The Ethnographic Atlas covers more than 1200 pre-industrial societies but has been seen skeptically; paper documents positive associations between the Atlas & self-reported data from 790,000 individuals across 43 countries

Tabulated nonsense? Testing the validity of the Ethnographic Atlas. Duman Bahrami-Rad, Anke Becker, Joseph Henrich. Economics Letters, Volume 204, July 2021, 109880. https://doi.org/10.1016/j.econlet.2021.109880

Highlights

• We validate the Ethnographic Atlas, a popular anthropological database.

• We benchmark the ethnographic data with self-reports from survey respondents.

• Ethnographic data and contemporary self-reports are positively correlated.

• Our results provide evidence for the validity of ethnographic accounts.

Abstract: The Ethnographic Atlas (Murdock, 1967), an anthropological database, is widely used across the social sciences. The Atlas is a quantified and discretely categorized collection of information gleaned from ethnographies covering more than 1200 pre-industrial societies. While being popular in many fields, it has been subject to skepticism within cultural anthropology. We assess the Atlas’s validity by comparing it with representative data from descendants of the portrayed societies. We document positive associations between the historical measures collected by ethnographers and self-reported data from 790,000 individuals across 43 countries.

Keywords: Ethnographic AtlasValidationCulture

3.2 Results

Twelve domains are equivalently represented in the DHS and the Atlas: (1) patrilocality, (2) matrilocality, (3) polygyny, (4) reliance on animal husbandry, (5) reliance on agriculture, (6) length of post-partum abstinence, (7) breastfeeding duration, (8) insistence on virginity, (9) a preference for sons, (10) prevalence of domestic violence, (11) age difference between husband and wife, and (12) geographical location.2

Throughout, we find positive associations between the ethnographic information from the Atlas and the self-reported individual-level data from the DHS (Table 1). Columns (1) to (5) list the results for variables that capture different aspects of kinship organization and subsistence modes. Almost all associations are positive, statistically significant, and sizeable. For example, a one standard deviation increase in the prevalence of historical patrilocality is associated with a 0.8 percentage points increase in the likelihood that an individual lives patrilocally today. This amounts to about twelve percent of the unconditional probability

of living patrilocally in this sample (0.7). We can only speculate about the lack of association for reliance on agriculture, which could be due to differences in pre-industrial and contemporary agriculture, or the fact that the DHS variable captures only one specific aspect of contemporary reliance on agriculture.

Columns (6) to (11) list results for variables that capture social norms, customs, or preferences. Again, the associations between the historical and contemporary measures are positive throughout, in most cases statistically significant, and often meaningful in terms of size. For example, a one standard deviation increase in the historical length of post-partum abstinence is associated with a twelve percentage points increase in how long respondents today abstain after childbirth. For the preference of female virginity before marriage the association between the two measures is very small. This can plausibly be attributed to the lack of variation in the contemporary sample: about 93% of respondents express this attitude. Again, we can only speculate about the lack of association between the historical age of an infant at the onset of weaning in an ethnic group and the average breastfeeding duration of its descendants. It could be that male ethnographers could not make informed guesses about this dimension, or that breastfeeding practices have undergone substantial change during the past century.

Finally, we show that geographical location of the centroid of an ethnic group as reported by ethnographers is related to where people actually live today. For each individual in the DHS for whom we have information on geographical location, we calculate the distance in kilometers to the centroid of the homeland of her ancestral society. Figure 1 in the supplementary material shows the distribution. The median distance is 168 kilometers and a non-negligible fraction of about 12 percent live as close as 50 kilometers to the centroid of their ancestral homeland.

2  Table 2 in the supplementary material describes how these dimensions are measured in the Atlas and the DHS