RESEARCH METHOD
Multivariate analysis
PCA, proxy relationships, categories and major gradients in complex data.
01
What the method shows
Covariation among proxies, sample grouping, major gradients and relationships among categories or spatial units.
How we use it in HEL
For standardised tables with enough samples and well-defined variables; PCA and tests are interpreted with archaeological context.
02
How the method works
Multivariate analysis searches for structure in datasets where each sample carries many variables. PCA transforms correlated variables into new orthogonal axes that successively capture as much of the dataset variance as possible.
Scores locate samples in this new space, while loadings show which variables define each axis. This can reveal clusters, gradients and covarying elements that would be difficult to see across many separate plots.
PCA does not explain causation and is sensitive to scaling, transformation and outliers. Statistical structure must be returned to the archaeological question and spatial context.
GIS, geochemistry, robust control groups and follow-up tests of explicit hypotheses.
03
In practice
- Input / materialclean numerical table plus contextual categories
- Scale / resolutionsample, profile, site and comparative dataset
- Sampling / preparationdepends on the primary method; metadata and replication are critical
- Typical outputPCA scores/loadings, boxplots, post-hoc tests and other ordinations
04
Limits and control
PCA and statistical differences are not causal explanations by themselves; scaling, compositional structure, missing values and group size must be checked.
Třebokov: PCA and spatial analysis of multivariate geochemical signals.
05
References and HEL studies
Selected studies underpinning the method descriptions and examples used by Human Earth Lab.