Skip to content
Avanti Docs

Strand B: video, TV and recorded lessons vs a live teacher vs self-study in low- and middle-income countries

Searched 2026-11-01. The archived copies are in gs://avantifellows-research-sources/ai-coach/medium/ (every PDF starts with %PDF; the three .html files are PMC full-text pages). The extracted text is in the sibling strandB_txt/ folder. Every number below was read in the archived copy unless a row says otherwise. Where I worked out a 95% CI myself, it is the estimate ± 1.96 × the reported SE and is marked “computed”.

1. Synthesis: what we know, and how sure we can be

Section titled “1. Synthesis: what we know, and how sure we can be”

  • Almost no study compares video and a live teacher head-to-head on the same content. Nearly every study compares adding video or broadcast lessons against business-as-usual, and business-as-usual is often weak teaching or no school at all. I found only three designs that come close to a direct substitution test, and none is at Grade 11-12. Büchel et al. (El Salvador) compares software with extra teacher-led classes. Lichand et al. (São Paulo) compares remote learning, mainly TV, with in-person classes. Telesecundaria (Mexico) compares TV-based schools with conventional schools. Confidence that this gap is real: high. I found no LMIC study, and no Indian study, of recorded vs live coaching for JEE/NEET-type exam preparation.

  • Video that supports a teacher in the classroom has a mixed record, and the largest, best-designed trial is negative.

    • Beg et al. (Pakistan, Grade 8, 59 schools) found +0.30 SD on the project test and +0.27 SD on the government’s standardised PEC exam, at about $9 per student.
    • de Barros (Haryana; Avanti’s own Sankalp programme; 240 schools; n = 24,584; pre-registered) found −0.15 SD in maths and about zero in science after roughly 10 months. Observed instructional quality also fell by 0.46 SD.
    • Seo (Tanzania, Grade 11) used videos that worked through past exam papers. On the national O-level exam the effect was about zero, and the paper rules out anything above 0.13 SD.
    • The idea that “engaging the teacher makes video work” therefore rests mainly on one small trial. The largest replication points the other way. Confidence that teacher-supporting video reliably helps: low.

  • Broadcast lessons that replace the subject teacher, with a local adult facilitating, look positive where the alternative was weak teaching or no school.

    • Ghana: live interactive satellite lessons raised primary maths by +0.23 SD (147 schools, RCT). The gain held a year later. Reading fluency did not improve.
    • China: satellite and DVD lessons by top teachers raised maths by +0.18 SD and Chinese by +0.23 SD, measured 7-10 years later (difference-in-differences, not randomised).
    • Mexico: Telesecundaria raised schooling and earnings (+12.5-13.9% per extra year of schooling). The peer-reviewed test-score paper (Borghesan & Vasey 2024) was not opened.
    • Karnataka: +0.1-0.4 SD, but I saw this only second-hand, through the World Bank review.
    • These are mostly primary or lower-secondary students in rural areas with a facilitator in the room. The comparison is “a strong teacher on a screen” against “a weak teacher in person”, not video against a good live teacher. Confidence: moderate for those settings; low for transfer to Avanti’s Grade 11-12 students.

  • Video or content that bypasses the classroom (self-study on a device, or remote/TV-only learning) is the clearest losing case.

    • Beg et al.’s tablet arm used the same expert content and lowered scores by 0.43 SD.
    • In São Paulo, remote learning cost secondary students 0.32 SD on the state standardised test, measured against in-person classes. Remote learning there was mainly broadcast TV. That is about 0.009 SD lost per week of closure.
    • The Escueta et al. review covers mostly high-income studies. In it, fully online courses trail face-to-face, while blended courses do not.
    • Confidence: moderate to high that unstructured self-study from recorded content does worse than a live classroom for typical low-income students.

  • What helps is adaptive practice plus time on task, not watching.

    • Mindspark (Delhi, after school) raised maths by 0.36 SD and Hindi by 0.22 SD in 4.5 months.
    • At scale in Rajasthan, Mindspark ran inside school hours and replaced 25-50% of class time. It raised maths by 0.22 SD and Hindi by 0.20 SD after 18 months.
    • In El Salvador, offline Khan Academy (CAL) classes beat equal extra time with teachers: 0.21-0.24 SD against 0.15 SD. Teachers added nothing to the software.
    • Ma et al. (rural China) found that the computer itself adds about nothing beyond the same practice on paper.
    • Passive video (Seo; Sankalp) shows no such gains. Confidence: moderate.

  • Avanti’s own two RCTs both show no gain on independent, grade-level exams.

    • Chennai (Chiplunkar, Dhar & Nagesh): the after-school programme combined 15 five-minute videos a week with facilitators and peer learning. Basic ASER skills rose sharply (maths 0.86 SD, reading 0.52 SD). The Class X board exam total, however, was −0.28 SD (p = 0.064; not significant at 5%; maths −0.01, science −0.21, both not significant).
    • Haryana Sankalp (de Barros): −0.15 SD maths, as above.
    • These are the most decision-relevant studies in this strand, and neither supports “video improves exam performance”.

  • The kind of outcome test matters, and the evidence leans negative on independent exams.

    • Positive on a standardised exam: Beg’s PEC (+0.27).
    • Zero or negative on a standardised exam: Seo’s O-level (~0), Chiplunkar’s Class X (−0.28, not significant), Lichand’s state test (−0.32).
    • Researcher-designed tests carry most of the large positive numbers: Ghana’s EGMA, Mindspark, and the ASER scores in Chiplunkar.

  • Moderators that recur:

    • Video and software help most where teachers are weakest. In Beg, gains were larger with less-experienced teachers and in low-baseline schools. In Büchel, teacher-led gains fell with grade level and with student ability, while software gains did not.
    • Results vary widely by site. Sankalp’s district effects differed by 0.41 SD.
    • Gender was rarely a significant moderator in these trials. In São Paulo, the remote-learning losses were larger for girls.

  • Bottom line for Avanti: the evidence does not support recorded or broadcast lessons as a replacement for live teaching of motivated Grade 11-12 students. Nor does it reliably support them as an in-class supplement. The pattern that holds across studies:

    • live or structured instruction beats passive video;
    • active, adaptive practice with feedback beats watching;
    • unsupervised device self-study loses.

    The positive video results come from places where the counterfactual teacher was very weak. Overall confidence: low to moderate. There is no LMIC RCT at Avanti’s grade level and exam type.

2. Opened sources

Section titled “2. Opened sources”
# Citation / URL / archived file Design; country; population Comparison (what vs what) Outcome and delay Effect (SD) and precision N / clusters Moderators Where read (≤15-word quote) Critical flags
1 Beg, Lucas, Halim & Saif, “Engaging Teachers with Technology Increased Achievement, Bypassing Teachers Did Not”, NBER WP 25704 (Mar 2019, rev. Oct 2020); published AEJ: Policy 14(2) 2022. https://www.nber.org/papers/w25704 — beg_2019_engaging_nber.pdf Two parallel cluster RCTs; Punjab, Pakistan; government middle schools Classrooms (Gr 8): 9-minute expert videos projected on an LED screen during class, with the teacher present (video supporting teacher), vs business-as-usual. Tablets (Gr 6): the same type of content on personal student tablets (self-study, bypassing teacher) vs business-as-usual Project maths+science test (researcher-designed, aimed at higher-order skills) and the provincial PEC standardised exam; about 4 months Classrooms project test 0.296 (SE 0.130); PEC 0.269 (0.103); combined 0.263 (0.127); pass PEC +4.7 pp (0.023). Tablets −0.426 (0.149). LASSO-control specification; limited-control versions 0.256 (0.135), 0.221 (0.129), −0.420 (0.163) Classrooms 59 schools (30 T / 29 C), 2,622 project-test obs; Tablets 71 schools (19 T / 52 C), 3,058 obs Largest gains in lowest-baseline schools and students, and with less-experienced teachers; no consistent difference between girls’ and boys’ schools; teacher attendance up p.4: “eLearn Classrooms increased student achievement by 0.30 standard deviations (SD)”; Table 2 Small number of clusters; randomisation-inference p = 0.057 (project) and 0.014 (PEC). Two arms in different grades and samples, so not a head-to-head test. Cost $9/student at scale (Classrooms) vs $131 (Tablets). I did not open the published AEJ version
2 Johnston & Ksoll, “Effectiveness of Interactive Satellite-Transmitted Instruction: Experimental Evidence from Ghanaian Primary Schools”, CEPA WP 17-08 (Aug 2017); published EER 91 (2022) 102315. https://poverty-action.org/publication/effectiveness-interactive-satellite-transmitted-instruction-experimental-evidence — johnston_2017_ghana_satellite.pdf Cluster RCT; rural Ghana; Grades 2-5 (Making Ghanaian Girls Great) Live remote teacher in an Accra studio, two-way, with an in-person facilitator in the room (replaces the subject lesson), vs business-as-usual EGMA maths and EGRA reading (researcher-administered standard instruments); 1-2 years, plus one year after exposure ended Maths 0.23 (SE 0.03) pooled (Table 2, col 10); 1-year effect 0.22; still 0.21-0.22 one year after the programme ended. Reading words per minute 0.02 (0.03), not significant 147 schools (70 T / 77 C); 12,848 student-round observations Treated×female 0.06 (not significant); no significant difference below the median baseline score p.2: “Satellite classes improved students’ math scores by 0.23 standard deviations.” Primary only. Facilitators were trained existing teachers (5-day residential training). Bundled with girls’ after-school sessions. Cost: the World Bank review gives about USD 22/student (not read in the WP itself)
3 Bianchi, Lu & Song, “The Effect of Computer-Assisted Learning on Students’ Long-Term Development”, IPR WP-20-53 (Dec 2020); published JDE 2022. https://www.ipr.northwestern.edu/documents/working-papers/2020/wp-20-53.pdf — bianchi_2020_china_satellite.pdf Difference-in-differences on a staggered county rollout (2004-07) of the Modern Distance Education Program; rural China; middle school (exposed at age ≤14); outcomes from the CFPS survey Recorded and satellite lectures by top national teachers, shown in rural classrooms under local teachers (a mix of replacing and supporting), vs no programme CFPS maths and Chinese tests (survey-administered, independent of the programme); 7-10 years later Maths 0.183 (SE 0.080); Chinese 0.227 (0.064); +0.85 years of schooling (0.304) 4,996 individuals CAL×female: maths 0.103 (not significant), Chinese 0.135*; CAL×teacher education and ×teacher-student ratio not significant for maths p.18: “CAL individuals’ scores were on average 0.18σ higher in math” Not randomised; take-up inferred from administrative statistics. A bundle (DVDs, satellite, computer rooms). The authors attribute the effect mainly to lectures by high-quality teachers. Cost ≤ CNY 111 (~USD 17)/student
4 Navarro-Sola, “Secondary Schools with Televised Lessons: The Labor Market Returns of the Mexican Telesecundaria”, WP (7 Dec 2021). https://ideas.repec.org/p/hka/wpaper/2021-053.html — navarrosola_2021_telesecundaria.pdf Staggered-rollout difference-in-differences plus IV; Mexico; junior secondary (ages 12-14); ENOE labour survey TV broadcast replaces subject teachers, with one support teacher per grade, vs no telesecundaria (a mix of no school and conventional school) Schooling and earnings about 20 years later; no test-score outcome +12.5-13.9% income per extra year of schooling induced; 10 more enrolments per extra telesecundaria per 50 children ~875,000 individuals Compliers: 26-70% would otherwise not have attended any secondary school Abstract: “an additional year of education increases long-run income by 12.5–13.9%” Not a learning outcome. PISA 2003 descriptive figure (p.36, fn 58): 94% of telesecundaria students below maths competency vs 58% in conventional schools (selection, not causal). Cost per student about half: 6,811 vs 12,460 pesos (2002)
5 Chiplunkar, Dhar & Nagesh, “Too Little, Too Late: Improving Post-Primary Learning Outcomes in India”, WP (29 Jun 2020). https://riseprogramme.org/sites/default/files/inline-files/Dhar.pdf — chiplunkar_2020_too_little.pdf Cluster RCT (public lottery); 24 Chennai Corporation schools; Grade 9; Avanti after-school programme Blended: 15 five-minute Tamil videos a week (1h15m) + peer worksheets (1h30m) + facilitator mini-lectures (1h40m), after school, vs business-as-usual. Video supports a facilitator; it adds time ASER basic reading and maths (end of Gr 9); Class X board exam (independent standardised exam), about 1 year later ASER maths 1.106 levels (SE 0.244) = 0.86 SD; reading 0.559 (0.152) = 0.52 SD (the authors’ conversion). Class X total −0.278 (0.150), p = 0.064, RI p = 0.056; maths −0.015 (0.231); science −0.205 (0.183) 24 schools (12/12); ASER N = 695; Class X N ≈ 755 None reported for exam outcomes p.8: “the effect size is 0.52 SD in reading and 0.86 SD in Math”; Table 7 ASER tops out at Grade 2-4 level, so the large gains are foundational, not grade-level. The board-exam estimates are negative though not significant at 5%. The paper says “no treatment effect”. 15.7% attrition, no differential attrition found. Video was one of three components, so its own effect cannot be isolated
6 de Barros, “Explaining the Productivity Paradox: Experimental Evidence from Educational Technology”, WP (27 Sep 2022); J-PAL evaluation “Do Students Benefit from Blended Instruction?” (AEARCTR-0003665). https://www.povertyactionlab.org/sites/default/files/research-paper/wp_5802_Explaining-the-Productivity-Paradox_India_Sept2022.pdf — debarros_2022_productivity_paradox.pdf Pre-registered three-arm cluster RCT; Haryana government senior secondary schools; Grades 9-10; Avanti’s Sankalp ICT arm: teachers blend in-class teaching with 2.5-minute Hindi curriculum videos (smart TV + tablet) + workbooks + training and coaching. Video supports the teacher; teaching hours unchanged. Workbook arm: the same without technology. Control: business-as-usual Paper maths and science tests (designed by the researcher, IRT-scaled, CBSE/NCERT-mapped); about 10-11 months ICT vs control maths −0.15 (SE 0.06); science −0.03 (0.05). ICT vs workbook maths −0.10 (0.06). Workbook vs control maths −0.06 (0.06). Instructional-quality index −0.46 SD (ICT); attitudes index −0.26 / −0.24 240 schools (80 per arm); n = 24,584 Uniform across grades and baseline levels; districts differ by up to 0.41 SD p.3: “students in schools assigned to the ICT intervention performed 0.16 standard deviations lower in mathematics” (text 0.16; Table 4 −0.15) High take-up (about 1,292 minutes of video use per ICT school; 88% on days without visits). Researcher tests, not board exams. Time-on-instruction fell 6-7 pp in both treatment arms. Cost not found in the paper
7 Seo, “Do School Electrification and Provision of Digital Media Deliver Educational Benefits? First-year Evidence from 164 Tanzanian Secondary Schools”, IGC WP E-40308-TZA-2 (Sep 2017). https://www.povertyactionlab.org/sites/default/files/research-paper/WP4802_Do-school-electrification-and-provision-of-digital-media-deliver-educational-benefits-in-Tanzania_Seo_Sept2017.pdf — seo_2017_tanzania_wp.pdf (+ J-PAL summary seo_2017_tanzania_jpal_summary.pdf) 6-arm cluster RCT; northern Tanzania; Grade 11 (O-level) Solar power + TVs, with or without English or bilingual videos (lectures solving 10 years of past biology and geography exams + motivation videos), vs control. Videos supplement the teacher, used in class or after hours National O-level exam (CSEE), Nov 2016; about 1 year Solar-enabled programmes pooled 0.052 (SE 0.039), 95% CI [−0.024, 0.129]. On video subjects: English-videos indicator −0.015 (0.046); bilingual −0.033 (0.044). Solar + English videos vs control 0.040 (0.072); solar + bilingual 0.081 (0.067) 164 schools; 10,171 students Bilingual-with-solar had the largest point estimates (not significant) Abstract: “rule out impacts larger than 0.13σ and 6.7 pp in the first year.” Closest in design to exam-prep videos. Solar + English schools watched 4.2 more TV hours a week, yet showed no exam gain. Without solar there was no TV, so the video-only arms were barely treated. Cost $6.41/student. Year-2 results not found
8 Lichand, Doria, Leal-Neto & Fernandes, “The impacts of remote learning in secondary education during the pandemic in Brazil”, Nature Human Behaviour 6 (2022) 1079-1086; doi:10.1038/s41562-022-01350-6. https://pmc.ncbi.nlm.nih.gov/articles/PMC9391221/ — lichand_2022_remote_brazil_pmc.html (+ correction lichand_2022_remote_brazil_correction_pmc.html) Difference-in-differences (Q1 to Q4 2020 vs 2019) on administrative data; São Paulo State; Grades 6-12 Remote learning (mainly broadcast TV plus an app) replaces in-person classes, vs in-person learning in the prior year State quarterly standardised tests (AAP), maths + Portuguese; about 35 weeks of closure −0.319 SD (SE 0.0001; IPW-reweighted, Table 1 col 5); ≈ −0.009 SD per week. Remote learning delivered 27.5% of the in-person equivalent. Dropout risk +6.2 pp (+365%) ~7.1 million student-quarter test observations Larger harm for girls, non-white students, schools in low-income areas, and schools without prior online experience Text: “students lost approximately 0.009 s.d. of learning each week relative to in-person classes” Not randomised. Pandemic stress is confounded with mode. Tests were taken remotely in 2020. Still the most direct LMIC secondary-school estimate of broadcast/remote vs in-person
9 Büchel, Jakob, Kühnhanss, Steffen & Brunetti, “The Relative Effectiveness of Teachers and Learning Software: Evidence from a Field Experiment in El Salvador”, Bern Social Sciences WP 36 (Apr 2020); published JOLE 40(3) 2022. https://repec.sowi.unibe.ch/files/wp36/Buechel_etal_2020_CAL.pdf — buchel_2020_elsalvador_bern.pdf Class-level RCT; rural Morazán, El Salvador; Grades 3-6 Equal extra time: T1 extra teacher-led maths classes vs T2 CAL (offline Khan Academy: videos + exercises) with a technical supervisor (self-study) vs T3 CAL with a teacher; control Maths test (IRT, researcher-administered); one school year (46 × 90-minute sessions) ITT vs control: T1 0.15 (p = 0.01); T2 0.21 (p = 0.00); T3 0.24 (p = 0.00). T3 − T1 = 0.09 (p = 0.10); T3 − T2 ≈ 0.02 (p = 0.65). IV for full attendance: 0.36-0.38 198 classes (158 in 29 programme schools + 40 pure control); 3,197 students The teacher-led effect fell with grade and baseline ability; CAL did not PDF p.2 (abstract): “teachers add little to the effectiveness of learning software.” Primary. Attendance 59%. Teachers had very low content knowledge. The software is adaptive practice with videos, not lecture video alone. The table reports p-values, not SEs
10 Muralidharan, Singh & Ganimian, “Disrupting Education? Experimental Evidence on Technology-Aided Instruction in India”, NBER WP 22923 (rev. Jul 2017); AER 109(4) 2019. https://www.nber.org/papers/w22923 — muralidharan_2017_mindspark_nber.pdf Individual-lottery RCT; Delhi; Grades 6-9, after school Mindspark adaptive software + small-group instructor (blended, adds time) vs lottery losers Independent IRT tests in maths and Hindi; 4.5 months Maths 0.36 (SE 0.063); Hindi 0.22 (0.076) (ITT); IV for 90 days of attendance: 0.59 / 0.36 619 students Similar absolute gains; larger relative gains for weaker students Abstract: “lottery winners scored 0.36 higher in math and 0.22 higher in Hindi” Adaptive-software contrast, not video. Attendance 58%. Gains were mostly on below-grade content; school exams improved in Hindi but not maths
11 Muralidharan & Singh, “Adapting for Scale: Experimental Evidence on Technology-Aided Instruction in India”, NBER WP 34205 (Sep 2025, rev. Nov 2025). https://www.nber.org/papers/w34205 — muralidharan_2025_mindspark_scale_nber.pdf Cluster RCT; Rajasthan Adarsh government schools; Grades 1-8 Mindspark inside school hours, replacing 25-50% of maths and Hindi class time, vs business-as-usual Independent IRT tests; 6 and 18 months 18 months: maths 0.22 (SE 0.036), Hindi 0.20 (0.032); Grades 6-8 maths 0.25 (0.038). Year 1: maths 0.15 (0.03) 80 schools (40/40); 8,733 students (Year 2) Gains proportional to time on the platform Abstract: “treated students scored 0.22 higher in Mathematics and 0.20 higher in Hindi” A follow-on to the Mindspark lineage. Substitution design (no added time), unlike the 2019 study
12 Ma, Fairlie, Loyalka & Rozelle, “Isolating the ‘Tech’ from EdTech: Experimental Evidence on Computer Assisted Learning in China”, NBER WP 26953 (Apr 2020); EDCC 72(4) 2024. https://www.nber.org/papers/w26953 — ma_2020_isolating_tech_nber.pdf Multi-arm RCT; rural Shaanxi; Grades 4-6 boarding students After-school CAL vs the same content on pencil-and-paper vs control Maths grades and a standardised maths test; one semester Overall CAL programme: positive on grades, no positive effect on maths test scores. Isolated technology effect: “small to null”. Boys: programme +0.10σ on the maths test 4,024 students; 130 schools; 352 classes Boys gained, girls did not Abstract: “we generally find small to null effects.” Not video. Shows that the screen adds little over the same practice on paper. I did not read the table SEs
13 Angrist, Bergman & Matsheng, “School’s Out: Experimental Evidence on Limiting Learning Loss Using ‘Low-Tech’ in a Pandemic”, IZA DP 14009 (Dec 2020); NHB 2022. https://docs.iza.org/dp14009.pdf — angrist_2020_botswana_iza.pdf Household RCT; Botswana; Grades 3-5; school closures Live remote one-to-one phone tutoring (15-20 minutes a week) + SMS vs SMS only vs control Phone-administered numeracy test; about 12 weeks Phone + SMS 0.121 SD (p = 0.008); SMS alone no effect; IV for full participation 0.167 4,550 households reached — PDF p.5: “The combined phone and SMS intervention increases learning by 0.121 standard deviations (p=0.008).” Live-remote-teacher contrast; primary; phone-administered test
14 Crawfurd, Evans, Hares & Sandefur, “Live tutoring calls did not improve learning during the COVID-19 pandemic in Sierra Leone”, JDE 164 (2023) 103114. https://pmc.ncbi.nlm.nih.gov/articles/PMC10198741/ — crawfurd_2023_sierraleone_radio_tutoring_pmc.html Household RCT; Sierra Leone; primary Teacher tutoring calls added to radio lessons vs radio + SMS Maths and language tests; end of the 16-week programme No effect on maths or language (I did not extract coefficients) 4,399 students No difference by girls/boys or by public vs private teacher Abstract: “had no effect on mathematics or language test scores” 34-43% did not listen to the radio at all. Shows broadcast reach ≠ use
15 Rodriguez-Segura, “EdTech in Developing Countries: A Review of the Evidence”, World Bank Research Observer 37(2) 2022, 171-203; doi:10.1093/wbro/lkab011. https://openknowledge.worldbank.org/handle/10986/40100 — rodriguezsegura_2022_edtech_review.pdf Narrative review of causal studies in LMICs Groups studies into “improvements to instruction” (live broadcast, pre-recorded video, TV, audio) and “self-led learning” — Median effect in “improvements to instruction” 0.28 SD, 75th percentile 0.38 (k not stated on that page). Table S1.7 secondary figures: Naik et al. Karnataka 0.1-0.2 maths, 0.2-0.3 science, USD 1.7/student/year; Wennersten BridgeIT 0.36 English (not randomised) — — p.184: “none of the papers included here speak to whether EdTech can fully replace classroom instruction” Narrative review, no pooling or bias adjustment. A median of reported effects inherits publication bias
16 GEEAP, 2023 Cost-Effective Approaches to Improve Global Learning (“Smart Buys”), World Bank/FCDO/UNICEF/USAID, May 2023. https://documents1.worldbank.org/curated/en/099420106132331608/pdf/IDU1977f73d7122b1147771980c1c5a14598eef8.pdf — geeap_2023_smart_buys.pdf Expert-panel evidence review Hardware alone is a “Bad Buy”; adaptive software where hardware already exists is “Promising” — No pooled ES “over 550 evaluations” (search snippet; not checked) — PDF p.13: “Education technology should complement and enhance teachers’ engagement and not seek to replace it” Panel judgement, not a meta-analysis
17 Escueta, Quan, Nickow & Oreopoulos, “Education Technology: An Evidence-Based Review”, NBER WP 23744 (Aug 2017); published JEL 58(4) 2020 as “Upgrading Education with Technology”. https://www.nber.org/papers/w23744 — escueta_2017_edtech_nber.pdf Narrative review (mostly US/HIC) Online vs face-to-face vs blended courses (7 RCTs) — Fully online slightly below face-to-face; blended not significantly different 9 experimental online-course studies — p.71: “without some degree of face-to-face teaching, learning outcomes may suffer” HIC and higher education; listed for the substitution contrast only. I opened the 2017 WP, not the 2020 JEL version
18 Orozco-Olvera & Rascón-Ramírez, “Improving Enrollment and Learning through Videos and Mobiles: Experimental Evidence from Northern Nigeria”, World Bank PRWP 10413 (Apr 2023). http://documents.worldbank.org/curated/en/099446204182379796/pdf/IDU0b5f8c3270fed204eeb0a99b0973dd775f03d.pdf — orozcoolvera_2023_nigeria_videos.pdf Cluster RCT; rural northern Nigeria; ages 6-9 Community aspiration-video screenings for parents (not instruction); ± mobile literacy apps Literacy and numeracy; 12 months Screenings alone: out of school −42%, but learning “did not improve”. With apps: +0.46 literacy / +0.63 numeracy — Similar for boys and girls Abstract Not instructional video. Opened, then excluded from the video-vs-teacher answer

3. Leads not opened (or opened only through a secondary source)

Section titled “3. Leads not opened (or opened only through a secondary source)”
Lead Status What is known (and from where)
Borghesan & Vasey (2024), “The Marginal Returns to Distance Education: Evidence from Mexico’s Telesecundarias”, AEJ: Applied 16(1):253-85, doi:10.1257/app.20220065 Not opened: AEA behind Cloudflare; VoxDev summary returned 403 Abstract (IDEAS/RePEc, read): MTE estimates show “significant improvements in both math and Spanish after just one year”. The 0.35 SD maths / 0.23 SD Spanish figures come only from a search-engine snippet of VoxDev; unverified. This is the key missing telesecundaria test-score paper and should be opened by hand
Naik, Chitre, Bhalla & Rajan (2020), World Development 127:104736; IIMB WP 512 IIMB repository blocked by Sucuri WAF Only via the World Bank review’s Table S1.7 (read): Karnataka, 1,823 rural schools, technology-assisted lessons replacing a third of class time, 0.1-0.4 SD, USD 1.7/student/year
Beg et al. (2022), AEJ: Policy 14(2):61-90 (published version) Cloudflare Numbers above are from the Oct 2020 NBER revision and may differ slightly from the published version
Johnston & Ksoll (2022), EER 91:102315 (CC-BY) ScienceDirect blocked automated download Search snippet says a one-year numeracy gain of 0.21 SD (unverified)
Bianchi, Lu & Song (2022), JDE 102919 (published) Not OA Opened the 2020 WP instead
Fabregas, “Broadcasting Education at Scale: The Long-Term Effects of Television-Based Schools” (Mexico) No PDF found Seminar abstracts (snippets only): +8 pp lower-secondary completion, +0.4 years, +8% hourly earnings
Amazonas “Ensino Médio Presencial com Mediação Tecnológica” (Centro de Mídias) No causal evaluation found Only descriptive theses and programme-staff claims
Jamison et al. (1981), Nicaragua radio mathematics; Näslund-Hadley et al. (2014), Paraguay audio (0.16 SD per the World Bank review) Not opened Pre-recorded audio, primary/preschool
Wennersten et al. (2015), BridgeIT India (teacher-shown videos, Gr 5-6; not randomised) Not opened World Bank review: 0.36 SD English, USD 10.50/student
Gambari et al. (2016), Nigeria senior-secondary video geometry (120 students) Not opened World Bank review: positive, no SD conversion possible
Seo, Tanzania follow-on trial (AEARCTR-1072, ~171 schools, cross-randomised maths textbooks, videos, scholarships) and Year-2 results of the GivePower trial No results located Registry descriptions only
Sankalp follow-up (de Barros), second cohort / second year; any published version Not located Worth asking Avanti or J-PAL directly. Avanti also holds its own Haryana RCT report
Hassan et al. (2021) Bangladesh telementoring; Radhakrishnan et al. (2021) Nepal; Schueler & Rodriguez-Segura (2021) Kenya Not opened As cited in Crawfurd et al. (read): 0.56/0.66 SD; ~0.2 SD; no effect
Bettinger et al. (Russia) CAL substitution; Major, Francis & Tsapali (2021) BJET meta-analysis of personalised learning in LMICs Not opened —
COVID TV programmes (Pakistan Taleem Ghar, Peru Aprendo en Casa, Bangladesh Sangsad TV, Mexico Aprende en Casa) No test-score impact evaluations found World Bank feature (snippet): impact on learning “difficult” to assess
India recorded vs live online coaching / JEE-NEET test prep Nothing causal found Search returned only provider marketing
Last 30 days (Oct 2026) Nothing found that fits The Mindspark-at-scale WP (Nov 2025) is the newest lineage item