mixeddata

Mixeddata refers to data sets that contain variables of substantially different measurement scales, typically combining numerical variables (continuous or discrete) with categorical variables (nominal or ordinal). It is common in many domains such as healthcare, economics, and social sciences, where records include age and income alongside gender or education level. Mixeddata can also describe data with time-to-event components that include numeric time metrics and category indicators.

Descriptive summaries for mixed data differ by type: numeric variables use means, medians, and standard deviations, while categorical variables use frequencies and proportions. For analysis involving similarity and clustering, mixed data requires specialized distance measures such as the Gower distance, which normalizes across variable types and can handle missing values. Dimensionality reduction and visualization for mixed data often rely on methods designed for mixed types, such as Factor Analysis for Mixed Data (FAMD), which combines PCA for numeric variables with multiple correspondence analysis for categorical variables. When such specialized methods are not available, one-hot encoding followed by standard PCA is possible, though it can inflate dimensionality and complicate interpretation.

Imputation and modeling for mixed data also require approaches that accommodate multiple variable types. Model-based imputation and k-nearest neighbors methods using Gower distance are common, and certain algorithms—such as decision trees and tree ensembles—natively handle mixed data, while others require preprocessing. software and libraries exist to support mixed-data analysis, including R packages for FAMD and related methods, and Python implementations that offer similar functionality. Mixeddata analysis aims to extract structure, handle missing values, and support predictive modeling without forcing a single data type.