Skip to content
Back to search results

Asset Index Construction

Introduced by
Deon Filmer & Lant Pritchett
Year
2001
Type
Composite socioeconomic-status proxy index
Family
Wealth and asset measurement
Font size

What does it do?

Asset index construction builds a proxy for household wealth or socioeconomic status from observable possessions — durable goods, housing quality, and access to utilities — when reliable income or consumption data are unavailable.

The dominant approach, popularized by Deon Filmer and Lant Pritchett in 2001, applies principal component analysis (PCA) to a set of asset variables and uses the first principal component as a set of weights, producing a single wealth score for each household. The method underlies the wealth quintiles reported in Demographic and Health Surveys and many other household surveys across low- and middle-income countries.

The intuition behind the method

Measuring income or consumption well is expensive and error-prone, especially where much economic activity is informal or in kind. But surveys can cheaply record what a household owns and how it lives: whether it has electricity, a refrigerator, a motorcycle, a finished floor, piped water, a flush toilet. Households that are better off tend to own more of these and better versions of them, so the assets move together. PCA finds the single underlying axis along which the assets co-vary the most and reads that axis as 'wealth'. Each household's position on the axis — a weighted sum of its standardized assets, with weights learned from the data rather than assumed — becomes its wealth score, and households are then ranked and grouped, usually into quintiles.

History

Filmer and Pritchett's 2001 Demography paper showed that a PCA-based asset index could substitute for missing expenditure data, validating it against consumption in datasets where both were available and demonstrating its power to explain educational enrollment in India. The DHS Program subsequently adopted the asset index as the basis of its standard 'wealth quintiles', making it the most widely used socioeconomic-status proxy in international health and demographic research. Vyas and Kumaranayake's 2006 Health Policy and Planning guide gave practical advice on variable selection, coding, and interpretation. Later refinements — separate urban and rural indices combined onto a common scale, multiple correspondence analysis for categorical data, and inter-survey comparability methods — addressed the original method's known distortions.

When to use it

Use an asset index when you need a household wealth or socioeconomic-status proxy but lack trustworthy income or consumption data, which is the common situation in DHS, MICS, and many census or rapid-survey settings. It is ideal for ranking households relatively, defining wealth quintiles, and studying socioeconomic gradients in health, education, and demographic outcomes. It is not a substitute for a poverty line or a measure of consumption poverty — the score is ordinal and relative, not absolute — and it should be avoided when assets carry very different meaning across the populations being compared (for example, pooling urban and rural households on one index without care), where separate or adjusted indices are preferable.

Limitations and considerations

  • The index is ordinal and relative — it ranks households but has no monetary scale, so it cannot identify who falls below a poverty line.
  • It tends to capture an urban/modern-assets axis and can cluster rural households together (truncation and clumping), distorting comparisons across urban and rural areas.
  • Results depend heavily on which assets are included and how they are coded; including weakly correlated or context-specific assets changes the weights and the ranking.
  • PCA assumes the first component represents wealth, but assets reflect lifestyle, location, and household size too, so the axis may be contaminated by non-wealth variation.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Produces a household wealth proxy from cheap, reliably reported asset data when income or consumption data are missing or unreliable.
  • Weights are learned from the data by PCA rather than assumed, removing the need for arbitrary analyst-chosen weights.
  • Asset stocks are stable and less subject to seasonality and recall error than income or consumption flows, giving a smooth measure of long-run living standards.
  • Yields a single continuous score that ranks households and slots cleanly into quintiles for analyzing socioeconomic gradients in outcomes.

Applications

  • Construction of the wealth quintiles used throughout the Demographic and Health Surveys to disaggregate health and demographic indicators by socioeconomic status.
  • Analysis of socioeconomic gradients and inequities in child mortality, immunization, schooling, and health service use.
  • Targeting of programs and benefit-incidence analysis where consumption data are unavailable.
  • Equity monitoring in global health, ranking households to measure how outcomes vary from poorest to richest groups.
  • Proxy-means-test design and validation, where asset indices inform poverty-targeting scorecards.

How it works

How it works

1

The analyst assembles a set of asset, housing, and utility variables — ownership of durables (radio, television, refrigerator, bicycle, car), dwelling characteristics (floor, wall, and roof material, number of rooms), and access to services (electricity, water source, sanitation). Binary indicators are coded 0/1 and categorical housing variables are expanded into dummies. Variables with almost no variation, or that mean different things in urban versus rural settings, are screened out, since PCA captures only variables that co-vary across households.

2

Principal component analysis is run on the (standardized) asset variables. PCA finds the linear combination of variables with maximum variance: the first principal component is the eigenvector f_1 of the covariance (or correlation) matrix associated with the largest eigenvalue. Its loadings indicate how strongly each asset contributes to the underlying axis of common variation, which is interpreted as the wealth or socioeconomic-status dimension. Only the first component is retained.

Formula
\Sigma\, f_1 = \lambda_1 f_1, \qquad \lambda_1 = \max_{\|f\|=1} f^{\top}\Sigma f
3

Each household's wealth score is the first component evaluated at its own asset profile: A_i is the sum over assets of the loading f_k times the standardized value (a_ik minus the asset's mean, divided by its standard deviation). Standardizing puts all assets on a common scale so that the score is not dominated by high-variance variables, and the resulting score is a continuous, unitless index centered near zero, higher for asset-rich households.

Formula
A_i = \sum_{k=1}^{K} f_k\, \frac{a_{ik} - \bar{a}_k}{s_k}
4

Households are ranked by their wealth score and typically cut into quintiles (or other groups) to define wealth categories from poorest to richest. These categories are used to analyze how outcomes — child mortality, schooling, health service use — vary with socioeconomic status. The score is ordinal: it ranks households reliably but its absolute level has no monetary interpretation, so analyses use relative position rather than the raw value.

Common pitfalls

  • Interpreting the wealth score or quintile cutoffs as absolute living standards or comparing scores across surveys as if they were on the same scale.
  • Failing to standardize the asset variables, so high-variance assets dominate the first component and distort the weights.
  • Pooling urban and rural households on a single index when assets mean different things in each, masking rural differentiation; separate or area-adjusted indices may be needed.
  • Including assets with almost no variation or strong urban bias, which loads the index toward the urban/poor split rather than the full wealth gradient.

Frequently asked questions

Why use PCA instead of just counting assets or using income?

Counting assets treats every item as equally informative, but a television and a piece of land do not signal wealth equally; PCA learns data-driven weights so that assets contributing more to the common variation count more. Income or consumption would be preferable in principle, but in many surveys they are absent, unreliable, or highly seasonal. The asset index sidesteps that problem by using stocks that are easy to report and stable over time, capturing long-run living standards rather than a noisy snapshot of a flow.

Why is the asset index only ordinal and not an absolute measure of wealth?

The PCA score is a position on a relative axis derived from a particular sample of households and a particular set of assets; it has no monetary unit and its mean is roughly zero by construction. It tells you that one household is wealthier than another within the sample, which supports ranking and quintile analysis, but it cannot say whether a household is poor in an absolute sense or compare levels across surveys built from different asset lists. For absolute poverty you need a consumption measure and a poverty line.

What are 'truncation' and 'clumping' in asset indices?

Clumping occurs when many households share the same or very similar asset bundles (common in poor rural areas), so their scores pile up in a narrow band and the index fails to distinguish among them. Truncation occurs when the index spreads richer households over a wide range while compressing the poor, so the distribution is uneven. Both arise because the index is dominated by assets that the better-off own and the poor uniformly lack. Using assets with variation among the poor, or constructing separate rural indices, mitigates the problem.

Related methods

Sources

  1. Filmer, D., & Pritchett, L. H. (2001). Estimating Wealth Effects without Expenditure Data—or Tears: An Application to Educational Enrollments in States of India. Demography, 38(1), 115-132.

    Open DOI reference
    Copied.
  2. Vyas, S., & Kumaranayake, L. (2006). Constructing socio-economic status indices: how to use principal components analysis. Health Policy and Planning, 21(6), 459-468.

    Open DOI reference
    Copied.