Health Policy and Planning Advance Access originally published online on October 9, 2006
Health Policy and Planning 2006 21(6):459-468; doi:10.1093/heapol/czl029
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Constructing socio-economic status indices: how to use principal components analysis
HIVTools Research Group, Health Policy Unit, Department of Public Health and Policy, London School of Hygiene and Tropical Medicine, London, UK
Correspondence: Seema Vyas, HIVTools Research Group, Health Policy Unit, Department of Public Health and Policy, London School of Hygiene and Tropical Medicine, Keppel Street, London, WC1E 7HT, UK. Tel: +44 (0) 20 7612 7828; Fax: +44 (0) 20 7637 5391; E-mail: seema.vyas{at}lshtm.ac.uk
Theoretically, measures of household wealth can be reflected by income, consumption or expenditure information. However, the collection of accurate income and consumption data requires extensive resources for household surveys. Given the increasingly routine application of principal components analysis (PCA) using asset data in creating socio-economic status (SES) indices, we review how PCA-based indices are constructed, how they can be used, and their validity and limitations. Specifically, issues related to choice of variables, data preparation and problems such as data clustering are addressed. Interpretation of results and methods of classifying households into SES groups are also discussed. PCA has been validated as a method to describe SES differentiation within a population. Issues related to the underlying data will affect PCA and this should be considered when generating and interpreting results.
Key Words: socio-economic status, principal components analysis, cluster analysis, methodology
1A vector that results in a scalar multiple of itself when multiplied by a matrix is known as an eigenvector, and the scalar is its associated eigenvalue. Eigenvectors can only be found for square matrices (though not all), and for an n x n matrix, there are n eigenvectors. For a more detailed description of matrix algebra, and in particular eigenvectors and eigenvalues, see Manly (1994).
2Brazil is a lower-middle-income country with a GNI per capita of US$3090. With a GNI per capita of US$110 Ethiopia is one of the world's poorest countries ([http://www.worldbank.org]). The urban population was 83% in Brazil and 16% in Ethiopia in 2003 (UNDP 2005). We used the 1996 Brazil DHS and 2000 Ethiopia DHS.
3The construction of a number of binary variables from categorical variables is another way to organize the data, although nominally new variables are created. For example, the categorical variable RELIGION, with the values Christian, Muslim, Jewish, Buddhist, converted to binary form would mean the creation of four new variables CHRISTIAN, MUSLIM, JEWISH, BUDDHIST, all of which took on the value of 0 or 1. As the nature of categorical variables is that there is no hierarchical relationship between the variables (which is why they cannot be converted into a meaningful quantitative scale), their conversion into binary variables and inclusion as additional variables does not change the relationship between the variables nor add any additional variation or correlation in the dataset. Rather, having individual variables, PCA can determine which of the particular religion variables can differentiate between households.
4PCA is not invariant to differences in the units of measurement among variables, therefore it is usual to standardize the variables in this instance (Bolch and Huang 1974). Standardization is the process of transforming variables so that the new set of scores has a mean equal to zero and standard deviation equal to one. The correlation matrix is a standardized version of the co-variance matrix.
5Factor score for ownership of a bicycle not stated in Houweling et al. (2003).
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. H. Rah, P. Christian, A. A. Shamim, U. T. Arju, A. B. Labrique, and M. Rashid Pregnancy and Lactation Hinder Growth and Nutritional Status of Adolescent Girls in Rural Bangladesh J. Nutr., August 1, 2008; 138(8): 1505 - 1511. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. L Iannotti, N. Zavaleta, Z. Leon, A. H Shankar, and L. E Caulfield Maternal zinc supplementation and growth in Peruvian infants Am. J. Clinical Nutrition, July 1, 2008; 88(1): 154 - 160. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. H Baqui, A. M Rosecrans, E. K Williams, P. K Agrawal, S. Ahmed, G. L Darmstadt, V. Kumar, U. Kiran, D. Panwar, R. C Ahuja, et al. NGO facilitation of a government community-based maternal and neonatal health programme in rural India: improvements in equity Health Policy Plan., July 1, 2008; 23(4): 234 - 243. [Abstract] [Full Text] [PDF]
|
|