Chapter 1: The Design and Content of Household Surveys
This chapter lays the groundwork for analyzing household survey data by explaining how
surveys are structured, how data are collected, and the implications for statistical analysis and
policy research. It covers everything from sampling methods to data reliability.
1.1 Survey Design
Simplest household survey- A comprehensive, up-to-date list of all units (households or
individuals) in the target population. Typically based on recent census data or household
listings. Survey design is the foundation of any household survey. It determines how data are
collected, who is included, and how reliable and representative the results will be. A
well-designed survey ensures that the information gathered is accurate, cost-effective, and
suitable for policy and research needs.
Survey Frames and Coverage
● Definition: A sampling frame is a list or database from which survey participants are
selected.
● Key Point: A good frame should cover the entire population of interest. If not, it leads
to coverage bias. Undercoverage occurs when some groups (e.g., nomads, urban
slum dwellers) are left out, which can bias results.Some groups may be
intentionally left out (e.g., military barracks, institutionalized populations, remote or
insecure areas, homeless). These exclusions must be documented, as they affect
representativeness. Overcoverage happens when ineligible units (e.g., vacant
homes) are included.
● Challenges: In developing countries, comprehensive and updated sampling frames
are often missing, especially for rural or mobile populations.
Strata and Clusters
● Stratification divides the population into subgroups (strata) such as regions,
urban/rural, or income levels.
Each stratum is sampled separately, ensuring representation and improving
precision.
○ For example, urban and rural areas may have different living standards,
so stratifying ensures both are well represented.
○ Stratification reduces the variance of estimates if differences within
strata are less than between strata.
● Clustering involves grouping units (e.g., households within villages or city
blocks) and sampling clusters first, then sampling units within clusters.
Grouping households into clusters (e.g., villages, blocks) for logistical efficiency.
○ This approach reduces travel and administrative costs.
○ However, people within clusters tend to be similar (intra-cluster
correlation), which increases sampling error compared to a simple
random sample.
○ The design effect quantifies the increase in variance due to clustering.
Two stage design- Often, surveys select clusters (primary sampling units, PSUs) first, then
households (secondary sampling units) within each cluster.
● Example: A survey might first select 100 villages (clusters), then 20 households from
each village.
Unequal Selection Probabilities, Weights, and Inflation Factors
● In practice, some groups may be sampled at higher rates (oversampling) for
analytical reasons.
○ For example, oversampling poor households allows for more precise
poverty estimates.
● Weights are assigned to each observation, typically as the inverse of the
probability of selection. They correct for unequal probabilities and ensure
results represent the population.
○ For example, if a household had a 1 in 100 chance of selection, its
weight is 100.
● Inflation factors are similar, scaling up the sample to match the total
population.
○ Proper use of weights and inflation factors is essential for unbiased
estimates and correct standard errors.
Sample Design in Theory and Practice
● Ideal sampling should balance cost-efficiency and representativeness.
● Practical considerations often force trade-offs between theory (e.g., random sampling)
and real-world constraints (e.g., accessibility).
● Theoretical sample design involves specifying how to select the sample to
achieve desired statistical properties (e.g., minimum variance,
representativeness).
● Practical realities often require compromises:
○ Nonresponse (households refusing or unavailable)
○ Inaccessible areas (due to conflict, disaster, or geography)
○ Budget and time constraints
● Field teams may substitute households or clusters, which can introduce bias if
not done systematically and transparently.
Panel Data
● Panel (longitudinal) surveys repeatedly interview the same households over
time.
○ This allows analysis of changes in welfare, the impact of shocks, and
causal inference.
○ Panel surveys track the same households or individuals over time, offering
insights into dynamics like income mobility or poverty transitions.
● Advantages: Track dynamics such as income mobility, the effect of
interventions, or life-cycle changes.
● Challenges: Attrition (households dropping out), increased cost, and
complexity in data management and analysis, changes in household composition.
1.2 The Content and Quality of Survey Data
Individuals and Households
● Varies by survey but generally refers to people who live together and share
meals or resources. The definition affects who is included in the data and
how results are interpreted.
● Household is usually defined as people living together and sharing resources
(food, shelter, income).
○ The definition affects who is included—e.g., boarders, extended family,
domestic staff.
● Individual-level data (e.g., education, health, employment) enables analysis of
intra-household differences (such as gender or age disparities).
● The unit of analysis (household or individual) must be clear for correct
interpretation.
Reporting Periods
● Longer periods might lead to recall bias, while shorter periods may miss seasonality
or irregular events.
● Reference (reporting) period is the time span respondents are asked to recall
(e.g., last week, last month, last year).
○ Short periods (e.g., last 7 days) reduce recall bias but may miss
infrequent events.
○ Long periods (e.g., last year) risk inaccurate recall, especially for
irregular income or expenditures.
● Seasonality: Consumption and income can vary by season (e.g., harvests,
festivals), so the timing of the survey matters for interpretation.
Measuring Consumption
● Why consumption? In developing countries, consumption is often more reliable
than income due to informal economies and income volatility.
● Components: Food (purchased, home-produced, received as gifts), non-food
goods, services, and sometimes durables.
○ Home-produced goods must be valued at market prices, which may
require imputation.
● Data collection methods:
○ Recall: Respondents report what they consumed over a past period.
○ Diary: Households record consumption as it happens, usually over 7–14
days.
○ Each method has trade-offs in accuracy, respondent burden, and cost.
● Imputation: For missing prices or quantities, imputation uses average prices
from local markets or similar households.
● Must distinguish between:
○ Food vs. non-food consumption
○ Cash vs. in-kind consumption
○ Home production vs. market purchases
Measuring Income
● Harder to measure accurately due to:
○ Irregular income flows (e.g., agriculture, informal work)
○ Reluctance to disclose earnings
○ Problems with valuing self-employment or in-kind income
● Income is difficult to measure due to multiple sources (wages,
self-employment, remittances), irregularity, and reluctance to disclose.
● Comparison with Consumption: Income data are often underreported or
misreported; thus, consumption is preferred for welfare analysis in many
contexts.
1.3 The Living Standards Surveys (LSMS)
Design Features of LSMS Surveys
● Multi-purpose: Combines data on education, health, consumption, labor, and more.
● Standardized structure to allow comparability across countries and time.
● Emphasis on both monetary and non-monetary indicators of well-being.
● Multi-purpose, modular design: LSMS surveys include modules on income,
consumption, health, education, labor, agriculture, etc.
○ Modular design allows countries to adapt the survey to local needs.
● High standards: Emphasis on careful questionnaire design, interviewer training,
and data quality control.
● Flexibility: LSMS surveys can be repeated over time (cross-sections or panels)
to track trends and dynamics.
What Have We Learned?
● Methodological insights: LSMS experience has led to improved sampling,
questionnaire design, and fieldwork practices.
● Policy impact: LSMS data have shaped poverty measurement, targeting, and
evaluation of development interventions, design social protection programs, and
evaluate the impact of policies like cash transfers.
● Lessons: Importance of pre-testing, interviewer training, and adapting to local
context for high-quality data.
● Example: The LSMS in Vietnam included modules on health, education, and
agriculture, enabling analysis of how school fees affect enrollment and how health
shocks affect poverty.
● Challenges: Ensuring up-to-date sampling frames, minimizing non-response, and
maintaining data quality in difficult field conditions.
