The school in the village goes to class ten. Class eleven and twelve — what everybody calls plus two — are in the town, which is two and a half hours by bus when the road is open and is not always open between June and September.
So at sixteen, every student who intends to continue leaves. Not the best ones. All of them.
I have written about what happens when the top of a class is extracted by a scholarship. This is the same shape at a hundred times the scale and it is entirely unremarked, because it is not a policy, it is a fact about where buildings are.
The decision made at sixteen is described, by everyone involved, as an education decision. It is a migration decision. A family is choosing whether to fund a room in the town, or a hostel, or a relative's floor, for two years, at a cost that for most households is comparable to a wedding. And the two years are not two years. A child who has lived in the town from sixteen to eighteen has friends there, a coaching centre there, a sense of what a day is like there. Almost none come back to live. The plus two decision is where the village population is actually determined, about twenty years in advance.
There is an economy around this that nobody counts. In the town, rooms are let by the month to fifteen- and sixteen-year-olds, four to a room, with a landlady who provides food and a degree of supervision that varies enormously. The tuition centres are built around exactly this population. It is a functioning market and it is not predatory in the main — I lived in one of those rooms and the woman who ran it was kind to me and is still kind to me — but it is a market in the housing of children, arranged privately, regulated by nobody, and invisible to every education statistic collected.
And it does not treat everyone the same.
A son in a room in the town is a normal arrangement. A daughter in a room in the town is a negotiation, and it is the negotiation that decides a very large number of girls' education in this country, and it is not conducted on educational grounds. It is conducted on the grounds of what will be said, whether there is a relative in that town, whether the landlady is known to somebody, and what her brother thinks. I have watched two families make opposite decisions about equally capable daughters on the basis of whether an aunt happened to live within walking distance of the campus.
That is the mechanism. Not fees, not attitudes about girls' education in the abstract — most of these families want their daughters educated and will say so — but the presence or absence of an aunt.
The counterargument, and it is the one the district makes, is that you cannot put a plus two in every village. That is true. The enrolment does not support it, the science stream needs a laboratory and a teacher who exists, and a badly staffed plus two in the village would be worse than a well-staffed one in the town, and I would not send my own students to it.
So I am not asking for a building in every village. I am asking for the thing that the private market has already proved is the binding constraint, which is a place to sleep. A district that built two hostels — one for girls, run properly, with a warden and a published fee — would change the plus two decision for more households than any scholarship scheme it has ever run, and it would do it by removing the only part of the calculation that is not about the child.
I said this at a district education meeting last year. Somebody agreed with me and then explained that hostels come under a different budget line and that the education office does not build residential buildings.
I have twenty-two students in my Kathmandu classroom this year and nineteen of them left home at sixteen. I asked, once, out of curiosity, how many had gone back for Dashain. Eleven. How many expected to live there again: one, and she said it in a way that suggested she was reporting her mother's expectation rather than her own.

The presence or absence of an aunt. That is the variable, it is not in any dataset, and it decides more girls' education than any scheme I have seen evaluated.