A Note on Efficiency Gains from Multiple Incomplete Subsamples

11/10/2015 - 15:30
11/10/2015 - 16:30
Saraswata Chaudhuri, PhD (McGill University)
Purvis Hall, 1020 Pine Ave. West, Room 24 (McGill University)

We demonstrate efficiency gains in estimation by optimally using multiple subsamples all but one of which are incomplete following a monotone pattern. The finite dimensional parameter of interest is defined by moment restrictions on a target population which is some arbitrary union of the possibly different subpopulations for the multiple subsamples. A form of the missing at random (MAR) assumption is made for identification. MAR also makes the information contained in each incomplete subsample usable and thus contributes to efficiency gains. We show that the characteristics and possibility of such efficiency gains can be very different from those in the two subsamples contexts that have been studied extensively in the literature. Implication of these results on possible sampling strategies is briefly noted. We also show that a set of unconditional and conditional moment restrictions exhausts all the relevant information in the subsamples and can be easily used in a Frisch-Waugh-Lovell type sequential way, by virtue of monotonicity, for efficient estimation of the parameter of interest.



Saraswata Chaudhuri is an assistant professor of Economics. His research focuses on econometrics with micro-level data. He received his PhD in Economics from University of Washington and was an assistant professor at University of North Carolina before moving to McGill.

Web: http://saraswata.research.mcgill.ca/

Last edited by on Thu, 11/05/2015 - 11:18