Virtual methods for studying human remains are becoming increasingly popular in bioarchaeology, and the rate of technological innovation in the last few years has been such that we now have multiple options to choose from when collecting data. This raises the question of whether datasets generated with different methods are transposable. In the study reported here, we investigated whether it is valid to combine 3D data obtained with a MicroScribe digitising arm and 3D data collected via photogrammetry. We did so by simulating a population-based analysis similar to those commonly undertaken in bioarchaeology. Our sample comprised 19 crania from two ethnic groups, Ancient Egyptians and Guanches, and the landmarks we employed pertained to facial shape.

The analyses yielded several findings. First, we found that photogrammetry was significantly more precise than the MicroScribe digitising arm. Second, the photogrammetry-based method revealed the existence of facial shape differences between the two ethnic groups that were not captured by the MicroScribe-based method. Third, we found that the two methods did not consistently capture the same facial shapes—they did for one of the ethnic groups but not for the other. Fourth, the analyses indicated that using the two methods can result in ethnic group-level differences in facial shape when they are applied to individuals from a single ethnic group. Lastly, the two methods of data collection yielded different patterns of variation in facial shape. Together, these findings suggest that combining 3D landmark coordinates collected with a MicroScribe and those obtained via photogrammetry may introduce considerable error into an analysis, and, consequently, bioarchaeologists should be cautious about doing so.

The sample comprised 19 crania from two different ethnic groups: Pre-Spanish Guanches from the Canary Islands (n=11) and Ancient Egyptians (n=8). Collected in the 19th century, the crania are currently curated at the University of Edinburgh’s Anatomical Museum. Individuals were recorded as female or male based on the original collection records (i.e. we did not sex them ourselves). Eight of the individuals were female and 12 were male. Only adult crania were included in the sample in order to avoid the confounding effects of ontogeny; individuals were judged to be adult on the basis of dental eruption. Further information about the sample can be found in the Supplementary Information.

The specific photogrammetry method we used was outlined by Evin et al. (2016). We took 150 photographs of each cranium with an eight megapixel digital single-lens reflex Canon EOS 77D camera and a 50mm lens. With the cranium placed on a rotating table, the photographs were shot at intervals of approximately 10°. Following Evin et al. (2016), a 3D scale was employed as a target marker. We then used Agisoft Metashape (Agisoft, 2019) to create a 3D model of the cranium from the photographs. We aligned all the photographs with the accuracy level set to ‘high’, and then produced a depth map, again with the accuracy level set to ‘high’. Next, we used the depth map to create a mesh. Thereafter, we exported the mesh as a 3D model (.ply). Lastly, the 3D model was imported into MorphoDig (Lebrun 2018) and the 3D Cartesian coordinates of 13 facial landmarks captured twice (Figure 1). All landmarks were Type I landmarks, according to Bookstein’s (1997) widely used scheme. We selected Type I landmarks because they are the most reliable type of landmark (Bookstein, 1997) and thus should maximise the probability of the two methods yielding data that are statistically indistinguishable.

The digitising arm we used was a MicroScribe MLX (https://gomeasure3d.com/microscribe/) with a standard stylus tip. The landmarks we recorded were the same as the ones we captured on the photogrammetry-derived models (Figure 1). A single observer (KAP) operated the MicroScribe to avoid the problem of inter-observer error. KAP has considerable experience collecting data with a MicroScribe (e.g. Plomp et al. 2013, 2015, 2019ab, 2020, 2021ab). As with the photogrammetry, the coordinates of the landmarks were collected twice.

We employed the dataset in five analyses. In the first, we investigated whether the landmarks recorded with the two methods of data collection had the same level of precision. To accomplish this, the entire dataset was subjected to generalised Procrustes analysis (GPA), which converts the Cartesian coordinates for a given landmark configuration into Procrustes coordinates by removing translational and rotational effects and scaling the landmark configuration to centroid size (Slice 2007). The Procrustes coordinates are provided in this dataset.