Coverage

A psephologist attacks this before anything else, so it sits on screen rather than in a footnote. A mobile frame in India systematically under-covers older rural women, the poorest households, and anyone whose handset is shared and controlled by someone else.

Three mechanics answer it. Numbers are drawn by random digit dialling over allocated blocks rather than a purchased list, because a bought list is a convenience sample wearing a probability costume. Constituency is self-reported and verified, never inferred from the prefix, because portability and migration broke that link years ago. And on a shared handset the instrument asks for the adult with the next birthday and calls back for them, which is slower and is the difference between a representative sample and one that is male by nine points.

Weighting can lift a cell that is under-represented. It cannot lift a cell that is empty. For the weakest cells a small face-to-face probability sample runs alongside and calibrates, roughly four per cent of interviews carrying a disproportionate share of the credibility.