Household Survey Data on Remittances in Sending Countries
Johan A. Mistiaen
International Technical meeting on Measuring RemittancesWashington DC - January 24-25, 2005
Sampling and Questionnaire Design: Options and Uses
World Bank - Development Data Group
OverviewWhy Collect Micro-Data from Remittance Senders?Sampling Frame Design Options Why is Sampling a Critical Issue? Plan A: Build a Representative Sampling Frame Plan B: Some Micro-Data is Better Than None On Sample Size
Questionnaire Design and Implementation A Core Module: Towards Data Consistency Implementation Challenges
Ideas for a Research Agenda
Sampling Design OptionsWhy is Sampling a Key Issue?
A representative sampling frame is the cornerstone of sample-based statistical analysis:
Without it we cannot obtain sample-based statistics or inferences that are representative of the population of interest.
For instance, representative sample data is needed to compute “propensity to remit” estimates.
Sampling frames of the population sub-groups that send remittances are non-existing need to build them
Need to define our target population (domain of analysis)
All persons above 18 years of age that were born in a foreign country.
Unlikely standard frames can be used…
Sampling Design OptionsPlan A: Build a Representative Sampling Frame
Option I: Finding Option I: Finding AllAll Needles in the Haystack Needles in the Haystack
Current Population Registers
Data systems that record selected info on the de jure population in a country; including data that identify residents by street address, age and country of birth.
Construct address referenced listings of all members in the respective target sub-population groups by geographical areas (asap) which
become the “clusters” of our sampling frame.
Sampling Design OptionsPlan A: Build a Representative Sampling Frame
Option I: Finding Option I: Finding AllAll Needles in the Haystack Needles in the HaystackCan apply standard techniques to select a representative (stratified) sample of each sub-group (i.e. by country of birth) with associated sampling weights (the inverse selection probability).Work ongoing to implement this approach in some EU member states.Already in design phase to draw samples of African-born residents in Belgium.Advantages: Representative sample Relatively easy to maintain sampling frame
Sampling Design OptionsPlan A: Build a Representative Sampling Frame
Option II: Finding the key HaystacksOption II: Finding the key HaystacksPopulation Census Data
Typically collect data on “country of birth” (sometimes also include street addresses)
Identify all geographical areas (as small as possible) from the census that contain target sub-population group members; these become “clusters” in our sample frame.
Examples: UK 2001 Population Census US 2000 Population
Census
From Population Census data From Population Census data it is possible to build a “frame” it is possible to build a “frame” of Enumeration Areas/Blocs of Enumeration Areas/Blocs (100?-150? hhs) in the UK that (100?-150? hhs) in the UK that contain people born in contain people born in specificspecific foreign countriesforeign countries
Data on “country of birth”was also collected via the
“long form” of the 2000 UScensus (1 out of 6 hhs)
Sampling Design OptionsPlan A: Build a Representative Sampling FrameOption II: Finding the key HaystacksOption II: Finding the key Haystacks
A Two-Step Sampling Approach
Step 1: Draw sample of clusters (can adjust probability of selection on the proportion of target sub-population).
Step 2: Conduct a “screening” or “re-listing” exercise to identify current incidence of the target population.
Draw sample based on screened clusters
If needed, adjust initial cluster sample ex-post (if step 2 conducted “on-the-go”) either via re-weighting methods or with supplementary sampling.
Sampling Design OptionsPlan A: Build a Representative Sampling Frame
Options I and II: Limitations and Caveats
Frame Errors: All Needles?…“illegal” immigrants… Population registers vs. population census data Pilot attempts to supplement main sampling frame
by “snowball” sampling (i.e. referrals), through relevant organizations, and at key likely contact points (Groenewold and Bilsborrow, 2004).
Population register approach potentially feasible in most EU member states; but few useable population registers elsewhere (Bilsborrow et al., 1997).
Sampling Design OptionsPlan A: Build a Representative Sampling Frame
Options I and II: Limitations and Caveats
“sensitive data”: Government cooperation critical
“updating” of population census based frames… without screening all relevant clusters will need to account for modeling
errors.
Sampling Design OptionsPlan B: Some Micro-Data is Better Than NoneAggregation Point Sampling
Listing of migrant (foreign-born) meeting points Religious venues, community centers, international phone
businesses, employment offices, etc…
Will capture both legal and undocumented immigrantsEx-post determination of respondent selection probabilities Based on “visit frequency” profiles (e.g., what aggregation
points in the sample are visited, how often, when, etc…)
Can yield a (representative) sampleApplied successfully to interview Ghanaian and Egyptian born persons in Italy (Groenewold and Bilsborrow, 2004).
Sampling Design OptionsOn Sample Size
Osili (2004): Sampled 112 Nigerian born residents in the Chicago area to study remittances
Average annual per capita remittances: $6,000Standard deviation: $11,250 95% confidence interval = [$3,750 ; $8,250]
Average annual per capita income: $25,500 Mean Propensity to Remit = 0.23 95% confidence interval = [0.15 ; 0.32]
Increasing sample size to 400 would halve the standard error
Optimal sample size will be a function of the distribution of the variable of interest and the targeted precision
Questionnaire Design and Implementation
A Core Module: Towards Data Consistency Core data collection Consistent across countries and within
countries Modular: stand alone or tag-on to other survey
Implementation Challenges Minimizing Non-Response
Questionnaire design, interviewer selection/training, collaboration with community groups, etc.
Understanding/Correcting for Non-Response
Ideas for a Research AgendaStatistical and econometric analysis to obtain better measures of the “propensity to remit” and its determinants; both household characteristics and market variables (e.g., transaction costs…).Small area estimation of the “propensity to remit” by combining survey and census data.