On the Use of Biomineral Oxygen Isotope Data to Identify Human Migrants in the Archaeological Record: Intra-Sample Variation, Statistical Methods and Geographical Considerations

Abstract

Oxygen isotope analysis of archaeological skeletal remains is an increasingly popular tool to study past human migrations. It is based on the assumption that human body chemistry preserves the δ18O of precipitation in such a way as to be a useful technique for identifying migrants and, potentially, their homelands. In this study, the first such global survey, we draw on published human tooth enamel and bone bioapatite data to explore the validity of using oxygen isotope analyses to identify migrants in the archaeological record. We use human δ18O results to show that there are large variations in human oxygen isotope values within a population sample. This may relate to physiological factors influencing the preservation of the primary isotope signal, or due to human activities (such as brewing, boiling, stewing, differential access to water sources and so on) causing variation in ingested water and food isotope values. We compare the number of outliers identified using various statistical methods. We determine that the most appropriate method for identifying migrants is dependent on the data but is likely to be the IQR or median absolute deviation from the median under most archaeological circumstances. Finally, through a spatial assessment of the dataset, we show that the degree of overlap in human isotope values from different locations across Europe is such that identifying individuals' homelands on the basis of oxygen isotope analysis alone is not possible for the regions analysed to date. Oxygen isotope analysis is a valid method for identifying first-generation migrants from an archaeological site when used appropriately, however it is difficult to identify migrants using statistical methods for a sample size of less than c. 25 individuals. In the absence of local previous analyses, each sample should be treated as an individual dataset and statistical techniques can be used to identify migrants, but in most cases pinpointing a specific homeland should not be attempted.

Fig 1. Comparison of the different measures of scale for each site in the Post-Infant Dentition data-subset plotted by site sample size, for all sites where N>4.

On the left, (a) the statistic value for the range, SD, IQR and MAD_norm and MAD_Q3. On the right, (b) the spread encompassed by the range, and the boundary values of the methods of 2SD, 1.5IQR, 3MAD_norm and 3MAD_Q3. Criteria of each method (2SD, 1.5IQR, 3MAD_norm, 3MAD_Q3) as defined in the text.

Fig 2. Histogram of the difference between each individual’s oxygen isotope value and (a) the site mean oxygen isotopic value or (b) the site median oxygen isotopic value, for all specimens from sites where N>4 in the Post-Infant Dentition data-subset.

Fig 3. The proportion of outliers found at each site using the five different outlier identification methods, plotted by site sample size, for all sites where N>4 in the Post-Infant Dentition data-subset.

For some of methods, there are multiple apparent curves for sample sizes between 5 and 25. This is an artefact of producing estimates of either one, two or three outliers within the sample, producing “curves” of 1/x where x is the number of outliers (i.e. 1, 2 or 3).

Fig 4. Density plot of each individual’s oxygen isotope value expressed as a z-score for all specimens from sites where N>4 in the Post-Infant Dentition data-subset, with a fitted normal distribution curve overlaid.

Z-scores as defined in the text.

Fig 5. Map of all sites represented in the Post-Infant Dentition data-subset.

Map created using ArcGIS version 10.2.2 for PC. Source: US National Parks Service.

Fig 6. Map of all European sites represented in the Post-Infant Dentition data-subset, with regional modern precipitation oxygen isotopic values indicated.

Map created using ArcGIS version 10.2.2 for PC. Precipitation data taken from Bowen 2015 [64] and Bowen and Revenaugh 2003 [65].

Fig 7. Histogram of oxygen isotope values for all European individuals in the Post-Infant Dentition data-subset.

Fig 8. Oxygen isotope values for all individuals from European sites in the Post-Infant Dentition data-subset, plotted by: (a) and (b) country, (c) latitude, (d) longitude, (e) altitude, (f) modern precipitation oxygen isotopic values (δ¹⁸O_MAP).

(c) Latitude is positively correlated with δ¹⁸O_PO4: r_s = 0.246, P<0.001, N = 1266. (d) Longitude is negatively correlated with δ¹⁸O_PO4: r_s = -0.272, P<0.001, N = 1266. (e) Altitude is weakly negatively correlated with δ¹⁸O_PO4: r_s = -0.092, P<0.005, N = 1266. (f) δ¹⁸O_MAP is not correlated with δ¹⁸O_PO4: r_s = 0.019, P = 0.494, N = 1266.

Fig 9. The limits of the ‘local’ signal of the different outlier identification methods for European sites, calculated: (a) by country, (b) by lat/long grid square, (c) by altitude, (d) by site-specific modern precipitation oxygen isotopic values (δ¹⁸O_MAP).

Plotted for groupings with N>4. Criteria of each method (2SD, 1.5IQR, 3MAD_norm, 3MAD_Q3) as defined in the text. (a) Full data in S4 Table, the countries are statistically different: H(8) = 297.67, P<0.001. (b) Full data including alphanumeric codes in S5 Table, the grid squares are statistically different: H(12) = 312.50, P<0.001. (c) Full data in S6, data binned in 150m intervals, the altitudinal groups are statistically different: H(4) = 155.94, P<0.001. (d) Full data in S7 Table, δ¹⁸O_MAP data binned in 1‰ intervals, the precipitation oxygen isotope groups are statistically different: H(3) = 93.78, P<0.001.

Fig 10. Histograms of data in the Post-Infant Dentition data-subset from the sites of Ban Non Wat, Kaminaljuyu and Velim-Velištak.

Data from: Ban Non Wat—King et al. 2013; Kaminaljuyu—White et al. 2000, Wright et al. 1998, Wright et al. 2010; Velim Velištak—Lightfoot et al. 2014. Note that specimens from Kaminaljuyu have been analysed in multiple studies.