240 likes | 755 Views
The Wealth Index. MICS3 Data Analysis and Report Writing Workshop. Background. Economic status is known to be strongly correlated with demographic and health behaviour
E N D
The Wealth Index MICS3Data Analysis and Report Writing Workshop
Background Economic status is known to be strongly correlated with demographic and health behaviour However, income and expenditure data are usually not collected in large scale surveys that focus on non-economic issues, such as mortality, child health, and other demographic/social issues
Income and expenditure data Difficult and time-consuming to collect (recall problems, large modules…) Misstatement, particularly of income Seasonality, current versus long-term wealth, methodological constraints, incompatibility
Solution? Proxies such as employment, education, ownership of assets used Assets were sometimes used in producing simple counts, or prices of assets were used as weights Without good indicators on household wealth, analyses usually remain incomplete
Solution? In the late 1990s, a technique was developed to derive information on “long-run wealth” from data already collected in large-scale surveys: assets or possessions of the household… …and called the “Wealth Index” An opportunistic approach to make use of data already available in most household surveys, and to produce an index of wealth which would perform well in explaining differentials
Construction of the WI Use information on assets or household possessions, thought to be indicative of wealth Generate weights (factor scores) for each of the assets through principal components analysis Weights summed by household, household members ranked according to the total score of the household in which they reside Divide the households into quintiles – each containing 20 percent of the household members
Construction of the WI Uses principal components analysis (PCA) to determine the weights (factor scores) We take a large number of assets that may not tell us much individually, but are correlated since they are all related to an underlying factor – in this case, “wealth” The program analyzes the pattern of correlations between the possession of assets and assigns weights to asset variables based on their relation to one another
Number of persons per sleeping room Material of dwelling floor Material of the roof Material of the walls Fuel used for cooking Electricity Radio Television Mobile telephone Non-mobile telephone Refrigerator Watch Bicycle Motorcycle/scooter Animal-drawn cart Car/truck Boat Source of drinking water Type of sanitation facility Ownership of animals Ownership of land Furniture Additional household items MICS3 Assets & Facilities
Selecting ‘assets’ Select those that are thought to reflect material wealth Avoid variables such as nutrition (which is not an asset), or outcome variables, such as education The more indicators are selected/used, the better – but select only theoretically sound variables
Variables Run frequencies of all variables Check for outliers, unexpected values, or large numbers of missing cases – if necessary, regroup or recode Dichotomize all categorical or ordinal variables Use continuous/interval scale variables as they are
Improving the WI Check total variance explained by the first component. Should be greater than 10 percent
Improving the WI • Check the “component score coefficient matrix” (especially if the eigen value of the first component is less than 10 percent) • Assets owned by very few households are likely to have low scores (Do not contribute to the model). Combine such assets with others that might be related conceptually (in terms of wealth)
Improving the WI In this model, persons per sleeping room and calamine/cement fibre roof are negatively correlated with wealth – cement roof, wood floor, parquet/polished floor are positively correlated Combine assets which have the same signs These values are summed over each household to generate the total index value of that household
Uses of the WI The wealth index has become a standard background variable used in household surveys The only “index” constructed by using a statistical technique Poor - nonpoor differences in a variety of health and demographic outcomes – e.g. rich-poor ratios can be calculated to show the extent of differences between socioeconomic groups Can be used to show changes in the extent of disparities
Uses of the WI Always check for denominators of the quintiles in the tabulations – if necessary, dichotomize and use the “poor 40 percent” and “rich 60 percent” Usually, outcome indicators display regular patterns by quintiles. Absence of such regular patterns does not necessarily mean that the calculation of the index is problematic
Issues Does not allow comparisons across countries Urban bias Long-term wealth versus current economic status “Household”, institutional households, populations in special circumstances Meaning of the index values Association between assets/facilities in the index and the dependent variables