Start with the number that breaks the assumption most people carry into this topic. Government school teachers in India are, on paper, unambiguously the better hires. They clear a competitive recruitment process, hold mandatory formal training and education qualifications, and earn roughly four to five times what a private school teacher in the same village earns. And yet a landmark study following the same villages found private school teachers absent roughly a third less often than their government counterparts. Only 1 in 3,000 government head teachers has ever dismissed a teacher for excessive absence. Something other than raw teacher quality is deciding which of these two systems actually functions. Explaining what that something is takes four pieces of economic theory. Fixing it, it turns out, does not require inventing anything new, because Estonia, Sweden, and an Indian NGO have each already run the experiment.
Chapter 1Why does a better qualified, better paid teacher show up to work less often?
This is a principal agent problem in close to its textbook form, and it is worth naming precisely rather than vaguely.
A principal, here the state, hires an agent, the teacher, to do a job the principal cannot cheaply and continuously observe. The government cannot station an inspector in every classroom every day, so it has to trust that pay, qualification, and a sense of public duty will produce effort. Private schools face an identical monitoring problem, with one structural difference that changes everything downstream of it. A private school's management sits physically close to the classroom, parents paying fees directly complain fast and loudly when a teacher is absent, and critically, a private school can simply fire a teacher who does not show up. A government school, once a teacher clears recruitment, offers close to permanent tenure, and the 1 in 3,000 dismissal rate found in the original Kremer, Chaudhury, Muralidharan, and Rogers study is not a rounding error, it is close to a formal guarantee that absence carries no real consequence.
Nationally, government teacher absenteeism runs at 24% to 25% of the working calendar, rising as high as 38% in some states, and even when a teacher is physically present, studies found they spend less than half the school day actually teaching. Economists Chaudhury and Muralidharan estimated the direct fiscal cost of this absenteeism, salaries paid for days not worked, at roughly 1.5 billion dollars a year, before counting the cost of the actual learning never delivered. This is moral hazard operating exactly as the theory predicts. Remove the credible threat of losing the job, and average effort falls, regardless of how qualified or well paid the person doing the job originally was.
Chapter 2Does this actually show up in learning outcomes, or is it just an inconvenient statistic about attendance?
It shows up, and the cleanest evidence is not an observational comparison, which always risks comparing different kinds of families, but an actual randomised trial. Karthik Muralidharan and Venkatesh Sundararaman ran the Andhra Pradesh School Choice Project, offering private school scholarships by lottery to over 6,000 students across 180 villages and tracking winners against otherwise identical lottery losers for four years. Private school teachers in the study had markedly lower formal qualifications and training than their government counterparts, exactly as the premise assumes. They also delivered slightly better test score gains than the government schools the lottery losers stayed in, at roughly one third the cost per student. Muralidharan's own framing of the result is the sharpest way to state it: since less qualified, worse paid private teachers matched or beat government outcomes at a third of the cost, the real puzzle is not why private schools are cheap, it is how much further ahead government schooling could get if it spent its existing budget on accountability rather than assuming input quality alone would carry the outcome.
Chapter 3So why do parents keep leaving, even the ones who could plausibly push government schools to improve instead?
This is where the economics moves from teacher behaviour to parent behaviour, and it is the part that actually matches the shape of a genuine game theory problem, not just a loose metaphor for one.
Picture the two possible collective outcomes. If most parents who have the means, the informed, the politically connected, the ones with time to attend a school committee meeting, stayed in the government system and pushed for accountability together, the system has real scale advantages, established infrastructure, existing teacher training pipelines, and enough politically engaged parents to make absenteeism costly for a local administration to ignore. Call that the cooperative outcome. It plausibly beats what any single family can buy privately. But no individual family can unilaterally produce that outcome by staying enrolled alone, and every family choosing where to send one specific child faces a short, fixed, non repeatable window, roughly ages four to eighteen, in which to make the bet. If enough other families have already left, a single family's decision to stay changes nothing about the peer group, the accountability pressure, or the school's political weight, while the actual downside, a specific child losing years of foundational learning while the system theoretically improves, is entirely and irreversibly borne by that one family. The individually rational move is to exit regardless of what other parents are doing, which is precisely the structure of a prisoner's dilemma, mutual cooperation would beat mutual defection, but defection is the dominant strategy for each player taken alone, and the aggregate result is everyone ends up worse off than the cooperative outcome that was theoretically available.
Chapter 4Is there a more precise name for what happens after that exit starts?
There is, and it belongs to the economist Albert Hirschman, whose 1970 framework of exit, voice, and loyalty is close to a perfect description of what India's own enrolment data has been recording in real time. Hirschman's insight was that a dissatisfied customer of any institution, a company, a political party, a school system, has two available tools, voice, staying and agitating for improvement, or exit, leaving for an alternative. His deeper and more counterintuitive point was that making exit too easy can actively weaken voice, because the people most capable of exercising effective voice, the educated, the resourced, the ones a bureaucracy is most likely to actually listen to, are also the people with the cheapest exit option. When they leave first, voice does not merely lose a few members, it loses its most effective ones, and the institution is left facing only the complaints of people with the least power to force change.
This is exactly the mechanism visible in India's own recent numbers. Private unaided school enrolment rose by 4.8 million in a single year while government enrolment fell by a comparable margin, and the families making that move first are disproportionately the ones with the income to pay fees, the literacy to research alternatives, and, not incidentally, the social capital to have made the loudest, most effective noise about a locally absent teacher had they stayed to make it. Their exit does not just shrink the government system's enrolment count. It quietly removes the demographic most likely to have generated the political pressure that might have fixed the 1 in 3,000 dismissal problem in the first place, leaving the system to answer to a remaining parent body with structurally less leverage, which makes the next round of exit even more rational for whoever is left with the means to make it. That is not a one time market adjustment. It is a self reinforcing spiral, and Hirschman's framework predicts both the direction and the mechanism of it, not just the outcome.
Chapter 5Does the private school choice add up to more than just avoiding a bad system?
One more layer explains why parents choose private schooling even where the actual measured learning gain over government schooling is, per Muralidharan's own RCT, modest rather than dramatic. Economist Michael Spence's signalling theory offers the missing piece. A credential can be valuable to a family independent of how much it actually improved a child's underlying ability, purely because it communicates something to a future employer or an admissions officer who cannot directly observe that ability either. An English medium private school education functions as exactly this kind of signal in the Indian labour market, employers reading it as a proxy for a certain kind of family investment and preparedness regardless of the measured test score gap behind it. This is the rat race element working exactly as the phrase implies, a positional good where relative standing against other applicants matters as much as absolute learning, which means even a family who correctly understood that the actual pedagogical gain from private school was marginal could still rationally choose it, because the signal itself carries labour market value that a government school diploma, fairly or not, is currently perceived to carry less of.
Put the four theories together and the simple version of this story, government hires better teachers but parents irrationally chase private school status anyway, turns out to get the facts backwards in the most important place. Government teachers are indeed better credentialed. That credential is being neutralised by a principal agent failure so complete that dismissal for absence is statistically almost fictional. Parents leaving is the individually rational move in a structure that rewards defection regardless of what other parents do, and the families most capable of forcing the system to fix itself are systematically the first to leave rather than stay and fight, exactly as Hirschman's framework predicts. Layer signalling on top and even a family who has done the maths correctly still has a rational reason to pay for the label. None of this is a story about villains or a story about panic. It is a structure, and structures respond to redesign in a way that villains and panics do not, which raises the actual question this piece exists to answer. Has any country redesigned its way out of exactly this trap.
Chapter 6Estonia solved the principal agent problem this piece just diagnosed, without ever calling it that
For twenty years, Finland was the pilgrimage site education ministers flew to. That pilgrimage is now going to the wrong country. Estonia placed first in Europe across every subject in PISA 2022, producing 1.5 times more top performers and half as many low performers than the OECD average, and did this while spending 11,116 dollars per student, meaningfully below the OECD average of 13,210 dollars. Estonia is not outperforming the rich world by outspending it. It is outperforming the rich world on a discount.
What Estonia actually does is, read against the theory two sections above, a direct fix to the exact failure mode diagnosed there. Every teacher holds a master's degree, a genuinely high qualification bar. But unlike India, Estonia pairs that qualification with real school level autonomy, hiring, curriculum design, and method all sit with the school rather than a distant ministry, and Estonia ranks first in the entire world for how much freedom PISA's own survey finds its teachers actually have. Autonomy paired with a demanding entry bar is, functionally, a monitoring solution, a school that hired and trained its own teacher has both the information and the standing to manage that teacher's performance the way a private school's management does in the Indian data above, without needing the qualification trade off Indian private schools make to get there. Estonia found a way to keep the government teacher's credential and gain something close to the private school's accountability, rather than being forced to choose one or the other the way India currently is.
Chapter 7Even the old idol is proof no model survives on autopilot
Before treating any single country as a permanent template, Finland's own recent numbers are the necessary cold shower. Finnish basic skills have been declining for a decade. Since 2012, underachievement in mathematics is up 12.6 percentage points, in reading up 10.1 points, in science up 10.6 points, each roughly 3 to 4 points worse than the EU's own average decline over the same period. Finland has had to respond the way any serious system should, not with denial, but with an actual policy correction, adding extra weekly literacy and numeracy instruction hours starting in 2024 specifically to arrest the slide.
The lesson here is not that Finland failed. It is that an education system is a living thing that decays the moment a country stops actively tending it, precisely the discipline the correction demonstrates and precisely what a genuinely evidence driven system does that a merely well branded one does not. Copying Finland's 2005 policy manual in 2026 would mean copying a system already mid decline. The habit worth copying is the willingness to look at your own falling numbers and act on them inside two years, not the specific policy that happened to work twenty years ago, and it is the same habit, seen from a different angle, that separates Estonia's continuous school level monitoring from the Indian government system's 1 in 3,000 dismissal rate.
Chapter 8The attention span problem is not theoretical anymore, and the country that pioneered classroom tech just reversed itself
This is the part of the international picture most directly relevant to a country that has spent the last decade treating a tablet in every classroom as self evidently progress. Sweden made mobile phones and tablets compulsory even in preschools by 2019, becoming Europe's most aggressive adopter of digital first education. Sweden is now dismantling that policy, starting with a nationwide school phone ban from autumn 2026, alongside a legal guarantee that every student gets access to a printed textbook, over 2 billion krona in cumulative grants for physical books and teacher guides since 2025, mandatory staffed school libraries, and a return to pen and paper for national tests from 2026.
The reversal was not ideological. It was forced by the country's own numbers. PISA 2022 found 24.3% of Swedish ninth graders did not reach basic reading comprehension, underperforming the UK, the US, and Finland, and PIRLS data separately showed Swedish reading comprehension scores actually declining between 2016 and 2021. The single most important finding behind the reversal came from the OECD's own analysis of Sweden's one to one laptop policy across 2012 to 2022, that it did not improve outcomes overall and produced small negative effects in mathematics specifically among students from less educated family backgrounds. Handing out devices without a clear pedagogical purpose behind them did not close the gap between richer and poorer students. It widened it.
Sweden is not an outlier making a lonely choice. By the end of 2024, 85% of countries across Central and South Asia had already introduced school phone bans, against 40% in Europe and North America and 28% in Latin America and the Caribbean, and Denmark, the Netherlands, France, Italy, Chile, and South Korea have all moved the same direction in the past two years. For a country whose own ASER data shows three in four Class 3 students still cannot read a Class 2 level text, the Swedish finding is not a foreign curiosity. It is a direct warning against solving a foundational literacy crisis by handing every child a screen and calling it modernisation.
Chapter 9If phones and tablets are the wrong tool, does that rule out short educational videos too?
Not automatically, and the evidence here is precise enough to draw an actual line rather than a blanket verdict, which matters given how often the fix proposed for exactly this crisis is some version of shorter, punchier digital content, twenty second facts replacing an Instagram Reels or TikTok feed rather than competing against a whole tablet. Two studies published within a year of each other land in genuinely opposite places, and both are worth knowing before building anything. A meta analysis of microlearning in language education found short video instruction produced a 1.43 standard deviation jump in speaking proficiency, a large effect by education research standards, while a control group taught identical material through a textbook showed no statistically significant gain at all over the same period. A separate academic system, ReelsEd, used large language models to convert full lectures into structured short clips and beat traditional long form video on engagement, quiz performance, and task efficiency in a controlled study, without raising cognitive load.
Set against that, a study published in Communications Psychology in June 2026 ran three separate experiments on over 150 college students, showing identical informational content, travel footage, nothing algorithmic or entertainment driven, as either one continuous video or the same material split into short clips. The short clips measurably fragmented both attention during viewing and memory afterward. Heavy short form video use more broadly has been linked in a 2019 to 2025 research review to shorter attention spans, weaker academic performance, and measurably abnormal white matter in brain regions tied to behavioural control, concentrated most in adolescents whose attention systems are still developing.
The variable separating these two outcomes is not runtime, it is exactly the same variable the Swedish tablet story turned on. Every study that found short video helping paired it with a single deliberate objective and a forced retrieval step, a quiz, spaced repetition, a reflective task, inside a closed system built for learning. The fragmentation findings describe unstructured, often algorithmically driven consumption, the actual design of a commercial Instagram Reels or TikTok feed, whose entire business model is maximising scroll time, not maximising what any one viewer remembers an hour later. Replacing doomscrolling with twenty second facts only helps if the twenty seconds sit inside a closed, curriculum aligned system that ends in a quiz, not inside the same feed architecture that fragments attention regardless of how accurate the fact being taught is. India already has a government platform built on close to the right architecture for this. DIKSHA, built as a teacher facing repository and scaled into a full national ecosystem under PM eVidya during the pandemic, is closed, curriculum aligned, and free of the infinite scroll incentive that undermines the same idea when attempted inside a commercial app.
Chapter 10India already built the actual fix for the teaching method itself. It has mostly used it on other countries.
Here is the part of this piece that should embarrass rather than merely inform, because it means a version of the solution was never missing, only underused at home.
Pratham, an Indian NGO, developed Teaching at the Right Level in the early 2000s, a method that groups children by actual demonstrated ability rather than by age or grade, and teaches reading and arithmetic at the level a child is actually at rather than the level the syllabus assumes they should be at. Abhijit Banerjee and Esther Duflo, who partnered with Pratham on the earliest randomised trials of the method, went on to share the 2019 Nobel Prize in Economic Sciences, work that drew directly on this partnership. Six separate randomised evaluations across seven Indian states found some of the largest effect sizes ever rigorously measured in education research. A related computer assisted maths version of the approach improved scores by 0.47 standard deviations, a genuinely enormous effect in education research terms. The method has since been exported and adapted as TaRL Africa, running in multiple African countries, and was named a Good Buy by the Global Education Evidence Advisory Panel in 2020, an international endorsement of a homegrown Indian idea.
The honest part of this story loops directly back to the principal agent problem this piece opened with. Early attempts to hand TaRL to government teachers in Bihar and Uttarakhand, despite well received training and ongoing Pratham support, simply did not stick, teachers did not adopt the method in practice even after being trained on it, the same unmonitored, unaccountable classroom behaviour that produces 24% absenteeism showing up here as quiet non compliance with a proven method instead. It took a redesign, adding structured mentor support and closer onsite monitoring within the government system itself, essentially manufacturing, by hand, a small piece of the accountability Estonia builds in structurally, before large scale versions in Haryana and Uttar Pradesh actually worked. The pedagogy was never the hard part. Getting an enormous, understaffed, inconsistently supervised government bureaucracy to actually change daily classroom behaviour was, and it is the identical obstacle in a different costume from the one this entire piece started with.
Chapter 11So what does the full picture actually say, theory and evidence together?
Line up the diagnosis and the fixes and one thread runs under all of it. The principal agent problem explains why qualification and pay alone were never going to be enough. The prisoner's dilemma explains why individual families cannot fix that problem by staying and hoping. Hirschman's exit and voice framework explains why the exit, once it starts, removes exactly the people capable of forcing a fix, making the spiral self reinforcing rather than self correcting. Signalling explains why even a family that understands all of this still has a private incentive to pay for the credential regardless. And every functioning fix this piece found abroad, Estonia's autonomy paired with a real qualification bar, Sweden's willingness to reverse a flagship policy the moment its own data turned against it, the narrow evidence backed case for closed system microlearning over open feed consumption, and TaRL's own hard won redesign after watching two states fail to adopt it, is a version of the same move, building real, continuous accountability into a system rather than assuming a good enough input, a qualified teacher, a printed textbook, a laptop, a fact delivered in twenty seconds, will carry the outcome on its own. India has the underlying material to do this today. It has its own Nobel connected pedagogy sitting half deployed in its own classrooms, a live global evidence base on exactly when technology helps and when it quietly widens the gap it was bought to close, and a government platform in DIKSHA built on close to the right architecture already. What it has not yet shown is the one habit every example in this piece shares, treating a bad number as an instruction to redesign the incentive behind it, rather than a statistic to survive until the news cycle moves on.