The Comparison of Log-Ratio Analysis and Correspondence Analysis Applied to Compositional Data
Compositional data is typically analyzed using log-ratio analysis (LRA). LRAs biggest weakness is its inability to handle data with missing values. Correspondence analysis (CA) is an alternative method to analyzing compositional data, and is a method that can handle sparse data (Greenacre, 2019). This study applied LRA and CA to data from the American Time Use Survey (US Bureau of Labor Statistics, 2020) with the goal of finding similar gender differences in both analyses. It was hypothesized that both methods will produce similar results and provide support in the future use of correspondence analysis for compositional data. The LRAs (weighted and unweighted) and CA both showed clear gender differences in daily time use. Males spent more time doing working activities and leisure/sport related activities. Females spent more time caring for household members, communicating via telephone, mail, or email, and doing household activities. A Procrustes analysis was used to assess the agreement between the weighted LRA and CA and found a Procrustes correlation of 0.99. This indicates that the analyses had high agreement about the gender differences in the ATUS. Future research should be done to assess the agreement between LRA and CA when applied to sparse data.
Cook-Thibeau, Hayley, "The Comparison of Log-Ratio Analysis and Correspondence Analysis Applied to Compositional Data" (2021). ETD Collection for Fordham University. AAI28718226.