Abstract
Knowing how many people live in a place looks like an unmixed good. Planners need it, hospitals need it, water systems need it, and famine relief cannot be sized without it. Yet a substantial share of the world’s population lives in states whose last full enumeration is decades old, and the reason is almost never technical. It is that in most modern constitutional orders the census is not a description of the population but an instrument that distributes seats, money, offices, language rights, and territorial claims. Once counting is welded to allocation, the count stops being an observation and becomes a move in a game where someone must lose. This paper argues that census paralysis is best understood not as state incapacity but as a rational equilibrium: where the allocation formula is already fixed and the count is the only remaining variable, uncertainty about population becomes a shared asset that no coalition member has an interest in destroying. It develops a typology of census failure — frozen settlements, inflated counts, postponement equilibria, category suppression, participation contests, and second-order fights over adjustment — and closes with the implications for megaregional analysis, where the units of interest have no political principal and therefore no constituency demanding that they be counted at all.
I. The Problem Stated
The Nigerian case makes the shape of the thing plain. Nigeria has held national headcounts in 1953, 1963, 1973, and 2006; the 1973 exercise was annulled over allegations of manipulation, and counts scheduled for 2016, 2018, and 2023 were each postponed. The 2023 census was called off days before it was to begin, with the outgoing president deferring the choice of a new date to his successor. Two decades on from the last count, a country of well over two hundred million people is governed on the basis of projections rather than an enumeration, and observers now note that scheduling a census close to the 2027 general elections would itself invite suspicion.
Ethiopia’s constitution requires a decennial count. The last one was held in 2007; a census planned for 2020 did not happen, and specialists attribute the delay in part to the way population figures feed regional budget allocations. Constituency demarcation in 2020 was carried out using data from 1994, when the country held roughly half its present population. Lebanon has not conducted a national census since 1932. Afghanistan has never completed one in the modern era. Iraq went twenty-seven years without a full count, and when it finally held one, it excluded questions on ethnicity and religious sect, leaving the country’s demographic composition formally unknown.
These are not places that lack clipboards. Ethiopia mustered a nationwide count in 2007 under considerably worse logistical conditions than it faces now. Nigeria demarcated enumeration areas, procured digital assistants, and trained staff for a census it then cancelled. The obstacle is not capacity. It is that everyone involved can calculate what the number will do.
II. Counting Is Not Describing
The census in a modern state is rarely a purely statistical act, because constitutional and fiscal machinery is wired directly into its output. Seats in legislatures are apportioned by population. Intergovernmental transfers are formula-driven on population. Constituency boundaries are redrawn on population. In federal systems, the creation of new subnational units — states, provinces, regions, local government areas — is often triggered by population thresholds. Minority language rights, in several constitutional orders, switch on at a specified percentage.
This wiring converts a measurement into a distribution. The enumerator arriving at a door is not gathering information about the household; he is, in effect, allocating a fraction of a legislative seat and a fraction of the national budget. Every party in the system understands this, which is why the arrival of the enumerator is treated with the seriousness normally reserved for an election.
The American founding made the linkage explicit rather than implicit. Article I ties both representation and direct taxation to the same enumeration, and the three-fifths compromise was not a statement about the humanity of enslaved persons in some abstract register but a bargain about how much representation slaveholding states would extract from a population that could not vote. That is the census in its purest political form: a fight over the denominator conducted in the language of arithmetic.
The wiring also means that the count and its consequences can be severed at either end. A state may refuse to count, as Lebanon and Ethiopia have. Or it may count and then refuse to act on the result, which is what the United States did after 1920, when the House of Representatives simply declined to reapportion itself for a decade rather than transfer seats from rural to urban districts. India has done something similar and more durable: seat allocation among states has been frozen against 1971 population figures, a freeze imposed by the 42nd Amendment in 1976 and later extended, with delimitation now to follow the first census taken after 2026. Southern states that succeeded in reducing fertility stand to lose share in any population-based redistribution, and would be further disadvantaged if the total seat count remains capped. The freeze is a fifty-year admission that the country could count itself but could not survive acting on the count.
III. The Option Value of Not Knowing
The deepest reason censuses go untaken is that ambiguity has value, and the value is shared.
Where no authoritative number exists, every group can maintain a maximal estimate of its own size. A Lebanese confessional community, a Nigerian geopolitical zone, an Ethiopian nationality, an Iraqi governorate — each can assert a share, and no rival can conclusively refute it. Those claims are simultaneously believed by their holders and unfalsifiable, and they underwrite bargaining positions across every arena of politics. A census destroys this. It replaces a set of overlapping claims, all of which can be sustained, with a single official figure, most of which someone must accept as a loss.
The asymmetry is what makes the equilibrium stable. The gains from an accurate count are diffuse, delayed, and technocratic: better planning, better targeting, better infrastructure siting. The losses are immediate, concentrated, and legible: seats, revenue shares, offices, territorial claims. A coalition partner who might gain slightly from an accurate count and lose catastrophically from an unfavorable one will rationally prefer the status quo. And because the losses fall on identifiable actors while the gains fall on the population at large, the blocking coalition is always better organized than the enabling one.
This generates what is best called a postponement equilibrium. Nobody has to defend a boycott. Nobody has to argue publicly that the nation should remain uncounted. Delay is deniable and comes ready-supplied with respectable justifications: funding shortfalls, insecurity, technical preparation, an approaching election, a public health emergency. Nigerian commentary lists funding constraints, logistical difficulty, security concerns, and political considerations in a single breath, and each is true enough to carry the weight of a postponement on its own. The census is never refused; it is always merely deferred. And deferral, repeated, is indistinguishable in its effects from refusal while costing far less politically.
There is a further wrinkle. In several of these systems the allocation formula is already settled and publicly known. That is precisely what makes the count so dangerous. When the rule is fixed and only the input is uncertain, every actor can compute in advance exactly what any given result will cost him. There is no veil of ignorance left to make agreement possible. Bolivia demonstrated this in 2022, when the timing of the census — not its questions, not its methods, merely when it would be conducted and therefore which fiscal and electoral cycle its results would govern — became the object of a weeks-long general strike in Santa Cruz. The number had not yet been produced, and people were already in the streets over it.
IV. The Schedule Is a Constitution
The second locus of conflict is not how many but what kind. A census form is a taxonomy, and taxonomies determine which groups exist for official purposes.
This is why the fight over categories is often fiercer than the fight over the count. To include a category is to make it politically countable, and therefore claimable. To exclude it is to render the group statistically invisible and its claims unprovable. Neither is neutral, and there is no third option: leaving the question off the form is as consequential as putting it on.
Iraq’s 2024 census is the cleanest recent illustration. Questions on sect and ethnicity had blocked any count since 2003 until the political parties agreed to leave them out. The Kurdistan Regional Government itself proposed the exclusion, apparently out of concern that ethnic data from Kirkuk and other disputed areas could be used against the referendum provision of the 2005 constitution, given decades of Arabization and substantial Arab in-migration since 2003. One official defended the omission on the ground that ethnicity accounted for a small share of the questionnaire and that securing ninety-five percent of the data was better than losing all of it; a critic replied that the exclusion was deliberate and aimed precisely at obscuring the composition of Kirkuk and Nineveh. Both were right. The census became possible only by agreeing not to learn the thing most worth knowing.
Nigeria arrived at the same settlement in 2006, dropping ethnicity and religion from the form so that a count could occur at all. Myanmar’s 2014 census, its first in three decades, permitted enumeration but did not permit Rohingya respondents to identify themselves as such, with the result that roughly a million people were excluded from the published figures — an instructive demonstration that a census can be simultaneously conducted and refused.
Belgium offers the endgame. Language censuses there determined the linguistic status of communes, and each count therefore threatened to move the frontier between Flemish and French administration. Rather than continue fighting a war every ten years, Belgium fixed the linguistic boundary by statute in the early 1960s and abolished the language census outright. The country decided that some facts about itself were better left unrecorded, and it has stuck to that decision for over sixty years. It is worth noting that this was not a failure of a poor or fragile state. It was a wealthy, capable democracy concluding that a particular question was not worth the answer.
Ethiopia’s disputes run the same way. The 2007 count allowed more accurate self-identification than its predecessors, with consequences for the recorded size of the Oromo population, and subsequent proposals to remove compulsory ethnic and religious identification were read by critics as an attempt to conceal identity in the service of one elite faction against the federal order. Some groups recognized in earlier counts, such as the Qimant, were omitted in 2007 — a reminder that the category list is revised by someone, for reasons.
V. Older Than the Modern State
The suspicion attaching to enumeration is not a modern artifact, and Scripture treats it with a seriousness that repays attention.
The censuses in Numbers are commanded, ordered, and specific: a count of the men able to go to war, taken by tribe and by father’s house, and a second count a generation later before the division of the land. Enumeration there is not forbidden. It is an instrument of allotment — the inheritance is divided according to the numbers — which is exactly the point. The count is bound to the distribution from the beginning.
Yet Exodus 30:12 attaches a condition to the taking of the sum of the children of Israel: a ransom is to be given for each soul numbered, “that there be no plague among them when thou numberest them.” The count is permitted, but it carries a liability, and the liability is discharged by an acknowledgment that the people belong to God rather than to the one counting them. When David numbers Israel and Judah in 2 Samuel 24 and 1 Chronicles 21, Joab objects before the count begins, David’s conscience strikes him after it ends, and judgment follows. The plain reading is not that counting is wrong but that counting for the wrong purpose — to measure one’s own strength, to convert a people into an asset on a balance sheet — is a claim of ownership that the counter is not entitled to make.
The same instinct persists in the later record. The enrollment under Cyrenius in Luke 2 is undertaken for the purposes of an occupying power, and Acts 5:37 records that Judas of Galilee rose up in the days of the taxing and drew people after him. A census provoked armed revolt because everyone understood what it was for. The Roman state was not curious about Judea; it was assessing it.
That understanding — that the count is an assertion about who owns whom, and about what will be done with the answer — has never left. The peasant who hides a child from the enumerator, the household that declines to answer, and the political party that blocks the questionnaire are all making the same calculation the ancient world made. They are asking what the number is for. In most of the cases in this paper, the answer is a bad one for somebody.
VI. A Typology of Census Failure
The frozen settlement. A power-sharing arrangement is calibrated to a historical population distribution and cannot survive its revision. Lebanon’s confessional allocation traces to a count taken in 1932, under rules that included emigrants with maintained ties and thereby shaped the resulting balance; the National Pact and the later Taif revisions adjusted the ratios without ever refreshing the underlying data. Belgium fixed its linguistic frontier and abolished the count. India froze seat allocation to 1971 figures. In each case, the settlement is stable precisely because the number is not allowed to move.
The inflated count. Where the count is decisive and enforcement is weak, the incentive runs toward manufacturing population. Nigeria’s 1962 and 1963 counts and the annulled 1973 exercise all foundered on this, and the pattern has not gone away: the 2006 census remained disputed on methodology and alleged undercounting, particularly in Lagos and parts of the South-East, with Lagos State conducting a parallel count and demanding a recount when the two disagreed. Kenya cancelled the results for several northern districts after its 2009 count on grounds of implausibility, a decision the affected communities read as political. Inflation and cancellation are two faces of the same disorder: once numbers can be made, they can also be unmade, and neither side can trust the other’s figure.
The postponement equilibrium. Discussed above. Ethiopia and Nigeria are the exemplars. The tell is that each new deferral is justified on grounds unrelated to the previous one, and that the funding line item is the reliable veto point — Nigerian optimism about a count now hinges on a budget allocation to the population commission. A body that must ask permission for its own budget can be silenced without anyone appearing to silence it.
Category suppression. Iraq 2024, Nigeria 2006, Myanmar 2014, Belgium after 1961. The count proceeds by amputation. This purchases an enumeration at the cost of the information that made the enumeration politically fraught in the first place, which is a genuine trade rather than a fraud — but the resulting data cannot answer the questions that provoked the fight, and the underlying disputes are merely postponed into whatever proxy evidence remains available.
The participation contest. Where the count is contested but proceeding, groups compete over who is enumerated and where. Despite the exclusion of ethnicity from the Iraqi form, and despite an accompanying political agreement that the results would not bear on the boundaries of Kurdish-administered territory, Kurdish parties urged families to register in Kirkuk. Turkey’s foreign ministry protested the resulting movement toward Kirkuk on behalf of Turkmen and Arab residents, arguing that it would fold into the province’s population many people not originally from it. Bosnia’s 2013 census, its first since 1991, produced a three-year fight over whether long-absent emigrants counted as residents, because the answer moved ethnic shares that determine constitutional offices. North Macedonia abandoned its 2011 count mid-enumeration and conducted the 2021 count amid boycott campaigns, with an Albanian population share near the twenty percent threshold that triggers language rights under the Ohrid framework. In Northern Ireland, the religion question is followed as a running score. In each case the census functions as an election with different rules — turnout determines the outcome — and the incentives to mobilize, and to suppress, are identical to those of an election.
The second-order fight. In states where the count itself is not seriously in doubt, conflict migrates to the treatment of the count. The United States is the mature case: litigation over a citizenship question in 2019, a statutory bar on sampling for apportionment purposes, contention over disclosure-avoidance methods applied to the 2020 data, and a persistent dispute over prison gerrymandering, in which incarcerated persons are counted at the facility rather than at their home address, transferring representation from urban to rural districts. None of these fights is about whether to count. All of them are about what happens between the count and the consequence, which is where the remaining discretion lives.
VII. Who Pays for the Missing Number
The costs of not counting are not distributed evenly, and they do not fall on the people who benefit from the silence.
Projections drift, and they drift worst where growth is fastest and change is least regular — informal urban settlements, displaced populations, pastoralist regions, rapidly expanding secondary cities. These are also the populations least well captured by administrative records, so the error compounds rather than averaging out. Vaccination campaigns sized on stale denominators miss their targets. School construction follows a map of where children were two decades ago. Famine relief, epidemic response, and flood evacuation all depend on knowing how many people occupy a given place, and all are calibrated from the same aging baseline.
The political economy is grim in its consistency: the groups whose numbers are suppressed or unrecorded are generally the groups with the least leverage, and the absence of a count does not merely fail to help them — it forecloses the evidence they would need to make a claim. A group that cannot demonstrate its size cannot demand proportionality. Statistical invisibility is not an accident of underdevelopment. In a fair number of these cases it is the intended outcome.
One Nigerian legislator observed that low voter turnout and school enrollment figures sit awkwardly with the projected national population — which points at the other cost. Without a census, every rival estimate becomes admissible, and public argument degenerates into a contest of proxies that each side selects for advantage. The census, whatever its faults, at least concentrates the dispute into a single procedure that can be observed and criticized. Its absence disperses the dispute into everything.
VIII. What This Means for Megaregional Analysis
Four consequences bear directly on the megaregion work.
First, censuses count residence, and megaregions are constituted by flow. The de jure resident population of a corridor tells you relatively little about the functional population that commutes, delivers, studies, and trades across it. Even excellent censuses answer a question adjacent to the one megaregional analysis asks. Where journey-to-work data exist they help; in most of the world they do not exist at the necessary resolution, and cross-border corridors — Detroit–Windsor, San Diego–Tijuana, Singapore–Johor, the Rhine axis — compound the problem with unsynchronized census dates and incompatible definitions of urban area.
Second, where the count is decades stale, megaregional population is a model output rather than an observation, and its error is structured rather than random. Gridded population products calibrate against national censuses; where the census is missing, the product is extrapolating, and the extrapolation is least reliable precisely where the megaregion is growing fastest. The Lagos–Ibadan corridor, the Addis corridor, the Kinshasa–Brazzaville conurbation, and much of the Karachi and Kabul regions sit in exactly this zone. Analysis of these corridors should carry explicit intervals, not point estimates dressed as facts.
Third, contested counts distort at the corridor scale in ways that matter for the argument. Lagos disputed the 2006 federal figures and ran its own parallel enumeration — a state government contesting the population of the anchor of West Africa’s largest urban region because state revenue depended on the answer. Any comparative claim about that megaregion’s size rests on a number that is itself a political artifact.
Fourth, and most usefully for the wider project: megaregions have no political principal. No legislature is apportioned by megaregional population; no transfer formula keys to it; no office is allocated on the strength of it. The interstate compact literature and the tri-state work already established that jurisdictions underinvest in places whose benefits do not accrue to any single jurisdiction. The census case adds the statistical corollary. Because nobody’s seat depends on the megaregional number, no organized constituency demands that it be produced — and so it is not merely undercounted but institutionally uncounted, assembled after the fact by researchers from units drawn for other purposes. This is the same negative-space problem in a new register: the entity most worth measuring is the one no principal is charged with measuring, and its invisibility is a consequence of the allocation architecture rather than of any technical difficulty.
IX. Making the Count Survivable
If the diagnosis is that counting fails when it is welded too tightly to allocation, the design response follows.
Loosen the coupling in time. Tie allocation formulas to multi-census averages or lagged figures rather than to the most recent count alone. This reduces the stakes of any single enumeration without freezing the settlement permanently, which is the failure mode of the Indian and Belgian solutions.
Damp the marginal effect. Floors, caps, and phased transitions reduce how much a marginal person is worth. Where one household can move a seat, households will be manufactured.
Set the rule behind the veil. The Bolivian and Nigerian pattern arises because the formula is known and only the input is in play. Agreeing the allocation rule before the count is taken, with the rule binding for a fixed period afterward, restores some of the uncertainty that makes agreement possible.
Firewall the statistical census from the administrative state. Where enumeration is suspected of feeding conscription, taxation, land adjudication, immigration enforcement, or a citizenship register, refusal is rational and widespread. India’s linkage of census machinery to a population register generated boycott calls that had nothing to do with statistics. A hard legal separation, publicly and repeatedly demonstrated, is a precondition for cooperation.
Prefer category pluralism to category suppression. Allowing multiple self-identifications and publishing cross-tabulations is more informative than forcing a single box, and less destructive than deleting the question. Iraq’s compromise bought a count at the price of the information; a form permitting layered identification might have bought both, though only if the allocation formula did not turn on the answer.
Publish uncertainty. A single point estimate invites treatment as a sacred fact and cannot be argued with except by rejecting it wholesale. Confidence intervals and documented post-enumeration surveys make the number contestable in a bounded way, which is healthier than making it contestable only by riot.
Protect the counter. An enumerating agency whose head serves at pleasure and whose budget is a discretionary line can be shut down without anyone taking responsibility. Fixed terms, protected funding, and where trust is low, third-party audit or international technical observation, raise the political cost of the silent veto.
The candid limit of all this is that these measures address coupling, not conflict. Where population shares are the constitution — where confessional or ethnic proportions determine the presidency, the premiership, and the speakership — no amount of statistical design will make counting safe, because the count is a direct challenge to the settlement. In those cases the sequence has to run the other way. The allocation rule must be changed first, in a direction that makes group size less decisive, and only then can the country afford to learn about itself. Lebanon cannot count until it deconfessionalizes; it cannot deconfessionalize without a political settlement that no one has been able to construct in ninety years. That is a hard conclusion, but it is the honest one, and it explains why exhortations to simply hold a census have failed there for the better part of a century.
X. Conclusion
The apparent paradox — that a manifest public good goes unproduced by states fully capable of producing it — dissolves once the census is understood correctly. It is not a photograph of a nation. It is the input to a machine that distributes power, and the people best positioned to block it are precisely the people who can calculate what it will cost them. Governments that cannot count their citizens are usually not ignorant of how many there are. They are aware that the answer is expensive, and that the expense will land on somebody in particular, and that the somebody has a veto.
The deeper matter is the one the Exodus provision gestures at. Enumeration is never a neutral act, because it presumes an authority to enumerate and a purpose for the result. The question that determines whether a census is possible is not whether the state has the enumerators. It is whether the people being counted, and the factions representing them, believe they can live with what the number will be used for. Where they cannot, the count does not happen — not for a decade, not for two, and in Lebanon’s case not for ninety-four years and running.
