Spatial Correlation in Household Choices in
Rural Indonesia*
John Gibson, Bonggeun Kim and Susan Olivia
Received 21 January 2011; Accepted 23 June 2011
Many household choices in developing economies are correlated with choices of
nearby households, because nearby locations have unobserved factors in common
and households and their neighbors interact. However, models recognizing these
spatial correlations are rarely used because few surveys give exact household
locations. In the present paper, we use unusual data from rural Indonesia, where
distances between households can be measured, to examine spatial effects in
equations for non-farm enterprises’ share of household incomes and food’s share of
total household budgets. Our results indicate that ignoring spatial correlations in
household choices might cause bias and inferential errors and could distort recom-
mended policy interventions aiming to raise living standards in rural Asia.
Keywords: spatial correlation, non-farm enterprises, food Engel curves, Indonesia.
JEL classification codes: C31, O17.
doi: 10.1111/j.1467-8381.2011.02063.x
I. Introduction
Many household choices are correlated with choices made by nearby households,
especially in developing countries where residential location strongly affects
economic activity. These correlations can arise either because nearby locations
share unobserved factors (e.g. access to markets or infrastructure quality) or
because of interaction between one household and another (e.g. coordination
problems when deciding to switch from farm to non-farm production). Empirical
studies of rural and regional development use several proxies to capture these
location effects, including distance to major population centers and markets, and
economic and population density, but a common belief in this literature is that
geographical dummy variables can capture all unobserved local factors (Jonasson
and Helfand, 2010).
Moreover, it is usually quite easy to estimate these geographic dummy variables
due to the structure of household surveys; samples are usually clustered in groups
of 10–20 rather than spread randomly across space. Because households in the
*Gibson (corresponding author): Department of Economics, University of Waikato, Private Bag
3105, Hamilton, New Zealand. Email: jkgibson@waikato.ac.nz. Kim: Department of Economics,
Seoul National University, 599 Gwanak-ro, Gwanak-gu, Seoul 151-746, Korea. Email: bgkim07@
snu.ac.kr. Olivia: Department of Econometrics and Business Statistics, Monash University, Clayton,
Victoria 3168, Australia. Email: Susan.Olivia@monash.edu.
Asian Economic Journal 2011,Vol. 25 No. 3, 271–289 271
© 2011 The Authors
Asian Economic Journal © 2011 East Asian Economic Association and Blackwell Publishing Pty Ltd