A closer look at UNICEF’s Foundational Learning Skills (FLS) Module and PAL Network’s ICAN-ICAR household assessments.
This is the second post in our series exploring how we measure learning where children live, not just where they study.
The detective work of learning assessment
Imagine you are trying to understand how well children in a country can read. You could visit schools and test students there, but you would immediately face two problems. First, what about the children who aren’t in school, or those who have never enrolled at all? Second, even among enrolled children, testing within a single grade tells you who in that grade is learning, but not when in a child’s school journey most children actually acquire foundational skills.
Here’s where household-based assessments differ fundamentally from school-based ones — and why this difference shapes everything that follows. Rather than drawing samples from school enrollment lists, they start from national census data. And instead of testing a single grade, they assess an entire age cohort, going door to door to reach all children of school age, whether enrolled, frequently absent or never attended at all. This single design choice changes everything about what the data can tell us.
Since 2015, two publicly available household-based tools have been doing exactly this — UNICEF’s Foundational Learning Skills (FLS) module, administered in over 40 countries, and PAL Network’s ICAN-ICAR assessment, reaching 12 countries. Together they’ve assessed learning across more than 50 countries, reaching children that school-based assessments might not cover.
Sampling: A nationally representative foundation
Representativeness is a function of sampling design and random selection, which both household-based and school-based assessments are designed to achieve. The key difference lies in the sampling frame: unlike school-based evaluations that only reach students in the classroom, household-based assessments draw from a nationally representative pool – one that includes households across the country.
Both UNICEF’s FLS in MICS and ICAN-ICAR are built on nationally representative sampling frameworks, typically derived from population census data maintained by National Statistical Offices. This foundation ensures that households are selected from across the country using a structured probability design. In practice, this means that each household and individual has a known chance of being included, allowing findings to reflect the population covered by the survey.
From this national frame, both tools use a multi-stage approach. First, census enumeration areas (EAs) within geographic domains are randomly selected based on their population size. This is followed by random selection of households within the selected EAs. In MICS, the third stage is the selection of individuals within the households. For the FLS module, one child within the target age range is randomly selected. This supports broad national coverage while keeping the survey manageable, given that MICS collects data across multiple domains. ICAN-ICAR assesses all children within the eligible age range in each household, allowing for analysis within families.
These design choices also have implications for how results are interpreted. Differences in interpretation often relate to the sampling frame and to sample size. The sampling frame determines which population is covered, while sample size affects how precisely results can be estimated, including across regions or population groups. In both household- and school-based assessments, the use of probability-based multi-stage sampling reduces selection bias while producing a representative sample and allowing for inferences about the larger population. Precision is managed through proper stratification, adequate sample size and weighting.
Guardrails against bias
Household-based assessments build in several features to ensure results reflect all children, not just the easiest to reach. The first is callbacks. If a selected child isn’t home during the first visit, interviewers return multiple times if necessary. In MICS, at least three callback attempts are required before a case is considered incomplete. In school-based assessments, a child absent on test day is simply not included, and since absence is often linked to disadvantage, this can quietly skew results.
The second is population transparency. Because household surveys start from census data, followed by a household listing exercise, researchers know exactly who was selected, making it possible to analyse patterns in who didn’t participate and assess whether non-response introduces bias.
The third is consent and assent. Both FLS and ICAN-ICAR require explicit consent from caregivers and assent from the child themselves – a procedural safeguard that also generates data on who declines and why, offering another window into potential non-response patterns.
Complementary Implementation Models: National systems and community networks
The two tools differ in how they are implemented, reflecting their origins. MICS is implemented by National Statistical Offices, positioning it within official data systems. This supports standardization, comparability across countries, and alignment with national planning processes. ICAN-ICAR is implemented through networks of trained citizen volunteers. This approach builds community engagement and allows assessments to be carried out at scale while fostering local ownership of the findings.
Interviewer training: Building consistency at scale
Training is central to how household surveys maintain consistency across large and diverse samples. In MICS, interviewers are typically recruited through National Statistical Offices and undergo several weeks of structured training. Because MICS is a multi-topic survey, training is sequenced to build both technical understanding and practical skills. It often begins with classroom-based instruction using paper questionnaires, ensuring that interviewers understand the content and can continue data collection if digital tools fail. This is followed by supervised field practice in households. Only after this are modules introduced on computer-assisted platforms, along with additional field testing using the digital system.
Across both approaches, the objective is the same: to ensure that assessments are administered consistently, regardless of where they take place or who conducts them.
Fieldwork monitoring: Ensuring data quality
Training is only one part of ensuring data quality. Monitoring during fieldwork plays an equally important role.
In MICS, interviewers work in small teams under field supervisors responsible for operational logistics and quality control. Team supervisors assess security conditions, arrange travel and coordinate with local authorities. They monitor interviewer performance through direct observations, random household revisits to verify information or ensure that protocols have been followed and regular feedback sessions with interviewers. National Statistical Offices often deploy additional supervisory staff at regional or district levels to observe fieldwork and ensure that standards are maintained across locations. Direct field supervision is reinforced by daily synchronization of collected data with the Central Office, enabling real-time generation of field check tables, interviewer performance metrics and data quality dashboards. This allows national survey coordinators to flag issues for review or corrective action, providing the immediate feedback needed to field teams.
In ICAN-ICAR, trained citizen volunteers typically work in pairs under the close supervision of Project Management Teams (PMTs) and District or County Coordinators (DCs/CCs). Monitors conduct intensive field shadowing in at least five households per enumeration area, using a standardized Field Monitoring Checklist to directly assess enumerator performance and ensure strict adherence to standardized administration, sampling and ethical protocols. A critical feature of this process is independent scoring, where supervisors simultaneously record assessment results to calculate Inter-Rater Reliability (IRR), enabling on-the-spot corrections and identifying needs for targeted retraining early in the data collection process. For EAs not visited in person, supervisors maintain oversight through phone monitoring to track progress and provide real-time guidance to teams facing field challenges. This multi-layered approach is further strengthened by desk rechecks of all digital data synced via SurveyCTO and field rechecks, during which supervisors revisit sampled households to independently verify the accuracy and integrity of the collected information. If significant procedural errors or major data discrepancies are identified, monitors have the authority to order a complete resurvey of the community.
These layers of oversight help ensure that data collection remains consistent and credible, even when conducted at scale.
Language and context: Measuring learning fairly
Language is central to ensuring that assessments reflect what children actually know.
Both FLS and ICAN-ICAR invest in adapting their tools to local contexts. The FLS is translated in collaboration with national authorities including the NSO and Ministry of Education while ICAN-ICAR is adapted by country partners in close partnership with subject matter and curriculum experts to reflect the languages of instruction largely used in the country.
These processes go beyond translation. They involve validation steps to ensure that assessment tasks are meaningful and comparable across contexts.
Complementary approaches to understanding learning
Taken together, FLS and ICAN-ICAR illustrate how household-based approaches can generate different but complementary insights into children’s learning. Used alongside school-based assessments, these tools help build a more complete understanding of learning- one that reflects both how education systems function and how children experience learning in their daily lives.
As countries continue to strengthen how learning is measured, the value lies not in choosing one approach over another, but in using these tools together to inform better policy and action.
