Empirical Approach to Developing an Optimal Socioeconomic Status Index for Health Surveillance

Abstract

The association between multi-dimensional deprivation and public health is well established, and many area-based indices have been developed to measure or account for socioeconomic status in health surveillance. The Yost Index, developed in 2001, has been adopted in the US for cancer surveillance and is based on the combination of two heavily weighted (household income, poverty) and five lightly weighted (rent, home value, employment, education and working class) indicator variables. Our objectives were to 1) update indicators and find a more parsimonious version of the Yost Index by examining potential models that included indicators with more balanced weights/influence and reduced redundancy and 2) test the statistical consistency of the factor upon which the Yost Index is based. Despite the usefulness of the Yost Index, a one-factor structure including all seven Yost indicator variables is not statistically reliable and should be replaced with a three-factor model to include the true variability of all seven indicator variables. To find a one-dimensional alternative, we conducted maximum likelihood exploratory factor analysis on a subset of all possible combinations of fourteen indicator variables to find well-fitted one-dimensional factor models and completed confirmatory factor analysis on the resulting models. One indicator combination (poverty, education, employment, public assistance) emerged as the most stable unidimensional model. This model is more robust to extremes in local cost of living conditions, is comprised of ACS variables that rarely require imputation by the end-user and is a more parsimonious solution than the Yost index with a true one-factor structure.