An Optimal Spatial Sampling for Demographic and Health Surveys

44 Pages Posted: 15 Apr 2011

See all articles by Naresh Kumar

Naresh Kumar

University of Miami

Dong Liang

University of Iowa

Marc Linderman

University of Iowa

Jin Chen

Zhejiang University

Date Written: April 13, 2011

Abstract

This paper presents an optimal spatial sampling (OSS) design for fielding the demographic and health. The proposed design (a) develops a context specific sampling frame at a fine spatial resolution, (b) captures maximum spatial autocorrelation-controlled semivariance in the selected attribute (a composite index of population concentration and socio-economic characteristics in the context of this paper) of the sampling domain, (c) ensures spatial coverage and representation, (d) minimizes sample size, and (e) minimizes redundancy in the selection of sample sites. OSS was tested for drawing a sample for fielding a pilot General Social Survey (GSS) in Chicago metropolitan area (MSA) in the summer of 2010.

Fine resolution LandScan population data, coupled with the U.S. Census data, were used to develop a multivariate contextual sampling frame. Our analysis suggests that a set of 97 sample sites captured 80% of the total spatial autocorrelation-controlled semivariance in the composite index used for optimizing sample sites. Maximizing spatial autocorrelation-controlled semivariance using OSS also ensured representation of the population variance.

The OSS design outperformed other widely-used spatial sampling designs, such as Generalized Random Tessellation Stratified sampling (GRTS) in terms of spatial coverage and population representation. The domain (or area) of each optimal site, defined using the extent of local spatial autocorrelation, serves as a stratum and formulates bases for drawing inferences. The simulation experiment suggests that the relative efficiency of the OSS was better than that of other sampling designs. However, for a skewed quantity the efficiency of OSS drops and prediction bias (measured by percent difference between observed and predicted mean) increases. Therefore, it is important that the variable used for optimization of sample sites is normalized to achieve the best performance of the OSS.

Various methods, including reverse geocoding, can be used to develop enumeration list and draw respondent(s) from each stratum. Geocoding respondent is also useful for the collection of multi-layer socio-physical contextual data at reduced cost. This, in turn, is likely to extend the scope of the survey data to a multi-level, interdisciplinary setting.

Keywords: spatial sampling, sampling frame, geospatial analysis, socio-physical contexts, GIS, geospatial technologies

JEL Classification: C42, C21, C8

Suggested Citation

Kumar, Naresh and Liang, Dong and Linderman, Marc and Chen, Jin, An Optimal Spatial Sampling for Demographic and Health Surveys (April 13, 2011). Available at SSRN: https://ssrn.com/abstract=1808947 or http://dx.doi.org/10.2139/ssrn.1808947

Naresh Kumar (Contact Author)

University of Miami ( email )

1425 NW 10 Ave, Suite 308
Epidemiology and Public Health
Coral Gables, FL 33136
United States
305-243-4854 (Phone)
305-243-5577 (Fax)

HOME PAGE: http://eph.ccs.miami.edu

Dong Liang

University of Iowa ( email )

341 Schaeffer Hall
Iowa City, IA 52242-1097
United States

Marc Linderman

University of Iowa ( email )

341 Schaeffer Hall
Iowa City, IA 52242-1097
United States

Jin Chen

Zhejiang University ( email )

38 Zheda Road
Hangzhou, Zhejiang 310058
China

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
206
Abstract Views
1,912
rank
188,166
PlumX Metrics