Visualization of Data Envelopment Analysis Problems.

deaviz

High-dimensional visualization methods for data envelopment analysis (DEA), gathered into a single R package. It brings together techniques that have appeared in the literature but remained scattered and largely unimplemented: cross-efficiency matrix unfolding, the Porembski reference-set graph, PCA biplots, multidimensional-scaling colour-plots, self-organizing maps, the Costa bi-dimensional efficient frontier, parallel coordinates, radar charts, panel-data trajectory biplots, peer and reference networks, and a set of descriptive plots.

The methods follow the author's PhD thesis (Ashkiani, 2019).

Design

Everything is built around one validated object, dea_data, which every function accepts (plain data frames are coerced automatically). Functions follow a consistent naming convention plot_<source>_<method>:

• compute_* — analysis, returns values (efficiency scores, cross-efficiency matrices, multiplier weights, a fitted SOM).
• plot_io_* — plots built from inputs/outputs (the io).
• plot_cem_* — plots of the cross-efficiency matrix (the cem).
• plot_efficiency_* — plots of the efficiency scores.
• plot_panel_io_* — plots of panel (long-format) input/output data.

Most plotting functions share the same four arguments:

• interactive — FALSE (default) returns a static ggplot2 object; TRUE returns an interactive plotly object.
• labels — TRUE (default) shows the identifying text labels for that plot (DMU names on points and networks, DMU ticks on heatmaps, variable names on distributions and radar, and so on).
• title — an optional plot title.
• ... — passed through to the underlying geom (or to the relevant compute_* function for the plots that take raw data).

A shared colour scheme is used throughout: the colour-blind-safe Okabe–Ito palette for categories and viridis for continuous scales.

The Benchmarking package is the DEA engine. The modelling and plotting packages are in Suggests; each function checks for the ones it needs and errors informatively if they are missing.

Installation

# install.packages("devtools")
devtools::install_local("deaviz")          # from the package directory

# install the suggested engines you intend to use, e.g.:
install.packages(c("Benchmarking", "ggplot2", "plotly", "lpSolve",
                   "smacof", "kohonen", "MASS", "ggrepel"))

Quick start

library(deaviz)

# Build the data object (the example data is not i_/o_ prefixed,
# so name the inputs and outputs explicitly)
d <- dea_data(
  chinese_cities,
  inputs  = c("industrial_labour_force", "working_funds", "investments"),
  outputs = c("gross_industrial_output", "profit_and_tax", "retail_sales"),
  id      = "DMU"
)

# Analysis
eff <- compute_efficiency(d, rts = "crs")          # Farrell efficiency
ce  <- compute_cross_efficiency(d)                 # cross-efficiency matrix

# Visualization (static ggplot2 by default)
plot_efficiency_distributions(d)
plot_io_pca_biplot(d)
plot_cem_heatmap(ce)
plot_cem_unfolding(ce)
plot_io_costa_frontier(d)
plot_io_peer_network(d, size_by_peers = TRUE)

# The same plot, interactive (plotly), with labels turned off
plot_io_pca_biplot(d, interactive = TRUE, labels = FALSE)

With your own data you can skip the explicit inputs/outputs by prefixing columns i_ (inputs) and o_ (outputs); dea_data() then recognises them automatically.

Function overview

Citation

If you use this package, please cite the thesis it is based on:

Ashkiani, S. (2019). Four Essays on Data Visualization and Anomaly Detection of Data Envelopment Analysis Problems [PhD thesis, Universitat Autonoma de Barcelona]. https://ddd.uab.cat/record/240333

See citation("deaviz") for the BibTeX entry.

Development Note

This package implements my DEA-Viz methods. An AI coding assistant (Claude, Anthropic) was used to help implement and refine the R code from my specifications and prototype functions; I designed the methods, and reviewed, tested, and validated all code, taking full responsibility for the software.

License

AGPL-3.