Using data envelopment analysis to measure and improve organizational performance

• Published date: 01 September 2023
• Author: Thomas R. Sexton,Christine Pitocco,Herbert F. Lewis
• Date: 01 September 2023
• DOI: http://doi.org/10.1111/puar.13679

RESEARCH ARTICLE

Using data envelopment analysis to measure and improve

organizational performance

Thomas R. Sexton | Christine Pitocco | Herbert F. Lewis

College of Business, Stony Brook University,

Stony Brook, New York, USA

Correspondence

Thomas R. Sexton, College of Business, Stony

Brook University, Stony Brook, NY 11794-3775,

USA.

Email: thomas.sexton@stonybrook.edu

Abstract

Organizations are complex and have many goals while almost all analytical tools

measure performance using only one goal. Thus, analysts often rely on multiple

analytical tools to produce a bewildering array of performance measures that

often lack internal consistency and a clear focus. In this article, we show how data

envelopment analysis (DEA) builds a performance frontier (analogous to a produc-

tion frontier) that measures organizational performance in the presence of multi-

ple organizational measures. The DEA frontier produces target values for each

organizational measure based on the observed performance of organizations in

the comparison set. In addition, DEA provides factor performance levels for each

performance measure for each organization and can detect circumstances in

which an organization has a strong overall performance measure but still has

weaknesses in one or more measures. We will illustrate this approach with applica-

tions to several examples using real data. Analyzing organizational performance

data is critical in the organizational improvement process. The data must directly

reflect the organization’s goals and the analytical tools used must be appropriate.

However, organizations are complex and have many goals while univariate analyti-

cal tools measure performance relative to only one goal. Thus, analysts often rely

on multiple analytical tools to produce a collection of performance measures,

sometimes resulting in a bewildering array of measures that lack focus.

Evidence for Practice

•We demonstrate how data envelopment analysis (DEA), commonly used to mea-

sure efficiency, can be used to evaluate organizational performance in the pres-

ence of multiple performance measures (PMs). The key insight is that efficiency

measurement seeks to reduce inputs and increase outputs. We replace inputs

with PMs for which smaller values are preferred and outputs with PMs for which

larger values are preferred.

•The model provides performance targets for each performance measure for

each organization. It also incorporates site characteristics, which measure the cir-

cumstances in which each organization operates, such as organization size, own-

ership type, that are beyond the organization’s control.

•Our model allows for variable and constant returns to scale. Standard ratio anal-

ysis assumes constant returns to scale, that is, a given percentage increase in

input will lead to the same percentage increase in output. Miles per gallon may

be sensible for automobiles, but not for other PMs.

•This paper provides five examples of applications of our methodology in the

public sector: nursing homes in New York State; hospitals in New York State;

hospital-acquired infections (HAIs) in New York State; rail transportation in the

United States; and global greenhouse gas emissions.

Received: 22 January 2023 Revised: 22 May 2023 Accepted: 26 May 2023

DOI: 10.1111/puar.13679

1150 © 2023 American Society for Public Administration. Public Admin Rev. 2023;83:1150–1165.wileyonlinelibrary.com/journal/puar

INTRODUCTION

In this paper, we show how data envelopment analysis

(DEA), which is commonly used to measure organizational

efficiency, can also be used to measure organizational

performance more generally. See, for example, Charnes

et al. (1978) and Banker et al. (1984) for some of the earli-

est introduction to DEA. The key insight is that, in measur-

ing organizational efficiency, we want to reduce the

consumption of inputs and to increase the production of

outputs. In measuring organizational performance, we

can identify performance measures for which we prefer

smaller values (these are analogous to inputs and we

refer to them as SIPs for Smaller is Preferred) and other

performance measures for which we prefer larger values

(these are analogous to outputs and we refer to them as

LIPs for Larger is Preferred). Thus, DEA may be used to

measure organizational performance even when in a set-

ting in which there is no production taking place.

TRADITIONAL RATIO ANALYSIS

Financial analysts are fond of ratios. They define leverage

as the ratio of the firm’s total assets to its total stock-

holder equity, asset efficiency as the ratio of its sales to its

total assets, and profitability as the ratio of its net income

to its sales. Each of these ratios measures the firm’s finan-

cial performance in its own specific manner. Leverage

measures the firm’s success in using its investors’money

to increase its financial strength; asset efficiency measures

its success in using its assets to increase sales; and profit-

ability measures its ability to control costs and retain sales

revenue. In all cases, higher values are preferred.

Which of these three ratios is most important to a pro-

spective investor or to the firm’s directors? The DuPont

Model (Soliman, 2008) addresses this question but not

entirely satisfactorily. The DuPont Model is the mathemat-

ical identity:

Return on Equity ¼Leverage Asset Efficiency

Profitability,

providing us with yet another ratio, Return on Equity.

The DuPont Model is useful, but it overlooks several

crucial flaws. First, ratios assume constant returns to

scale. Why should we expect that large and small opera-

tions should have comparable ratios? Often, this

assumption is unwarranted. A small company can

achieve a higher return on equity because it produces

only the most profitable products. Larger firms produce

additional products, which are likely to be less profit-

able than those first few.

Second, ratios lack a natural comparison value. Does a

given ratio indicate good performance or bad perfor-

mance? Analysts sometimes refer the mean value of the

ratio among the organizations in a specified comparison

set, such as other firms in the same industry. But the mean

value indicates average, not best possible, performance.

Sexton et al. (2008) showed how a three-stage vari-

able returns-to-scale (VRS) DEA model would improve the

DuPont Model. By removing the assumption of constant

returns to scale, this model can admit small and large

firms in the model. The larger comparison set means that

evaluations will be more robust, that is, less dependent

on a small comparison set. Moreover, as described below,

DEA measures performance against a “frontier”defined

by firms whose performance cannot be exceeded by any

other firm (or weighted average of other firms) in the

comparison set, thereby replacing average performance

with best possible performance.

Traditionally, DEA has been used to measure organiza-

tional efficiency. The model identifies two distinct sets

called inputs and outputs. Conceptually, it follows the tra-

ditional concepts of economic production: inputs are con-

sumed to produce outputs. For example, an automobile

may consume a certain amount of gasoline to move a

certain distance. We define the efficiency of the automo-

bile as the ratio of miles traveled to gallons consumed,

that is, miles per gallon. This approach fails if multiple

inputs are required to produce various amounts of multi-

ple outputs or constant returns to scale is invalid; DEA

permits these generalizations.

Other models have been proposed that differ from

the conventional DEA model proposed by Charnes et al.

(1978). Models have been used in many sectors and in

different ways other than the traditional measure of orga-

nizational efficiency. Nonetheless the use of DEA to evalu-

ate performance has been questioned due to potential

bias that exists when measurement error exists. Barnum

proposed replacement of DEA conventional models

where inputs and outputs are nonsubstitutable with

models that account for nonsubstitutability (Barnum &

Gleason, 2008; Barnum et al., 2017). Fried et al. (2002) pro-

posed a three-stage model. In the first stage, DEA is

applied to inputs and outputs only, in the second stage,

stochastic frontier analysis (SFA) is used to regress the first

stage performance measures against environmental vari-

ables, and in the third stage, depending on the orienta-

tion, either inputs or outputs are adjusted in order to

account for environmental effects and statistical noise

(uncovered in the second stage) and DEA is used to re-

evaluate producer performance. In this model, emphasis

is placed on slacks instead of on radial performance

scores Fried et al. (2002). Coupet et al. (2021) used DEA to

measure performance for benchmarking nonprofits with

similar service missions because DEA produces valid effi-

ciency scores and is a good way for nonprofits to learn

from each other.

In this article, we show that DEA can also measure

organizational performance. The central idea is that inputs

are quantities for which smaller values are preferred, and

outputs are quantities for which larger valuesare preferred;

we prefer to use less input and produce more output. We

PUBLIC ADMINISTRATION REVIEW 1151