Cross-cultural databases for students, from the Ethnographic Atlas to D-PLACE

Suppose you want to know whether societies in which a bride moves in with her husband's family also tend to trace descent through men. One ethnography cannot tell you. You need the same question asked of hundreds of societies, answered in the same terms, with a record of where each answer came from.

Four resources do most of that work: George Peter Murdock's Ethnographic Atlas, the Standard Cross-Cultural Sample, the Human Relations Area Files and D-PLACE. They overlap, and students mix them up all the time. This piece sorts them out, reads one variable closely, and walks a small project from question to caveats.

Codes and texts

Three of these resources give you codes: a society, a variable and a value, such as "Hopi, residence after marriage, matrilocal". The fourth gives you the ethnographies themselves, indexed so you can find the paragraph behind a code. Good projects use both.

The Ethnographic Atlas

Murdock published the Atlas in installments in the journal Ethnology, starting with its first issue in January 1962, and gathered the results in a book in 1967. Additions and corrections followed, including J. P. Gray's corrected edition in 1999 and new installments on Siberia and easternmost Europe in 2004 and 2005. The version most people use, the one distributed through D-PLACE, describes 1,291 societies with over 90 variables. Kinship and marriage dominate: residence after marriage, descent groups, cousin marriage, payments at marriage. Subsistence, settlement, house building, class, inheritance and religion fill out the rest.

Each society carries an identifier such as Aa1, whose first letter places it in one of six world regions: Africa, the Circum-Mediterranean, East Eurasia, the Insular Pacific, North America and South America. Coverage is uneven. Kirby and colleagues note the Atlas's emphasis on Africa and North America, and in the D-PLACE file 698 of the 1,291 identifiers (54 percent) begin with A or N.

Every society is also pinned to a focal year, the period its codes describe. For 69 percent of Atlas societies that year falls between 1900 and 1950. The Atlas portrays societies as ethnographers described them then, not how those communities live now.

The Standard Cross-Cultural Sample

Murdock and Douglas White published the SCCS in Ethnology in 1969. They took the best-described society from each of 186 cultural provinces of the world, aiming for cases that were as independent as possible in history and in borrowing. Each is pinned to a specific subgroup and date. Because 186 cases are few enough to code completely, scholars have kept adding variables for decades. A revised edition of the 1969 paper counts more than 2,000 variables from nearly a hundred studies, and D-PLACE's copy holds 1,781 of them.

The sample was built as an answer to Galton's problem, the worry that neighboring and related societies are not independent data points.

eHRAF World Cultures

The Human Relations Area Files began in 1949 as a consortium of universities based at Yale. Its database, eHRAF World Cultures, holds full ethnographic texts that HRAF anthropologists have indexed by subject, paragraph by paragraph. The index is the Outline of Cultural Materials, a classification Murdock began in the 1940s, with more than 90 major categories and more than 700 subcategories. Category 750 is sickness, for instance, and 753 is theory of disease. Search a code across cultures and you get the passages themselves.

HRAF also keeps its own sample, the Probability Sample Files: one well-documented culture drawn at random from each of 60 culture areas (Lagacé 1979). Access to eHRAF requires membership, usually through a university library. Carol Ember's (2007) concordance matches HRAF cultures to SCCS and Atlas societies, which is how you get from a code to the pages behind it.

D-PLACE

D-PLACE, the Database of Places, Language, Culture and Environment, was introduced by Kathryn Kirby and colleagues in PLOS ONE in 2016. It is less a dataset than a set of connections. The first version linked the Atlas and Lewis Binford's hunter-gatherer dataset (339 societies, 40 variables) to three other kinds of information:

• Language. Each society is matched to a language or dialect with a Glottolog code, which places it in a language family tree, and for some families on a dated phylogeny.
• Place. Each society has a focal location and focal year, and coded values link, where possible, to the ethnographic sources the coders used, more than 4,000 of them.
• Environment. Climate, plant productivity, species counts, altitude and ecoregion were sampled at each location, where possible for the years 1901 to 1950 to match the focal years.

D-PLACE now also distributes the SCCS codes and J. G. Jorgensen's Western North American Indians dataset (172 societies, 429 variables). Everything can be shown on a map or a tree and downloaded as a spreadsheet.

Reading one variable before you use it

Take residence after marriage, which the Atlas codes as variable EA012. The codebook separates patrilocal residence, where a couple lives with or near the husband's male kin who form a localized patrilineal group, from virilocal residence, where the couple lives with the husband's kin but those kin are not gathered into such a group. Matrilocal and uxorilocal make the same split on the wife's side. Avunculocal means living with the husband's mother's brother (MB, in the kin-type notation where M is mother and B is brother), and there are ambilocal, neolocal and mixed categories besides.

Counting the codes in D-PLACE's release of the Atlas gives this:

These are counts of societies in the Atlas, with its regional lean, not a census of humanity. And if you merge "patrilocal" and "virilocal", as many projects do, say so and say why.

Descent shows a subtler trap. The variable "Descent: major type" (EA043) gives 590 patrilineal, 362 bilateral and 160 matrilineal societies, plus smaller categories and 17 missing. But its own note says it is not in the original Atlas. D-PLACE derives it from three other variables about patrilineal, matrilineal and cognatic kin groups. That is a reasonable shortcut, but you should know when you are using a constructed variable.

A small project, step by step

The numbers in this section are invented for teaching. They are not Atlas results.

1. Write the question and the coding rules before you look at any data. Here: residence "with the husband's kin" means patrilocal or virilocal; descent "patrilineal" means code patrilineal and nothing else.
2. Draw the cases. Imagine a class exercise that takes 30 Atlas societies at random by identifier. Three have no descent code, which leaves 27.
3. Make the table.

Patrilineal descent appears in 14 of 18 societies with residence among the husband's kin (77.8 percent) and in 2 of 9 others (22.2 percent). The odds ratio is (14 × 7) / (4 × 2) = 98 / 8 = 12.25.

4. Look at where the cases come from. Download the language family column with the codes. Suppose 12 of the 27 societies belong to one family, and 11 of those 12 sit in the top left cell. Set that family aside and the remaining 15 societies give 3 of 6 versus 2 of 9, an odds ratio of (3 × 7) / (3 × 2) = 21 / 6 = 3.5. The association keeps its direction but now rests on very few cases. Our statistics starter covers the tests you would run next.
5. Check the codes against the texts. Kirby and colleagues recommend recoding a random sample of societies from the original ethnographies. Say you recode residence for 20 of your societies from passages found in eHRAF:

You agree on 17 of 20, so observed agreement is 0.85. But two coders who assigned categories at random, at these same rates, would agree fairly often too. Cohen's kappa corrects for that. Expected chance agreement is (13/20 × 12/20) + (7/20 × 8/20) = 0.39 + 0.14 = 0.53, and

κ = (0.85 − 0.53) / (1 − 0.53) = 0.32 / 0.47 ≈ 0.68

That is solid agreement but far from perfect. Then read the three disagreements one by one. Often the culprit is time: the ethnography you read describes a different decade, or a different village, from the Atlas's focal year and place.

Caveats that travel with every code

• A code describes a time and a place. Kirby and colleagues found Atlas and Binford entries for the same society whose focal years differ by more than 50 years, as with the Chumash (1800 in the Atlas, 1860 in Binford).
• "Society" is a choice. Murdock's Tunava include two Paiute groups, those of Deep Springs Valley and Fish Lake Valley, which Binford treats as separate societies. D-PLACE flags such likely matches with a shared cross-dataset identifier but leaves the decision to combine them to you.
• Different coders, different communities. For the Pumé of Venezuela, the Atlas, working from a description of Pumé on the Cinaruco River, puts shifting agriculture at around 40 percent of subsistence. Binford's dataset, reflecting work with savanna Pumé, records hunting, gathering and fishing at 6, 41 and 53 percent of subsistence. Neither is wrong. They describe different groups.
• Sources have biases. What an ethnographer noticed and whom they talked to shape the code. Watts and colleagues (2022) argue for recording facts about each source alongside the codes, so such biases can be checked.
• Some variables come with warnings. The codebook entries for inheritance of land and of movable property repeat Murdock's own 1963 caution that those codes needed serious revision and should be used "only with circumspection". Read the notes before you read the numbers.
• Names change. D-PLACE replaced society names considered pejorative with preferred names and keeps the alternatives searchable, so try more than one name.

Where to start

For a class paper, start in D-PLACE. Pick the SCCS for fewer, more independent cases; pick the Atlas for kinship and marriage codes on many societies. Read the codebook entry for every variable, keep the language family and focal year columns, and use eHRAF to read the passages behind a handful of codes. Kirby and colleagues ask users to cite both the original dataset and the D-PLACE paper, with the date of access. A methods line might read: residence and descent codes from the Ethnographic Atlas (Murdock 1967; Gray 1999), accessed through D-PLACE (Kirby et al. 2016) at d-place.org on the day you downloaded them. Related pieces sit under Methods & Tools.

Sources and further reading

• Murdock, G. P. (1967). Ethnographic Atlas. Pittsburgh: University of Pittsburgh Press. The installments began in Ethnology 1(1), 1962: jstor.org/stable/3772932
• Gray, J. P. (1999). A corrected Ethnographic Atlas. World Cultures 10(1): 24-85.
• Murdock, G. P., and White, D. R. (1969). Standard cross-cultural sample. Ethnology 8(4): 329-369. jstor.org/stable/3772907
• Lagacé, R. O. (1979). The HRAF probability sample: retrospect and prospect. Behavior Science Research 14(3): 211-229. doi.org/10.1177/106939717901400304
• Ember, C. R. (2007). Using the HRAF collection of ethnography in conjunction with the Standard Cross-Cultural Sample and the Ethnographic Atlas. Cross-Cultural Research 41(4): 396-427. doi.org/10.1177/1069397107306593
• Kirby, K. R., Gray, R. D., Greenhill, S. J., Jordan, F. M., et al. (2016). D-PLACE: a global database of cultural, linguistic and environmental diversity. PLOS ONE 11(7): e0158391. doi.org/10.1371/journal.pone.0158391; data at d-place.org and, for the Atlas release counted here, doi.org/10.5281/zenodo.17602181
• Watts, J., Jackson, J. C., Arnison, C., Hamerslag, E. M., Shaver, J. H., and Purzycki, B. G. (2022). Building quantitative cross-cultural databases from ethnographic records: promise, problems and principles. Cross-Cultural Research 56(1): 62-94. doi.org/10.1177/10693971211065720
• Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological Measurement 20(1): 37-46. doi.org/10.1177/001316446002000104