Estimating Deaton's unit-value model: the duvm Stata module

duvm estimates the unit-value model of Deaton (1988, 1990, 1997): own- and cross-price elasticities of quantity, expenditure elasticities and quality elasticities for a group of goods, from the budget shares and unit values that households report, using the spatial variation of prices across clusters. The estimator is Deaton's two-stage procedure, closed-form and written in Mata: within-cluster regressions, between-cluster regressions of the cluster averages with the errors-in-variables correction, the quality parameter derived from weak separability, completion of the system with a composite good, and an approximate Slutsky symmetry. duvm adds what the published programs lack: sampling weights with the eective cluster sizes they imply; a linearized variance built from the inuence functions of every rst- and secondstage moment, including the generated-regressor terms of the rst stage, combined through the analytic Jacobian of the chain from the moments to the elasticities, and aggregated by cluster or by survey design (svyset: PSUs, strata, nite-population correction); a design bootstrap of both stages; a diagnostic of the price signal before estimation; standard e() results and estat. Every variance is validated against the bootstrap under simple random sampling of clusters, sampling weights and a stratied design: the linearized standard errors are within 13 percent of the bootstrap for all 115 parameters. An option compat reproduces the Stata code published with Deaton (1997), checked against a line-by-line transcription of his programs; Appendix A records the three places where that code departs from the book and the eect of each.