What Language Are We Really Measuring?

What Language Are We Really Measuring?  

How household assessment data are changing our understanding of language and learning

Language has long been recognised as one of the greatest challenges in assessing foundational learning across Africa.

Fewer than one in five African children are taught in their home language during the early grades of primary school, compared with more than 70% of children in regions such as South Asia and Latin America. That single comparison says a great deal about the scale of the challenge before we even begin discussing how learning is assessed.

The continent's extraordinary linguistic diversity is one of its greatest cultural strengths, but it also presents complex questions for education systems. Should children be taught in their home language? When should they transition to a national or international language? Which language should be used for learning assessments?

These questions are often framed as matters of policy or principle. They are undoubtedly important, but they can also obscure a more fundamental issue.

Before asking whether an assessment is administered in the 'right' language, we should first ask a simpler question: What language environment are children actually learning in?

One of the most important contributions of household assessment data is that they allow us to answer this question empirically rather than relying on assumptions.

Language is part of a child's learning environment

Learning assessments often treat language as a characteristic of the assessment itself. Reports describe the language or languages in which the assessment was administered and may discuss the rationale for those choices.

Children, however, experience language long before they sit an assessment.

The language spoken at home, the language used by teachers, the language of textbooks and the language of assessment all form part of a child's broader learning environment. These languages may coincide, but they often do not.

The scale of this mismatch is visible in the data. Colonial languages remain the dominant language of instruction in the early grades in most of the countries we examined - reaching 100% in Tunisia and in São Tomé and Príncipe. Malawi and Madagascar stand out as clear exceptions, with only around 5% of children taught in a colonial language during the early grades. Most Anglophone countries fall somewhere in between, with English used alongside - or instead of - national languages to varying degrees.

A child may speak one language at home, receive instruction in another, and be assessed in a third. Alternatively, the language of instruction and assessment may align closely even though neither is spoken at home. In some education systems, children are assessed in the same language they encounter both at home and at school.

These are very different educational experiences. Understanding them is essential if learning assessments are to be interpreted appropriately.

From assumptions to evidence

Historically, researchers have had limited opportunities to investigate these language pathways because most assessment datasets contained relatively little information about children's language environments.

Household-based assessments are changing this.

The Multiple Indicator Cluster Surveys (MICS), for example, ask children which language they speak most often at home. The PAL Network's International Common Assessment of Numeracy and Reading (ICAN-ICAR) asks adult respondents how frequently the assessment language is spoken in the household. When these data are combined with information on the language of instruction and the language of assessment, they create entirely new opportunities for understanding how language shapes learning.

Instead of assuming that children are assessed in familiar or unfamiliar languages, researchers can begin to measure the extent of alignment between home language, language of instruction and assessment language.

This represents a significant shift in how language can be studied.

Africa's language environments are remarkably diverse

Our report illustrates just how varied these language environments are.

Some assessment programmes already reflect considerable linguistic diversity. South Africa's PIRLS, for example, is administered in all eleven official languages. MICS includes multiple assessment languages in countries such as Eswatini, Lesotho, Malawi, Madagascar, Nigeria and Zimbabwe. The PAL Network's ICAN-ICAR assessments also employ multiple assessment languages where appropriate, while Tanzania administers the assessment entirely in Kiswahili.

Other programmes have increasingly expanded the use of African languages over time. The 2024 round of PASEC, for example, introduced multiple assessment languages in Mali, Niger and Senegal, extending approaches that had previously been implemented only in selected countries.

These developments demonstrate that multilingual assessment is both technically feasible and increasingly common.

More importantly, they highlight the importance of documenting language choices carefully so that researchers can understand the contexts in which assessment results are generated.

Alignment matters

One of the central figures in our report examines the alignment between children's home language, language of instruction and assessment language.

The results illustrate that language mismatch is not a single phenomenon.

Madagascar, Eswatini and Tunisia sit at one extreme, where all children are assessed in a language that is both their home language and their language of instruction — full alignment. Zambia sits at the opposite extreme: 72% of children are assessed in a language that is neither their home language nor their language of instruction. Guinea-Bissau and the Democratic Republic of the Congo fall in between, with roughly half of children facing a similar mismatch.

These distinctions matter because they influence how assessment results should be interpreted.

The issue is therefore not simply whether an assessment is administered in an African language or a colonial language. Rather, it is whether the language of assessment corresponds to the language in which children have had opportunities to develop literacy.
This is a more nuanced question, and one that publicly available microdata now allow researchers to investigate.

This pattern echoes findings discussed elsewhere in this series. In Uganda, children perform comparatively well on letter-sound recognition but score poorly on listening comprehension — a pattern more consistent with limited familiarity with the language of assessment than with weak underlying reading skills.

Language also shapes who is represented

Language influences not only performance but also participation.

In several MICS countries, children do not attempt the reading assessment if the available assessment languages do not correspond to either their home language or their language of instruction. In other cases, children decline to attempt the reading passage even though they complete the mathematics assessment.

As a result, the realised reading sample differs systematically from the realised mathematics sample.

This has an important implication that is easily overlooked. Researchers should be cautious when comparing reading and mathematics outcomes within a country because the two assessments may represent different populations of children. Likewise, comparisons of reading outcomes across countries should take account of differences in language eligibility and realised response rates.

Household assessment data make these issues visible. Without access to the underlying microdata, they would remain largely hidden.

Better language evidence for better policy

Language policy will continue to be debated, and rightly so. Decisions about language of instruction involve educational, cultural and political considerations that extend far beyond learning assessment.

What household assessment data can contribute is something different.

They provide the evidence needed to understand the language environments in which children are actually learning and being assessed. They allow researchers to move beyond broad generalisations and examine how language pathways differ between countries, between regions and even between groups of children within the same education system.

This evidence does not prescribe a single language policy. It does, however, improve the quality of the information available to those responsible for designing assessments and education systems.

From language of assessment to language environments

One of the broader themes running throughout our report is that publicly available microdata encourage researchers to ask better questions. The same is true for language.

Rather than asking simply 'What language was the assessment administered in?', we can ask 'What language environment did children experience before they were assessed?'

That shift in perspective changes the kinds of evidence we can generate and the conclusions we draw.

Language is likely to remain one of the most complex issues in foundational learning assessment. There are no simple solutions, nor should we expect there to be. But there can be better evidence.

Household assessment microdata provide an opportunity to move beyond assumptions and towards a more systematic understanding of how children's language environments shape both learning and the measurement of learning.

In doing so, they improve not only our interpretation of assessment results but also the assessments themselves.

AFLEARN Insight #1       AFLEARN Insight #2       AFLEARN Insight #3      AFLEARN Insight #4     AFLEARN Insight #5    AFLEARN Insight #6    Read the report