Subnational diversity in Sub-Saharan Africa: Insights from a new dataset

Abstract

This paper presents a new dataset on subnational ethnolinguistic and religious diversity in Sub-Saharan Africa covering 36 countries and almost 400 first-level administrative units. We use population censuses and large-scale household surveys to compile detailed data on the ethnolinguistic composition of each region and match all reported ethnicities to Ethnologue, a comprehensive catalog of world languages. This matching allows us to standardize the notion of an ethnolinguistic group and account for relatedness between language pairs, a correlate of shared history and culture, when calculating diversity indices. Exploiting within-country variation provided by our new dataset, we find that local public goods provision, as reflected in metrics of education, health, and electricity access, is negatively related to ethnolinguistic diversity, but only if the underlying basic languages are first aggregated into larger families or if linguistic distances between groups are taken into consideration. In other words, only deeprooted diversity, based on cleavages formed in the distant past, is strongly inversely associated with a range of regional development indicators. Furthermore, we show that subnational diversity has been remarkably persistent over the past two-three decades implying that population sorting in the short to medium run is unlikely to bias our main findings.

Measuring Regional Ethnolinguistic Diversity in Sub-Saharan Africa: Surveys vs. GIS

“…A standard approach to constructing any dataset on regional sociodemographic characteristics is to compile the information from relevant population surveys. In Gershman and Rivera (2018), we introduce and describe the product of a comprehensive effort to compile the data on subnational ethnolinguistic composition of Sub-Saharan Africa based on this approach. Our dataset contains 36 countries, almost 400 first-level administrative units, and 750 unique ethnolinguistic groups.…”

“…Neither of these conditions is expected to hold perfectly since both key inputs of the exercise are only approximations. However, given the increasing popularity of GIS applications in social sciences and the simplicity of implementing this approach, it is instructive to compare its outcomes to those of the survey-based method in Gershman and Rivera (2018).…”

“…The final population grid composed of 30 arc-second cells (approximately 1 square kilometer near the equator) is produced by disaggregating census data within administrative boundaries via a "smart interpolation" technique that employs spatial data and imagery analysis. For consistency, we use the same set of administrative boundaries as in Gershman and Rivera (2018).…”

“…Due to their direct link to Ethnologue's classification system, the WLMS languages are readily comparable to those in Gershman and Rivera (2018) and can be similarly aggregated at the more basic levels of the linguistic family tree, as discussed above. Interestingly, as shown in panel (d) of Figure 2, the survey-and WLMS-based ELF indices are much more strongly connected at higher levels of linguistic aggregation, with the correlation coefficient hovering around 0.85 for levels 18 in the rural sample.…”

“…This paper examines two approaches to constructing such a dataset for a large sample of regions within Sub-Saharan African countries. First, we briefly discuss the contribution of Gershman and Rivera (2018) who employ censuses and large-scale household surveys to measure subnational ethnolinguistic diversity. Second, we show how the GIS approach, based on combining digital maps of ethnolinguistic groups with disaggregated population data, may be used to accomplish the same goal.…”