Recorded presentation

Lynne C. Messer, MPH, Department of Epidemiology, The University of North Carolina at Chapel Hill, CB # 7435, School of Public Health, Chapel Hill, NC 27599, (919)843-9472, lmesser@email.unc.edu, Barbara A. Laraia, PhD, MPH, RD, Department of Nutrition, The University of North Carolina at Chapel Hill, Carolina Population Center, CB #8120, Chapel Hill, NC 27599-8120, Jay Kaufman, PhD, Department of Epidemiology, UNC School of Public Health, 2104C McGavran-Greenberg Hall, Pittsboro Road, CB#7400, Chapel Hill, NC 275997400, and Paul A. Buescher, PhD, DHHS, State Center for Health Statistics - North Carolina, 1908 Mail Service Center, Raleigh, NC 27699-1908.

Objective: To present a data reduction methodology for highly correlated census variables to estimate a single neighborhood deprivation index for use in multilevel analyses of preterm birth. Methods: Hierarchical models approximating community-level effects on preterm birth often make use of administrative data. Census variables, the most commonly used aggregated data representing area level socioeconomic status, are highly correlated; therefore a series of single item associations may represent the same underlying construct. A literature review produced sixteen studies using administrative data, mostly census, to assess contextual level effects on adverse birth outcomes. It found census variable selection, deprivation index usage and census index development methods inconsistent across studies. Informed by the literature, we selected 32 census variables to associate with preterm birth for White and Black women. Principal components analysis was used for data reduction and index development. Findings: Of the 32 census variables, 15 variables were associated with preterm birth for White or Black women. Principal components analysis produced 10 items with high factor loadings (~0.3) on one component representing neighborhood deprivation with an eigenvalue of 6.86. The neighborhood deprivation index had high internal reliability (Chronbach’s Alpha = 0.95). Conclusion: The principal components analysis 15 item factor yielded 68% of the variability in the latent construct, neighborhood deprivation, with high internal reliability. This finding was consistent across four geographical sites. The highly correlated nature of census variables associated with preterm birth makes principal components analysis a useful tool for research into neighborhood effects on health using census data.

Learning Objectives:

• To present a review of various census deprivation indices and a data reduction methodology for developing a single neighborhood deprivation index from highly correlated census variables.

Keywords: Statistics, Infant Health

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.

Recorded presentation