Comparison of log-ratio and log10 chemical elemental data analysis of Central Amazonian pottery and archaeological implications

Hazenfratz, Roberto; Mongeló, Guilherme Z.; Munita, Casimiro S.; Neves, Eduardo G.

doi:10.1007/s12520-024-01965-y

Cited by 2 publications

(1 citation statement)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is because the ALR is defined as log(X/ref) = log(X) − log(ref), for all X ̸ = ref, so if log(ref) is almost a constant, then one can interpret the ALRs effectively as log(X), which is easier than interpreting logratios. This also means that if there is any part for which its log-transform is almost constant (i.e., has very low variance), then the analysis of log-transformed data (not logratio-transformed) is sufficient for unsupervised learning if, in addition, the logratio structure is closely approximated (i.e., close to isometry) (Greenacre et al, 2021;Hazenfratz et al, 2024). This may be particularly appropriate for recipes used to make objects in which the major components do not appear to change significantly over time, but variation in their trace components may identify differences in their provenance and/or production.…”

Section: Variable Selectionmentioning

confidence: 99%

A Comprehensive Workflow for Compositional Data Analysis in Archaeometry, with Code in R

Greenacre,

Wood

2024

Preprint

View full text Add to dashboard Cite

Compositional data, which have relative rather than absolute meaning, are common in quantitative archaeological research. Such multivariate data are usually expressed as proportions, summing to 1, or equivalently as percentages. We present a comprehensive and defensible workflow for processing compositional data in archaeometry, using both the original compositional values and their transformation to logratios. The most useful logratio transformations are illustrated and how they affect the interpretation of the final results in the context of both unsupervised and supervised learning. The workflow is illustrated on compositional data from bronze ritual vessels to provide compositional fingerprints for the Shang and Zhou periods of the Chinese Bronze Age. Predictions, with caveats, of the fabrication age of the vessels are made from the compositional data -- in effect, compositional rather than typological seriation of the bronzes. In the Supplementary Material, we further explore the effect of zeros in the dataset and compare logaratio analyses with the chiPower approach, where we replace any value in the original data determined as being below the detection limit of the instruments for the element, with zeros. The data and R code for reproducing all the analyses are provided both in the Supplementary Material and online.

show abstract

Section: Variable Selectionmentioning

confidence: 99%