to know more about how RTI uses cookies and how to manage them please view our Privacy Policy here. You can “opt out” or change your mind by visiting: http://optout.aboutads.info/. Click “accept” to agree.
Statistical methods in public health and epidemiology: a look at the recent past and projections for the next decade
Levy, P., & Stolte, K. (2000). Statistical methods in public health and epidemiology: a look at the recent past and projections for the next decade. Statistical Methods in Medical Research, 9(1), 41-55.
This article attempts to prognosticate from past patterns, the type of statistical methods that will be used in published public health and epidemiological studies in the decade that follows the millennium. With this in mind, we conducted a study that would characterize trends in use of statistical methods in two major public health journals: the American Journal of Public Health, and the American Journal of Epidemiology. We took a probability sample of 348 articles published in these journals between 1970 and 1998. For each article sampled, we abstracted information on the design of the study and the types of statistical methods used in the article. Our major findings are that the proportion of articles using statistical methods as well as the mean number of statistical methods used per article has increased dramatically over the three decades surveyed. Also, the proportion of published articles using study designs that we classified as analytic has increased over the years. We also examined patterns of use in these journals of three statistical methodologies: logistic regression, proportional hazards regression, and methods for analysis of data from complex sample surveys. These methods were selected because they had been introduced initially in the late 1960s or early 1970s and had made considerable impact on data analysis in the biomedical sciences in the 1970s-90s. Estimated usage of each of these techniques remained relatively low until user-friendly software became available. Our overall conclusions are that new statistical methods are developed on the basis of need, disseminated to potential, users over a course of many years, and often do not reach maximum use until tools for their comfortable use are made readily available to potential users. Based on these conclusions, we identify certain needs that are not now being met and which are likely to generate new statistical methodologies that we will see in the next decade