This paper assesses the empirical representativeness of micro-data by comparing two
representative household surveys implemented in Malawi, currently the world’s poorest
country, in 2010. The comparison of descriptive statistics –mainly concerning household
demographics, asset ownership and living conditions– shows considerable similarities in
average values despite statistically identifiable differences. Pronounced differences mostly
occur when the wording and scope of the survey questions differ suggesting that the way
questions are formulated impacts the responses. Multivariate analyses are less
representative due to loss of observations. This results in statistical correlations of asset
indicators with human capital as represented by child health and education that are
susceptible to the (political) objective of the survey; composite indicators of wealth, in turn,
promise more representativeness and, hence, more accurate results. Care is required when
drawing policy recommendations as they depend on purpose and design of the survey, and
are only representative as far as the context of the data collection is reflected.