Some Issues on Application of Standardization and Decomposition Analysis

This article is to 1) address some issues when multiple populations and/or sample uncertainty are involved in application of standardization and decomposition analysis (SDA); and 2) introduce a computer program that provides opportunity of dealing with such issues. In the recent issues of AJPH, there are some interesting discussions on the application of a demographic technique – standardization and decomposition analysis (SDA) – for studying temporal trends in low-birthweight (LBW) rates in the U.S. [1-3]. The purpose of this study is to promote further discussion and application of SDA, as well as to shed some light on the issues debated between Schempf & Becker [2] and Yang et al. [3]. It is well-known that differences in crude rates between populations are often confounded by population compositions or distributions of confounding factors. SDA can be used not only to adjust the crude rate, but also to decompose the difference in the crude rates between populations into component effects, such as the “rate effect,” representing the “real” rate difference, and “factor composition effects,” representing the rate difference attributed to compositional differences or distribution differences in specific confounding factors [4-9]. For example, in Yang et al.’s study [1] change in LBW rate in the U.S. population during a given decade was decomposed into a rate effect and a factor effect; the former represents the “real” changes in parity-specific LBW rates, and the latter represents the changes in the observed crude LBW rates that were attributed to the changes in age and parity compositions in the U.S. child-bearing women population during a given decade. Although three factors (i.e., mother’s age, birth parity, and ethnicity) were actually involved in the study, mothers’ age and birth parity were combined as one factor, and one-factor SDA was conducted by ethnic group for periods 1980-1990 and 1990-2000, separately. The key dispute in the debate between Schempf & Becker and Yang et al. is about the standardization solution. In Yang et al’s study [1] rates and factor distributions were standardized based on different populations (e.g., 1980 and 1990), respectively. Schempf & Becker [2] suggest a symmetric solution using the average rate and average factor distribution as the weights. *Address correspondence to this author at the Children Research Institute (CRI), Children’s National Medical Center (CNMC), The George Washington University School of Medicine, 111 Michigan Ave. N.W., Washington, D.C. 20010, USA; Tel: 202-476-2978; E-mail: [email protected] To better understand the debate between Schempf & Becker [2] and Yang et al. [1, 3] I rewrite Equation 2 in Schempf & Becker [2] as following: