Mapping Indian Districts Across Census Years, 1971-2001

The Indian states have been the standard unit of analysis for research on India that uses official data sources. For many empirical questions, states are a natural starting point because state governments set political agendas and budgets and administer a wide range of services. In addition, the boundaries of many states have been unchanged for over half a century and those of all major states were largely unchanged between 1971 and 2000. This stability has resulted in the relatively easy construction and use of panel data sets at the state level and these data have been used to ask a variety of questions relating to the e ectiveness of public policy. The use of more disaggregated district data allows the study of outcomes across regions with similar historical contexts and political regimes. States have an average of 20 districts, so district level panels can also be much larger. Most district-level studies however have relied on cross-sectional analysis because district comparisons over time are complicated by multiple boundary changes. Between 1971 and 2001, the number of districts increased from 356 to 593, a rise of about 67%. The purpose of this paper is to provide information on boundary changes across districts that will facilitate the construction of district-level panel data sets. We use population data from the state and central volumes of the Census of India to document changes in district boundaries between 1971 and 2001. For each decade during the 1971-2001 pe- riod, we classify districts into three categories: those with unchanged boundaries, those created by partitioning existing districts and nally, districts whose current boundaries were located in multiple districts at the time of the previous census. We nd that 136 of the 356 Indian dis- tricts in 1971 (38%) were unaected by boundary changes over the subsequent three decades, 79 districts (22%) were cleanly partitioned into multiple districts over the same period, and the 1 remaining 141 districts experienced more complex changes. Unchanged districts obviously pose no problem for the construction of panel data and the number of these districts can be quite large for short panels. For partitioned districts we provide population weights that permit the construction of panels using boundaries of either later or earlier census years as the base. For districts that are neither unchanged nor partitioned it is in general only possible to generate accurate population weights across adjacent census years. We provide these weights separately for the three periods: 1971-1981, 1981-1991, and 1991-2001. In addition, we amalgamate neigh- bouring districts into composite regions with unchanged boundaries between each census year and 2001. These composite regions, along with the unchanged and partitioned districts, give us the complete set of geographical units with unchanged boundaries between any census year and 2001. The following section provides details on data sources and our methods and compares these to those used by other studies relying on multi-year district data. Section 3 summarises some basic patterns. Section 4 concludes with some caveats on using our data and points to the type of work needed to construct district-level series over long time periods.