Transacting knowledge when there are no schools during the COVID-19 lockdown in Nigeria: the SENSE-transactional radio instruction experience

The study addresses how to ensure continuity of basic literacy learning in Northeast Nigeria where insurgency (Boko Haram) and the COVID-19 pandemic disrupted access to physical schools. Beyond physical access issues, the authors note pedagogical and technological barriers that hinder literacy, numeracy, and life skills development, particularly for disadvantaged learners. In this context, the SENSE-Transactional Radio Instruction (TRI) model was implemented to provide radio-based, interactive, and culturally responsive instruction that could reach learners safely at home. The research question is whether participation in the SENSE-TRI intervention leads to improved literacy outcomes for grade 3 learners relative to comparable learners not exposed to TRI during school closures. The paper hypothesizes that leveraging proven TRI approaches, adapted and blended with SENSE strategies, would yield superior literacy outcomes and contribute evidence for effective learning in crisis-affected settings.

The theoretical rationale is grounded in the USAID-funded Strengthening Education in Northeast States (SENSE) project, which expanded during COVID-19 school closures to include a TRI component (“schooling on air”). TRI uses a 4S framework—sound pedagogy, stories, songs, and synchronized workbooks—aligned with Early Grade Reading best practices (read-aloud, songs, workbook activities) and synchronized with teacher training and community engagement. It aligns with UNICEF’s ACADA (Assessment, Communication Analysis, Design and Action) planning model emphasizing assessment, analysis, and action with monitoring and evaluation. Prior literature indicates that while digital and radio technologies can supplement (not replace) in-person learning, they can reduce social biases and enhance performance when threats are mitigated. Radio’s ubiquity and low cost make it effective in crisis contexts for mobilization and instruction (e.g., Nicaragua IRI math gains; Bolivia early childhood; crisis education initiatives). However, interaction is often a limitation in radio programs, and blended approaches are often advantageous. Adaptation depends on infrastructure (electricity, internet, devices), income, and readiness of stakeholders. The TRI approach has shown effectiveness in various contexts, but cultural localization (e.g., use of Hausa/Fulfulde) is crucial. The review underscores the importance of integrating media literacy, teacher empowerment for digital/blended learning, and context-appropriate pedagogy to optimize outcomes in crisis and normal conditions.

Design: Quasi-experimental study comparing a treatment group exposed to SENSE-TRI with a comparison group not exposed, during COVID-19 school closures when all learners were formally enrolled but learning from home. Setting and participants: Two insurgency-affected but relatively accessible states in Northeast Nigeria (Adamawa and Gombe), which also have high out-of-school rates. Total N=400 grade 3 learners (200 per state), evenly split between treatment and comparison groups. Sampling: From schools participating in SENSE, randomly selected schools formed the treatment group. Comparable nearby schools not exposed to TRI formed the comparison group. Within communities near intervention schools, a quota-random-walk household sampling selected grade 3 learners for inclusion. The proximity ensured comparability and minimized confounding. Sample size was set using 95% CI and 5% margin of error (minimum 381; rounded to 400). Intervention: SENSE-TRI broadcast 13 weeks of Mu Karanta and RANA radio lessons (Aug–Oct 2020), with episodes twice weekly, incorporating stories, songs, and synchronized workbooks designed for EGR. Instruments: An abridged Early Grade Reading Assessment (EGRA) administered mid-December 2020 comprised five untimed subtasks: (1) letter-sound identification (29 items), (2) syllable-sound identification (21 items), (3) familiar word reading (18 items), (4) invented word reading (12 items), (5) reading comprehension (4 questions on a connected text). Scores were the number correct. Data collection: Trained enumerators administered EGRA using Android tablets with Tangerine, following a random-walk approach across areas. Potential enumerator variability (psychosocial, sociocultural) is acknowledged. All data were collected post-broadcasts. Analysis: Group differences were first examined with a nonparametric equality-of-medians test due to mass at score extremes. Tobit regression models then estimated adjusted differences, accounting for lower and upper bounds of scores. Models included controls for student characteristics (state, gender, age, language at home, living with both parents, household size, hunger in past week, parental literacy), interactions (e.g., treatment×Adamawa, treatment×age, treatment×Hausa), and Local Government Area (LGA) fixed effects plus enumerator variability; robust standard errors; maximum likelihood estimation. Alternative specifications replaced treatment status with self-reported weekly radio listening frequency to assess dose–response effects.

- Descriptive and unadjusted differences (Table 2): Treatment learners vastly outperformed comparison learners across all EGRA subtasks. - Letter sounds: mean 24.01 vs 8.76; median 26 vs 4; % zero 3.0% vs 26.0%; p<0.001. - Syllable sounds: mean 15.10 vs 4.52; median 17 vs 0; % zero 7.5% vs 51.0%; p<0.001. - Familiar words: mean 12.66 vs 3.61; median 15 vs 0; % zero 8.5% vs 56.0%; p<0.001. - Invented words: mean 7.97 vs 2.04; median 9 vs 0; % zero 12.0% vs 72.5%; p<0.001. - Reading comprehension (max=4): mean 3.76 vs 3.02; median 4 vs 3; % zero 0.0% vs 3.5%; p≈0.0001. - Adjusted Tobit regression (Table 3): Treatment effects remained large and statistically significant after controls and LGA dummies. - Coefficients (SE): Letter sounds 17.944 (7.078), p<0.01; Syllable sounds 18.123 (6.336), p<0.01; Familiar words 16.695 (5.700), p<0.01; Invented words 17.311 (5.383), p<0.01; Reading comprehension 4.041 (1.440), p<0.01. - Speaking Hausa at home associated with higher scores; recent hunger associated with lower scores; state (Adamawa) negatively associated for some subtasks. - Interaction treatment×Hausa negative and significant, indicating the Hausa-home-language advantage is smaller or absent among treatment learners; in reading comprehension, non-Hausa speakers in treatment outperform Hausa speakers. - Pseudo-R2 ranged ~0.117–0.165; link test indicated no major misspecification except possibly letter sounds model. - Robustness with exposure (Table 4): Weekly radio listening correlated with higher scores; marginal predictions (Fig. 5) show weekly listeners outperform non-weekly listeners, with particularly noticeable effects on invented word reading. - Overall: Treatment group performance also surpassed initial SENSE baseline reading proficiency aggregates (AUN 2020), suggesting TRI effectiveness in crisis and non-crisis contexts.

The findings indicate that participation in SENSE-TRI during school closures is associated with substantially better foundational literacy skills across letter and syllable identification, familiar and invented word decoding, and reading comprehension. These advantages persist after adjusting for key demographics and contextual factors, and they are reinforced by dose–response evidence from self-reported weekly listening. Language context matters: while speaking Hausa at home generally correlates with higher literacy scores, TRI appears to attenuate this gap and even reverse it for reading comprehension, possibly reflecting TRI’s accessibility to non-Hausa speakers. Socioeconomic stressors (e.g., hunger) negatively relate to performance, underscoring the role of household conditions in mediating learning outcomes. While results support TRI’s potential to transact knowledge during crises, causal attribution is limited by the quasi-experimental design and absence of baseline measures. The TRI model’s culturally localized implementation, integration with workbooks and community engagement, and alignment with ACADA planning likely contributed to engagement and outcomes. The study adds evidence that in emergencies, well-designed radio instruction can provide safe, impactful learning opportunities and may complement blended approaches post-reopening.

Mid-term evaluation shows SENSE-TRI achieved statistically significant improvements across all five EGRA subtasks for grade 3 learners during COVID-19 lockdown in insurgency-affected states. The program delivered safe, conflict-sensitive instruction with minimal resources, and outcomes exceeded prior baseline aggregates, indicating progress toward program goals. While technology-enhanced approaches should complement, not supplant, in-person teaching, TRI can be a key component of blended learning. Going forward, strengthening factors that amplified impact—parental literacy and involvement, consistent access to radio, supportive home environments—could enhance outcomes. The authors propose integrating TRI within the ACADA framework (ACADA-TRI) for learning continuity in unusual periods. Future research should implement baseline and longitudinal designs, expand beyond grade 3 and basic literacy to broader subject areas and grades, and assess TRI’s effectiveness absent pandemic/insurgency conditions, including cross-national studies.

- No baseline assessment for learners; outcomes reflect levels at a single time point rather than gains, limiting causal inference. - Non-random assignment to treatment and comparison groups (quasi-experimental design) leaves potential for unobserved confounding. - Focus restricted to grade 3 learners and basic literacy subtasks; results may not generalize to other grades or competencies. - Conducted in two states (Adamawa, Gombe) during a specific crisis period; external validity to other regions or conditions may be limited. - Potential enumerator variability and household/contextual factors (e.g., hunger, access to radio) influence outcomes. - Some variables (e.g., books at home) had missing data and were excluded from final models; model fit (Pseudo-R2 ~0.12–0.17) typical for education but indicates unexplained variance.