• DOI:10.1093/OXFORDJOURNALS.AJE.A009552
• Corpus ID: 11472957

Invited commentary: Factor analysis and the search for objectivity.

@article{Martinez1998InvitedCF,
  title={Invited commentary: Factor analysis and the search for objectivity.},
  author={M. Elena Martinez and James Marshall and Lee Sechrest},
  journal={American journal of epidemiology},
  year={1998},
  volume={148 1},
  pages={
          17-9
        }
}

• M. E. Martinez, J. Marshall, L. Sechrest
• Published 1 July 1998
• Medicine
• American journal of epidemiology

A key concern in nutritional epidemiology is the complex nature of the exposure assessed by means of dietary questionnaires. Unlike other exposures, diet is so complex that special knowledge is required to ensure that the correct exposure measure is used. A clear definition of the food or nutrient of interest and knowledge of how it may affect the disease or outcome of interest is imperative. Food consists of nutrients as well as other substances, such as additives, naturally occurring compounds, and unknown components, that may affect disease risk. Therefore, it is often not adequate to represent food intake by nutrient intake. The approach of Slattery et al. (1) is innovative: Using data from a detailed diet history questionnaire administered to participants in a large case-control study, the authors employed exploratory factor analysis to typify food patterns that might be associated with colon cancer risk. They then evaluated the factors they identified as risk predictors in conventional logistic regression analysis. Factor analysis is one of the most widely used quantitative techniques in the social sciences. It was developed primarily for psychometric measurement; its origins and early development were largely in the field of abilities testing. Factor analysis has been employed virtually since its inception for two somewhat distinct purposes: data reduction and theory building (2, 3). It does this by characterizing the covariance among many variables in terms of a few underlying but unobservable quantities called "factors." The factor model is driven by the idea that correlated variables can be grouped or aggregated—i.e., that all correlated variables belong together, and they should be recognized as distinct from groups of variables with which they are not correlated. Factor analysis has great intuitive appeal in nutritional epidemiology: It offers a means of factoring intakes of a variety of foods as