Recorded presentation

Craig A. Mason, PhD, University of Maine, 5717 Corbett Hall, Room 3, Orono, ME 04469, (207)581-9059, craig.mason@umit.maine.edu and Shihfen Tu, PhD, College of Education and Human Development/UCEDD, University of Maine, 5717 Corbett Hall, Room 1, Orono, ME 04469.

A strategy for partitioning a population attributable fraction (PAF) in a multiple risk factor model is proposed. For example, maternal smoking is associated with lower infant birthweight, which is also associated with mild mental retardation. A researcher may therefore be interested in assessing the degree to which maternal smoking results in elevated rates of low birthweight among children in a population, and the degree to which this increased rate of low birthweight results in increased rates of mild mental retardation. This presentation introduces a procedure for estimating a PAF in sequential models, and presents both hypothetical and real-world data illustrating the application. This procedure is contrasted against two existing strategies for estimating a PAF in multiple risk factor models, neither of which translate into sequential models. The first of these techniques involves statistically controlling for confounding between risk factors. However, adjusting for the confounding between two risk factors does not adjust for the increased prevalence of the endogenous risk factor (e.g., low birthweight) due to the effect of the exogenous risk factor (e.g., smoking). A second common strategy involves stratifying the sample based upon combinations of the risk factors, and then estimating a PAF for each combination. These PAFs do not correspond to a sequential model and cannot be used to estimate a PAF for either risk factor when factors are correlated. The proposed technique addresses these limitations and results in transparent and logical partitioning of the overall PAF associated with both exogenous and endogenous risk factors.

Learning Objectives:

At the conclusion of this presentation, the attendee will (
• Understand the limitations and issues of estimating a population attributable fraction in multiple risk factor models (
• Understand the benefits of a sequential partitioning approach (
• Discuss the relevence of this approach in research areas such as the impact of smoking by mothers while pregnant

Keywords: Methodology, Statistics

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