THE PRODUCTIVE EFFICIENCY OF FEDERAL INSTITUTIONS OF BRAZILIAN HIGHER

EDUCATION

ABSTRACT

This paper discusses an important issue in the framework of Brazilian higher education, with an emphasis
on Federal Institutions of Higher Education (IFES). We examine the efficiency frontier of federal public
higher education through a non-parametric method called Data Envelopment Analysis (DEA). Dynamic
frontier was estimated and the Malmquist index was evaluated, determining productivity through panels.
Analysis used several educational administrative indicators (inputs and outputs of the productive process)
provided by the institutions themselves between 2004 and 2008. The total number of IFES (49) was
divided into two subsets – group A composed of 28 institutions and group B containing 21 – in order to
minimize heterogeneity in the sector. Estimation results for efficiency of dynamic models suggest as
frontier levels with elevated efficiency scores. On the other hand, productivity variations of IFES in each
panel show decreased productivity for most facilities. These findings demonstrate that although frontiers
exhibit low inefficiency scores, the technical efficiency frontier moved to a lower level. This indicates a
possible deterioration of the educational product over time.

Keywords: IFES. DEA. Efficiency Frontier. Higher Education.

RESUMO
Este trabalho discute um importante ponto que faz parte do arcabouço da educação superior brasileira,
tendo como ênfase as Instituições Federais de Ensino Superior (IFES). Examina a fronteira de eficiência
da educação superior pública federal através de uma metodologia não paramétrica denominada Análise
Envoltória de Dados (DEA). Foi estimada a fronteira dinâmica, bem como foi avaliado o índice de
Malmquist, que verifica a produtividade através de painéis. A mensuração foi realizada através de alguns
indicadores educacionais de gestão (os inputs e os outputs do processo produtivo) fornecidos pelas
próprias instituições, cujo período se estendeu de 2004 a 2008. O conjunto total das IFES (49) foi
dividido em dois subconjuntos – o grupo A contendo 28 instituições e o grupo B contendo 21 – a fim de
minimizar a heterogeneidade existente no setor. Os resultados da estimação do modelo dinâmico sugerem
fronteiras com níveis de scores de eficiência elevados. Por outro lado, a variação de produtividade das
IFES em cada painel mostrou queda de produtividade para a maioria das IFES. Esses resultados mostram
que apesar das fronteiras apresentarem baixos scores de ineficiência, houve um deslocamento da fronteira
técnica de eficiência para um nível inferior, indicando que pode estar havendo deterioração do produto
educacional ao longo do tempo.

Palavras-chaves: IFES. DEA. Fronteira de eficiência. Educação Superior.

Jel Codes: CO2. I23.

 2

INTRODUCTION

Performance of Institutions of Higher Education (IFES) has been the subject of growing attention
in recent years. Several studies address this matter at both national and international levels. Some of these
apply statistical tools to measure performance, while others use non-statistical instruments. These
investigations use “performance indicators” (such as the proportion of students in a specific year and the
cost per student) to determine efficiency in public or private IFES.
In order to measure their performance, IFES are treated as productive units the same as any other,
that is, they require inputs to attain a certain level of output. In addition, peculiarities inherent to the
educational sector are considered when calculating efficiency. Thus, a performance ranking is compiled
to determine the optimal allocation of resources to IFES.
The educational sector is substantially varied. It is therefore necessary to use extreme care in
constructing “performance indicators” for efficiency analysis in IFES. Two issues are of primary
importance: first, institutions operate under different conditions and environments, which are often not
simply explained. Second, the educational production sector contains many inputs and outputs.
The question of how public resources should be allocated in higher education has led most
research to focus on measuring the efficiency of public IFES. In recent years several studies have aimed
to measure efficiency levels and rank institutions accordingly. Furthermore, every country has its own
financing and resource allocation structures, which serve as a basis for assessing efficiency in higher
education.
In Brazil, the public sector has Federal, State and Municipal faculties and universities, while
private institutions may be profit or non-profit organizations. According to the higher education census
carried out by the Ministry of Education (MEC) in 2009, there are 2,314 IES in Brazil. Distribution by
administrative category indicates 90% are private institutions and 10% are public. Of the latter, 38% are
federal, 34% state and 28% are municipal institutions.
Federal Institutions of Higher Education (IFES) are mostly financed by the federal government
through combined taxes established in Article 212 of the 1988 Federal Constitution. Institutions also
receive funds from parliamentary amendments, contracts with public and private organizations, as well as
by their own funding. Government resource allocation is carried out by the Secretary of Higher Education
(SESu)/MEC through a resource allocation matrix aimed at favoring IFES efficiency.
The most common methods used for efficiency measurement in the educational field are statistical
and non-statistical techniques. Statistical methods are based on Ordinary Least Squares (OLS) regression
for analysis of the stochastic frontier. Linear programming (LP) techniques are applied to equate relations
between inputs and outputs. Data Envelopment Analysis (DEA) is most adequate for efficiency studies
using these techniques (LP).
DEA has been increasingly used to estimate efficiency in education. This is primarily because the
field is composed of multiple inputs and outputs, facilitating DEA frontier estimation. In addition, the
non-necessity of the production function does not cause model misspecification errors in estimation.
Thus, the aim of this study is to estimate the efficiency frontier of federal public education to
obtain the degree of productive efficiency for each federal higher Education institution (IFES), and then
compile a ranking from the most to least efficient.

2 LITERATURE REVIEW ON USING DEA METHODOLOGY TO ESTIMATION EFFICIENCY

IN HIGHER EDUCATION

The international literature contains several studies on the efficiency of public universities in
many countries, which mostly apply Data Envelopment Analysis (DEA). Ahn, Charnes & Cooper (1988),
compared US higher Education institutions aimed at research using three input and three output factors.
Public universities achieved greater levels of efficiency than private facilities.
 3

In a separate study, Rhodes & Southwick (1986), contrast the efficiency of 96 public and 54
private universities in the United States (US), applying the DEA model with five input and six output
factors. Results indicated that efficiency in private institutions at that time was higher than in public
facilities.
Breu & Raab (1994) used Data Envelopment Analysis (DEA) to assess efficiency in 25 of the best
US universities. Their findings confirm DEA as an appropriate method for measuring efficiency in higher
Education. In addition, an inverse relationship was found between the pre-established ranking and that
obtained with DEA.
SARRICO et al (1997) evaluated 90 higher Education facilities in the United Kingdom, in three
categories: (i) government/society; (ii) institutions: departments, staff and students, and (iii) potential
students. The authors used DEA methodology to determine efficiency levels and compared these with a
local ranking, the Times League Table. Results achieved through DEA indicated better efficiency
outcomes.
Førsund & Kalhagen (1999) investigated efficiency in Norwegian regional faculties in 1994, 1995
and 1996. Some institutions were found to be efficient with regard to education services, while inefficient
faculties showed significant variation between inefficiency levels. Additionally, productivity improved
during the years studied indicating a positive productivity effect, moving the efficiency frontier to a
higher level.
Thurlow & Field (2003) analyzed the technical efficiency of 45 British universities from 1980/81
to 1992/93. This period was chosen primarily because it represents a time of substantial changes in public
financing. The study recorded a significant increase in technical efficiency during this time, more
pronounced between 1987/88 and 1990/91.
Research by Afonso and Santos (2005) estimated efficiency of public universities in Portugal in
2003. Inputs were compiled from the number of teachers and university expenses and outputs were based
on graduation rates and the number of doctorate theses. Findings indicate a mean efficiency index of
approximately 55.3% and 67.8%, respectively among facilities investigated.
Abbot and Doucouliagos (2003) studied the technical efficiency scale of Australian universities
using DEA. Results point to performance homogeneity for the whole university system, indicating that
universities in that country operate at high efficiency levels. However, the authors confirm a possibility
for improved efficiency in some universities.
Joumady & Ris (2004) applied DEA methodology to measure efficiency differences in a group of
210 higher Education institutions from 8 European countries, using a sample of students graduated for
more than three years. Three models were assessed: the first focused on the competence of educational
services; the second (adjustment model) estimated Education efficiency after graduation, and the third
(global model) was designed to investigate the university’s overall performance. Outcomes were
substantially different for all three models, that is, efficiency varied in accordance with the model used.
In Brazil, analytical efficiency studies of federal universities have intensified in the last decade.
This is mainly in light of pressure from organizations linked to higher Education through the Ministry of
Education to evaluate the scope of this efficiency and its results for society as a whole.
Ramos and Souza (1997) analyzed the performance of federal higher education facilities using
DEA and found that about 39.1% of institutions evaluated achieved maximum efficiency, while 6.5%
were among the least efficient. In comparison with studies from other countries, the author’s results
demonstrate low efficiency during this period in public IFES.
Corbucci (2000) assessed MEC expenditure on federal higher Education institutions, establishing
efficiency and productivity indicators between 1995 and 1998. Outcomes confirm that despite a reduction
in operating costs for institutions analyzed, an increase was recorded in the number of stricto sensu
graduates and post-graduates, as well as higher scientific production. This indicates efficiency and
productivity improvements in these institutions.
Façanha and Marinho (2001) investigated performance differences between IFES in large regions
of Brazil from 1995 to 1998, using DEA to gauge efficiency. Measurement also considered IFES
 4

distribution into federal, state and municipal facilities. In relation to graduate teaching, findings
demonstrate that private and municipal IFES achieved higher relative efficiency than state and federal
facilities during the period studied. However, in Postgraduate teaching, results show asymmetry in
relative efficiency between IFES studied.
In his doctorate thesis, Belloni (2001) evaluated the productive efficiency performance of 33
Brazilian federal universities using DEA methodology. In contrast to results obtained by Ramos and
Souza (1997), only 6 of the 33 federal universities investigated were considered technically efficient. The
author determined that constant returns to scale did not apply in the case of public federal universities.
Estimates were therefore performed according to the DEA-BCC model, with variable returns to scale.
Finally, Oliveira and Turrioni (2005) assessed the relative efficiency of Federal Institutions of
Higher Education (IFES). Inputs and outputs were constructed using indicators from the Federal
Accounting Tribunal (TCU). The CCR – DEA model considered constant returns to scale. A total of 19
federal higher education institutions were evaluated, of which 5 were found to be technically inefficient.
The results of Ramos and Souza (1997) and Belloni (2001) are in contrast to those of Oliveira and
Turrioni (2005), possibly owing to the application of a model with constant returns to scale.

3 METHODOLOGY PROCEDURES

3.1 Data Envelopment Analysis - DEA

DEA methodology (Data Envelopment Analysis) was developed by Charnes, Cooper and Rhodes
(1978). This type of analysis generalizes the measurements of Dantzig (1951) and Farrel (1957), aiming
to measure the productive efficiency of production units with multiple outputs and inputs in order to
obtain an indicator satisfying Koopmans concept of efficiency.
DEA estimation is performed non-parametrically, measuring the efficiency of decision making
units (DMU) observed and comparing them among themselves to achieve a relative efficiency indicator.
This method uses DMUs as the best practices observed to construct an empirical production frontier
known as an efficient frontier.

3.1.1 SBM DEA

The slack-based DEA model (SBM) was introduced by Tone (1997, 2001) and contains two
assumptions:
i. Measurement is constant regarding the measurement unit of each input and output item.
ii. Measurement decreases monotonically at each input and output slack.

In order to estimate DMU efficiency with the SBM model, the following fractionated problem is
defined for (PL) at  , s  , s  .

1 m 
1 i 1 s i xio
SBM  min
 ,s  ,s 
 m
1 s
1  i 1 s i y ro
s 1
subjecto to x o  X  s 
y o  Y  s 
  0, s   0, s   0
 5

The model assumes that X  0 . If xio  0 , then the expression si xio is excluded. However, if
yio  0 , the positive number is very small so that expression si yro is disadvantageous.
The  value of the objective function satisfies the first assumption since the numerator and
denominator are measured in the same unit for each expression of this function. In addition, the objective
function value falls following increases at si and si when other terms remain constant; due to the
second assumption. Furthermore, 0    1 .
The SMB model can be defined by input oriented, output oriented and non-oriented structures.
Only the output oriented structure will be dealt with here, defined by the equation below:

SBM  O  O*  min 
1
 ,s 1 s 
1  s r y ro
s i 1
subjecto to x o  X 2
y o  Y  s 
  0, s   0

3.1.2 Panel data DEA: The Slack-Based Malmquist Index

In the case of Panel Data, linear programming DEA methodology may be applied ((input-oriented
or output-oriented) to calculate the Malmquist Index. The aim is to measure the productivity variation and
decomposition of this productive change in the technical and technical efficiency alteration. A more
detailed examination of the Malmquist productivity index is given in the equation below:

   
1
 1 ( x , y ) 2  2 ( x , y ) 2  2
MI   1 o o 1  2 o o 1  3
  
  ( xo , yo )  ( xo , yo )  
Where, MI is composed of four terms:  1 ( xo , yo )1 ,  2 ( xo , yo ) 2  ,  2 ( xo , yo )1  and  2 ( xo , yo ) 2  . The
first two are related to measurement in the same time period with t  1 or t  2 , while the latter two are
used for intertemporal comparison. Thus, if MI  1 technical efficiency improved; MI  1 signifies no
change in production technology; MI  1 shows a gain in production technology.
The output oriented SMB Malmquist index is defined by the following equation:

SBM  O  
 s x o , y o t  min 1 1 
 1 m


i 
 ,
 m i 1 
subjecto to x  X  t
o
s

n
1   i y iot   y sij  j i  1,..., q  4
j1

L  e  U
  0,  0
 6

3.1.3 Dynamic SBM DEA

Measuring intertemporal efficiency through DEA has been increasingly studied in recent years.
The first approach was window analysis (Klopp, 1985), followed by Fare et al (1994) who incorporated
the Malmquist Index in DEA structure. Sengupta (1995) and Fare and Grosspkof (1996) were the first to
develop the dynamic DEA model. Sengupta determined the dynamic efficiency of Farrell’s structure by
varying capital input in relation to input price changes over time. In contrast, Fare and Grosspkof
proposed a dynamic production frontier using an intermediate output that interconnects production
processes for each year.

Figure 1. Structure of Dynamic DEA

Source: Dynamic DEA: A slacks-based measure approach.

Based on the Fare and Grosspkof model, Tsutsui and Tone (2008) employed a dynamic DEA
structure using carry-over variables to estimate the production frontier over several time periods.
Additionally, frontier estimation is carried out with a non-radial model, that is, a slack-based model
known as the Dynamic SBM (DSBM). Its structure is shown in the figure above.
The distinguishing factor separating Dynamic DEA from other DEA forms is the existence of a
transition connecting the periods over time. Carry-overs, called links, can be categorized as:
Desirable (good) – desirable links are treated as outputs and the value of the link is the restricted access to
not less than that observed. Comparative scarcity of these links is considered inefficiency; for instance,
profit.
Undesirable (bad) – undesirable links are considered inputs. Their value is limited and cannot exceed
recorded value. Comparative excess of these links is seen as inefficiency; for example, loss and default.
Discretionary (free) – this link can be freely manipulated by the DMU and its value may increase or
decrease from that observed. Deviance in relation to actual value is not directly reflected in efficiency
evaluation, although continuity between the two periods explained in the next period exerts an indirect
effect on the inefficiency score.
Non-discretionary (fix) – in this case, the link is beyond DMU control, and its value fixed at an observed
level. This link also indirectly affects the efficiency score through the continuity between the two time
periods.

3.1.3.1 Production Possibility Set

Let n DMUs  j  1,..., n during T time periods t  1,..., T  , where each time period has m inputs
i  1,..., m , p non-discretionary (fixed) inputs i  1,..., p  , s outputs i  1,..., s  andd r non-discretionary
(fixed) outputs i  1,..., r  . In addition, (discretionary) inputs xijt i  1,...m , (non-discretionary) inputs
xijtfixo i  1,... p  , (discretionary) outputs yijt i  1,...s  and non-discretionary outputs yijtfixo i  1,...r 
 7

represent DMU values j and time period t. respectively. Carry-overs are symbolized into four
categories z good , z free , z bad , z fix . In order to identify time period (t), DMU (j) and item (i), for example,
the notation zitfree : free i  1,..., free; j  1,...n; t  1,...,T  , denoting all link free values observed until time
period T, is used.
           
Thus, the production possibility set xit  , xitfixo , yit  , yitfixo , zitgood , zitbad , zitfree , zitfix is
defined by:
n
xit   xijt tj , i  1,..., m; t  1,..., T 
j 1
n
xitfixo   xijtfixo tj , i  1,..., p; t  1,...,T 
j 1
n
yit   yijt tj , i  1,..., s; t  1,...,T 
j 1
n
yitfixo   yijtfixo tj , i  1,..., r; t  1,...,T 
j 1
n
zitgood   zijtgoodtj , i  1,..., ngood ; t  1,...,T 
j 1
n
zitbad   zijtbadtj , i  1,..., nbad ;  1,...,T 
j 1

zitfree : free , i  1,..., nfree; t  1,...,T 

n
zitfix   zijtfix tj , i  1,..., nfix; t  1,...,T 
j 1

tj  0 ,  j  1,..., n; t  1,...,T 

n

 t
j  1, t  1,..., T  5
j 1

Where tj  n t  1,..., T  is the intensity vector for time period t, and ngood , nfree , nfix are
the bad, free and fix links, respectively. The last restriction corresponds to the variable returns to scale
hypothesis. In the absence of this restriction, a model with constant returns to scale is proposed. To the
right of the above equations, variables assume positive values; on the left are the vector intensity
variables.
The continuity of carry-over links between time period t and t+1 is guaranteed by the following
condition:
n n

 zijt tj   zijt tj1 i; t  1,...,T 1 6

j 1 j 1

Where  is the standard symbol for good, bad, free and fix links. This restriction is essential to
the dynamic model as it connects activities between time periods t and t+1. using these equation for
production, we can express the DMUo (o  1,..., n) as follows:
n
xiot   xijt tj  sit , i  1,..., m; t  1,..., T 
j 1
n
xiotfixo   xijtfixo tj , i  1,..., m; t  1,..., T 
j 1
 8

n
yiot   yijt tj  sit , i  1,..., s; t  1,...,T 
j 1
n
yiotfixo   xijtfixo tj , i  1,..., r; t  1,...,T 
j 1
n
good
ziot   zijtgoodtj  sitgood , i  1,..., ngood ; t  1,...,T 
j 1
n
bad
ziot   zijtbadtj  sitbad , i  1,..., nbad ;  1,...,T 
j 1
n
ziotfree   zijtfree tj  sitfree , i  1,..., nfree; t  1,...,T 
j 1
n
ziotfix   zijtfix tj , i  1,..., nfix; t  1,...,T 
j 1
n

 t
j  1, t  1,...,T 
j 1

tj  0 , sit  0 , sit  0 , sitgood  0 , sitbad  0 e sitfree : freei, t  7 

Where sit , sit , sitgood , sitbad e sitfree are the slack variables denoting input excess, output deficit, link
deficit, link excess and link deviance, respectively.
3.1.3.2 Objective Function and Efficiency

     
Global efficiency evaluation of a DMUo (o  1,..., n) having t , st , st , stgood , stbad , stfree
can be done through structures input oriented, output oriented and non-oriented variables. We will only
address output orientation, given the definition of the research model. Output oriented global efficiency
 o* with good link is represented by:

1 T t  s   ngood sitgood 
1
 max  w 1 
1
  wi sit   good  8
 o* T t 1  s  ngood  i 1 i 1 z iot 

Subject to equations 6  and 7  , where wi is the weight for i and satisfies condition:

w 
i s 9
i 1

This objective function is an extension of the output oriented SBM model. It deals with output
inefficiencies including the link (good), which functions as an essential evaluation goal. Undesirable
inefficiency links are also calculated within the objective function, in the same manner as output
inefficiencies. However, undesirable links are not outputs; they only connect the two consecutive time
periods as per equation 6  . In equation 8 , each period within the brackets refers to the efficiency of
period t measured by slacks relative to the outputs and link, equal to the unit when all are equivalent to
zero. In addition, the right side of the equation 8 is the weighted average of efficiency gains over time,
required to be greater than or equal to 1. Thus, once global efficiency is defined, by reciprocity, output
global efficiency will be between 0 and 1.
 9

        
Using the optimal solution t* , st* , st* , stgood* , stbad* , stfree* , defines dynamic efficiency
output oriented  ot* as:

1
 ot*  , t  1,...,T  10
1  s  * good*
ngood siot 
1  i 1 wi siot  i 1 good 

s  ngood  ziot 

Therefore, output oriented global efficiency during period  ot*  is a harmonious mean for
 
efficiencies of periods  ot , demonstrated as follows:

1 T wt
1
  11
 ot* T t 1  ot*

4 EFFICIENCY MEASUREMENT AND RESULT ANALYSIS

4.1 DMU Selection

A set of 49 institutions are considered in order to calculate efficiency in federal public higher
education. IFES not included in the estimation are: Federal University Foundation of São Francisco,
Federal University of Recôncavo da Bahia, Federal University Foundation ABC, Federal University
Foundation of Pampa, Technological Federal University of Paraná, Federal University Foundation of
Grande Dourados. These were not included due to insufficient temporal dimension with the period
analyzed.
Federal Institutions for Higher Education (IFES) demonstrate a high degree of heterogeneity,
making calculation of the production frontier complex. Irrespective of the approach (statistical or non-
statistical), estimated models should incorporate the differences between institutions. These variances can
be determined from various standpoints (resources received, number of enrolled students, courses, among
others). However, a large university with many areas of knowledge in both research and outreach
programs demonstrates a significant difference in this sector when compared with institutions focused
largely on graduation.
Groups placed in the first phase were determined by analyzing three indicators. The first is the
total number of graduate students enrolled. The final two indicators are detailed in statistics compiled by
the National Council for Scientific and Technological Development (CNPq): i) the faculty to research
ratio, and ii) total investment in scholarships and research. These two groups are estimated as follows:

Table 1. Reference Group of Institutions

GROUP I GROUP II
University-Code University-Code University-Code University-Code
01 UFRJ 15 UFF 01 UFRRJ 15 UFERSA
02 UFRGS 16 UFPB 02 UFMS 16 UFTM
03 UFMG 17 UFLA 03 UFS 17 UFVJM
04 UFPE 18 UFG 04 UFMA 18 UFSJ
05 UFSC 19 UFSM 05 UFPI 19 UNIFAP
06 UNB 20 UFAM 06 UFT 20 UNIFAL
07 UFC 21 UFRPE 07 UFOP 21 UFCSPA
 10

08 UFV 22 UFU 08 UNIR

09 UFPR 23 UFAL 09 UFJF
10 UFBA 24 FURG 10 UFRA
11 UFSCAR 25 UFPEL 11 UNIFEI
12 UFPA 26 UFES 12 UFAC
13 UNIFESP 27 UFMT 13 UFRR
14 UFRN 28 UFMT 14 UNI-RIO
Source: MEC/INEP/DEED.
Preparation: Author.

4.2 Model Definition

The concept of efficiency is related to the use and allocation of resources. Thus, in order to obtain
reliable estimates in efficiency calculation, it is essential to use indicators that consistently represent the
characteristics of the educational production function.
Outputs and inputs selected to measure the efficiency of Federal Institutions for Higher Education
(IFES) in the present study were based on the main outputs and inputs used in various studies in recent
decades, as well as the reality of the Brazilian federal higher education system. These are as follows:

Output
Educational outputs can be defined as a function of services offered by Institutions of Higher
Education (IFES). The following variables were therefore defined as outputs for this study:
 Graduate/Undergraduate students (GSR).
 CAPES/MEC Concept for Postgraduation.
(CAPES - The Brazilian Federal Agency for the Support and Evaluation of Pós-Graduate Education)

Input
Educational inputs may be defined as variables enabling services to be offered by the IFES. In the
present study, the following variables were defined as inputs:

 Curent cost/student equivalent.

 Full-time student/teacher equivalent.
 Full-time student/staff equivalent.
 Teacher Qualification Index.

Once DMUs are defined and the model selected, the period of analysis for efficiency measurement
is determined. Criteria for establishing the period were based on the availability of data for indicators
used in this research. As such, the analysis period chosen is 2004 to 2008.
Variables used in this study were primarily obtained from the following organizations:
 The Ministry of Education (MEC), through the website: <http://www.mec.gov.br>.
 National Institute of educational Studies and Research (INEP), on the website
<http://www.inep.gov.br>.

4.3 Efficiency Estimation and Results Analysis

The model aimed to measure efficiency considering several indicators related to IFES
administration. Estimations were performed using DEA Solver Professional software version 7.0. In this
model, two analyses were conducted. The first sought more robust results with regard to IFES
 11

homogeneity, estimating the efficiency frontier of IFES in two groups: Group A (IFES considered large);
Group B (IFES viewed as small). The second estimated the Panel data DEA.

 Estimation Group A

In Dynamic DEA estimation, a discretionary carry-over variable was introduced to link the time
periods. Each term in the estimation represents a period with the inclusion of a link, that is, term 1 refers
to 2004 including the carry-over, with the same procedure for the other terms. Table below shows
findings for the period 2004=>2008.

Table 2. IFES Group A

2004=>2008
Overall
Rank IFES Term1 Term2 Term3 Term4 Term5
Score
1 UFAM 1 1 1 1 1 1
1 UFBA 1 1 1 1 1 1
1 UFCG 1 1 1 1 1 1
1 UFF 1 1 1 1 1 1
1 UFLA 1 1 1 1 1 1
1 UFMG 1 1 1 1 1 1
1 UFMT 1 1 1 1 1 1
1 UFPA 1 1 1 1 1 1
1 UFPB 1 1 1 1 1 1
1 UFPEL 1 1 1 1 1 1
1 UFRGS 1 1 1 1 1 1
1 UFRJ 1 1 1 1 1 1
1 UFRPE 1 1 1 1 1 1
1 UFSCAR 1 1 1 1 1 1
1 UFSM 1 1 1 1 1 1
1 UFU 1 1 1 1 1 1
1 UFV 1 1 1 1 1 1
1 UNIFESP 1 1 1 1 1 1
19 UFRN 0,92 0,93 0,94 0,99 0,87 0,89
20 UFPE 0,91 0,84 0,90 0,86 1 1
21 UFC 0,88 0,82 0,81 1 0,91 0,91
22 UFES 0,88 0,91 0,88 0,94 0,85 0,84
23 FURG 0,87 0,89 0,90 0,90 0,83 0,84
24 UFSC 0,86 0,87 0,83 0,83 0,89 0,90
25 UFAL 0,86 1 1 0,94 0,67 0,81
26 UFPR 0,85 0,84 1 0,83 0,78 0,83
27 UFG 0,85 0,79 0,79 0,90 0,87 0,91
28 UNB 0,84 0,87 0,82 0,82 0,84 0,84
Source: Estimates of the research.
Preparation: Author.

In dynamic analysis of the frontier during the entire intertemporal trajectory, around 64% of IFES
were located on the efficiency frontier, while 22% were below it and 14% were both on and below the
efficiency frontier. When analyzing the dynamic frontier over the whole intertemporal trajectory
2004=>2008, about 64% of IFES were on the efficiency frontier, making them part of the efficient set,
while 36% were below it and therefore placed in the inefficient group. In addition, the following 10
 12

institutions were not part of the dynamic frontier, making up the inefficient set: FURG, UFAL, UFC,
UFES, UFG, UFPE, UFPR, UFRN, UFSC and UNB. The UNB recorded the highest inefficiency level
during the intertemporal trajectory with an overall score of 0.84.
The advantage of obtaining an intertemporal frontier is the inclusion of input and output behavior
in efficiency scores during the time period analyzed, generating a more robust efficiency frontier. Another
fact for analysis is that the dynamic efficiency frontier is situated above the efficiency score at 0.80. This
indicates that IFES from group A have a relatively high frontier in relation to the overall score. Finally,
the graph below shows the technical dynamic efficiency frontier for this set of IFES.

Figure 2. IFES Group A: Frontier Dynamics

Source: Estimates of the research.

Preparation: Author.

Following dynamic analysis of efficiency frontiers, we examined the decomposition of

productivity change, technical and technical efficiency alterations over time. It is therefore necessary to
consider panel analysis. The table below displays the panel obtained during time period analysis.

Table 3. Decomposition of the Malmquist Index (2004=>2008)

IFES Índex Change Pure Índex Change Scale Índex of

Efficiency Efficiency Malmquist

FURG 0,92 0,91 0,84

UFAL 0,77 0,97 0,74
UFAM 1 1 1
UFBA 1 0,95 0,95
UFC 1,09 0,87 0,95
UFCG 1 0,87 0,87
UFES 0,91 0,91 0,83
UFF 1,09 0,94 1,02
UFG 1,15 0,86 0,99
UFLA 0,87 0,98 0,85
UFMG 1,04 0,98 1,02
UFMT 1 1 1
UFPA 1 1,02 1,02
UFPB 1 0,82 0,82
UFPE 1,26 0,85 1,07
 13

UFPEL 1,17 0,90 1,05

UFPR 0,98 0,87 0,85
UFRGS 1,06 0,89 0,95
UFRJ 0,98 0,99 0,97
UFRN 0,95 0,94 0,89
UFRPE 1 1 1
UFSC 1,01 0,91 0,92
UFSCA
0,97 1 0,97
R
UFSM 1 0,95 0,95
UFU 0,96 0,98 0,94
UFV 1,04 0,94 0,98
UNB 0,98 0,94 0,92
UNIFE
1 0,98 0,98
SP
Source: Estimates of the research.
Preparation: Author.

In this panel (2004=>2008), UFAM, UFMT and UFRPE maintained their constant productivity
with, with an index equal to the unit. UFF, UFMG, UFPE and UFPEL increased their productivity during
the panel, with UFPE achieving the best index (1.09). In relation to the 20 IFES with reduced
productivity, UFAL recorded the worst result, with an index of 0.74.
Decomposition of the Malmquist Index in the above table demonstrates that the loss of
productivity in Group A IFES is due to the effect of scale efficiency change, known as the displacement
effect. Approximately 82% of IFES obtained a scale efficiency change index lower than 1. Consequently,
the production frontier shifted to a lower level during the panel (2004=>2008).

 Estimation Group B

The dynamic efficiency frontier (DSBM) was then determined in order to obtain more robust
efficiency frontier scores for IFES in group B. Table below shows findings for the period 2004=>2008.

Table 4. IFES Grupo B

2004=>2008
Rank DMU Overall Score Term1 Term2 Term3 Term4 Term5
1 UFAC 1 1 1 1 1 1
1 UFCSPA 1 1 1 1 1 1
1 UFERSA 1 1 1 1 1 1
1 UFJF 1 1 1 1 1 1
1 UFMS 1 1 1 1 1 1
1 UFPI 1 1 1 1 1 1
1 UFRA 1 1 1 1 1 1
1 UFRR 1 1 1 1 1 1
1 UFRRJ 1 1 1 1 1 1
1 UFSE 1 1 1 1 1 1
1 UFT 1 1 1 1 1 1
1 UFTM 1 1 1 1 1 1
1 UFVJM 1 1 1 1 1 1
1 UNIFAL 1 1 1 1 1 1
1 UNIFAP 1 1 1 1 1 1
1 UNIRIO 1 1 1 1 1 1
17 UFMA 1 1 1 1 0,97
18 UFOP 0,99 1
0,95 0,93 0,93 0,93 0,95
 14

19 UNIFEI 1 1 1 0,80
20 UFSJ 0,90 0,77 1
21 UNIR 0,84 0,63 0,93 0,85 0,921 0,78
Source: Estimates of the research.
0,80 0,78 0,75 0,76
Preparation: Author.

A discretionary carry-over variable was once again introduced in Dynamic DEA analysis to link
the time periods. Intertemporal trajectory of the efficiency frontier indicated that about 76% of IFES were
on the efficiency frontier during the entire trajectory, 24% were both on and below the efficiency frontier
during the entire trajectory In analysis of overall score results, 3 universities were not part of the dynamic
frontier: UFMA, UNIFEI, and UNIR, with UNIR registering the highest degree of intertemporal
inefficiency with an overall score of 0.78.
The graph below illustrates the dynamic frontier of the overall score. Only 3IFES are below the
frontier and outside of the efficient group. Another point to be analyzed is that the dynamic efficiency
frontier is above the efficiency score of 0.78, indicating that IFES in group B have a relatively high
frontier in relation to the overall score.

Figure 3. IFES Group B: Frontier Dynamics

Source: Estimates of the research.

Preparation: Author.

In order to examine the dynamic productivity, a panel analysis was performed using the
Malmquist Index. The following table displays the panels obtained during periods evaluated.

Tabela 5. Decomposition of the Malmquist Index (2004=>2008)

IFES Índex Change Pure Índex Change Scale Índex of

Efficiency Efficiency Malmquist

UFAC 1,20 0,77 0,93

UFCSPA 0,93 1,12 1,04
UFERSA 0,84 0,93 1
UFJF 0,86 1 0,86
UFMA 0,95 0,88 0,83
UFMS 2,63 0,54 1,43
UFOP 1,12 0,65 0,73
UFPI 1,35 0,64 0,86
 15

UFRA 1,16 0,95 1,11

UFRR 1 0,88 0,88
UFRRJ 0,77 0,93 0,72
UFSE 1,22 0,86 1
UFSJ 1,61 0,96 1,54
UFT 1 0,88 0,88
UFTM 1 0,88 0,88
UFVJM 1 0,91 0,91
UNIFAL 1 0,95 0,95
UNIFAP 1 0,59 0,59
UNIFEI 0,97 0,99 0,96
UNIR 1,03 0,81 0,84
UNIRIO 0,94 0,83 1
Source: Estimates of the research.
Preparation: Author.

The panel (2004=>2008) indicates that IFES UFERSA, UFSE and UNIRIO maintained constant
productivity with an index equal to the unit. Only UFCSPA, UFMS, UFRA, UFRRJ and UFSJ increased
their productivity during the panel, with UFSJ achieving the best index of 1.54. The total amount of IFES
with reduced productivity was high; approximately 76% were from group B with UNIFAP recording the
lowest result of 0.59.
Decomposition of the Malmquist Index in the table above demonstrates that, as in group A IFES,
loss of productivity for those in group B was mainly due to the effect of scale efficiency change.
Approximately 91% of institutions obtained a scale efficiency change index of less than 1. Consequently,
displacement in the production frontier to a lower level occurred during the panel (2004=>2008).

5 CONCLUSIONS

The present study sought to analyze Brazilian Federal Institutions for Higher Education (IFES)
using a non-parametric method to measure their technical efficiency. A model was applied to determine
the maximum educational product obtained by each institution investigated, given that the product is a
function of educational resources.
This study addresses the educational production function, demonstrating its specificity and the
indicators that may be used to compile educational inputs and outputs. The DEA methodology is then
defined in order to carry out technical efficiency estimations for IFES. A literature review was also
performed regarding DEA application in education economics. After consolidation of methodological
procedures, IFES in the federal public higher education sector were defined to allow for efficiency
frontier estimations based on the proposed model. Estimations were therefore made based on a total set of
IFES comprising 49 institutions, divided into two subsets (group A – 28 institutions; group B – 21
institutions) to minimize heterogeneity in the sector.
Results of efficiency frontier estimations for IFES in the first subset (group A) demonstrated that,
Estimation of the dynamic frontier indicated that 64% of IFES were on the efficiency frontier, whereas
about 36% were placed below it. Moreover, the introduction of the student equivalent carry-over proved
to be a good proxy to determine the increase or decrease in productivity during the period analyzed.With
regard to the dynamic frontier IFES group B, a mean of 76% universities were on the efficiency frontier
for the entire intertemporal trajectory, while about 24% were under it. As in group A, the Introducing the
student equivalent carry-over proved to be a good proxy to determine the increase or decrease in
productivity during the period analyzed.
 16

It is important to note that the intertemporal efficiency frontiers of the federal public higher
education sector for IFES in groups A and B can be considered plausible for both the current financial
structure and the actual resource allocation model for these institutions. On the other hand, the Malmquist
Productivity Index showed reduced productivity for most facilities. Examining the decomposition of this
index confirms displacement of the frontier to a lower level. This suggests that the technical efficiency
frontier of federal public higher education decreased to a lesser level during the period analyzed.
This situation demonstrates that the educational product of IFES has deteriorated over time, in
agreement with findings when analyzing the evolution of resources allocated to IFES. That is, a reduction
in funds for IFES may compromise performance with respect to their educational product. In addition,
some nationally significant institutions did not achieve efficiency scores compatible with the resources
they received.
However, in order for these results to corroborate the true situation in federal public higher
education, it is necessary to analyze the outcome for each IFES separately and subsequently make
observations regarding causes and inefficiencies relative to productivity loss.
In conclusion, we hope that the academy and institutions examined here, as well as SESu/MEC
and ANDIFES (The National Association of Directors of Federal Higher Education Institutions) examine
these results with a view to their use as an aid in discussing the politicies periodically adopted for the
sector.

REFERENCES

ABBOT, M; DOUCOULIAGOS, C. The Efficiency of Australian Universities a Data Envelopment

Analysis. Economics of Education Review, v. 22, p. 89–97, 2003.

AFONSO, A; SANTOS, M. Students and Teachers A DEA Approach to the Relative Efficiency of
Portuguese Public Universities. Technical University of Lisbon, series Working Papers n. 2005/07,
2004.

AHN, T.; CHARNES, A.; COOPER, W. Some statistical and DEA evaluations of relative efficiencies
of public and private institutions of higher learning. Socio-Economic Planning Sciences, v. 22, n. 6,
p. 259-269, 1988.

BELLONI, J. A. Uma Metodologia de Avaliação da Eficiência Produtiva de Universidades Federais

Brasileiras. Tese (Engenharia de Produção), Programa de Pós-Graduação em Engenharia de Produção,
Universidade Federal de Santa Catarina, 2000.

BRASIL. República Federativa do Brasil. Constituição da República Federativa do Brasil: Art. 212.
Brasília. Senado, 1988.

BREU, T.M; RAAB, R.L. Efficiency and Perceived Quality of the Nation’s ‘Top 25’ National
Universities and National Liberal Arts Colleges: An Application of Data Envelopment Analysis To
Higher Education. Socio-Economic Planning Science, v.28, p.33-45, 1994.

CHARNES, A; COOPER, W.W; RHODES, E.Measuring the efficiency of DMUs. European Journal of
Operational Research, v. 2, p. 429–44, 1978.

CHARNES, A; COOPER, W. W. Measuring the Efficiency and Productivity of Decison Manking

Units. European Journal of Operational Research, vol. 3, p. 339-338, 1979.
 17

COOPER, W. W; SEIFORD, L. M; TONE, K. A Comprehensive Text with Models, Applications,

References and DEA-Solver Software. New York: 2. ed. Spring, 2007.

CORBUCCI, P. R. As Universidades Federais: Gasto, Desempenho, Eficiência e Produtividade.

IPEA, texto para discussão 752. Disponível em: <http://www.ipea.gov.br>. Acesso em: 5 mar. 2009.

DANTZING, G. B.Maximization of a linear function of variables subject to linear inequalities. In

T.C. Koopmans (ed.), Activity Analysis of Production and Allocation, New York :Wiley, 1951.

FARE, R; GROSSKOPF, S. Intertemporal production Frontiers: with dynamic DEA. Norwell:

Kluwer, 1996.

FAÇANHA, L. O; MARINHO, A. Instituições de Ensino Superior Governamentais e Particulares:

Avaliação Comparativa de Eficiência. IPEA, texto para discussão n. 813, 2001. Disponível em:
<http://www.ipea.gov.br>. Acesso em: 4 jan. 2010

FARE, R; GROSSKOPF, S.; LINDGEN, B.; ROOS, P. Productivity change

in swedish hospitals: A Malmquist output index approach. 1994. Apud COOPER, W. W; SEIFORD,
L. M; TONE, K. A Comprehensive Text with Models, Applications, References and DEA-Solver
Software. New York: 2. ed. Spring, 2007.

FARRELL, M.The Measurement of Productive Efficiency. Journal of the Royal Statistical Society,
Series A, 120, Part 3, 253-290, 1957.

FLEGG, A. T; ALLEN D.O; FIELD K. & THURLOW, T.W.. Measuring the Efficiency and
Productivity of British Universities: An Application of DEA and the Malmquist Approach.
University of the West of England, Department of Economics, series Discussion Papers n. 304, 2003.

FORSUND, F. R; KALHAGEN, K, O. Efficiency and Productivity of Norwegian Colleges. Oslo

University, Department of Economics, series Memorandum n. 11, 1999.

GOTO, M; NENOMOTO, J. Measurement of Dynamic Efficiency in Production: An Application of

Data Envelopment Analysis to Japanese Electric Utilities. Journal of Productivity Analysis, vol. 19, n.
2-3, p.191-210, 2003.

GOTO, M; NENOMOTO, J. Dynamic data envelopment analysis: modeling intertemporal behavior

of a firm in the presence of productive inefficiencies. Economics Letters, vol. 64, n. 51–56, 1999.

Instituto Nacional de Estudos e Pesquisas Educacionais (INEP). Censu da Educação Superior: 2004.
Brasília, 2009. Disponível em: <http://www.inep.gov.br>. Acesso em: 10 fev. 2010.

JOHNES, G; JOHNES, J. International Handbook on the Economics of Education. Edward Elgar

Publishing Ltd. Cheltenham, UK. Northampton, MA, USA. 2004. Cap 16, p. 613-627.

JOUMADY, O; RIS, C. Performance in European Higher Education: A Non-Parametric Production

Frontier Approach. Journal of Education Economics, 2. ed, p. 189-205, 2005.

KOOPMANS, T. C. An Analysis of Production as an Efficient Combination of Activities. In T. C.

Koopmans, ed., Activity Analysis of Production and Allocation, Cowles Commission for Research in
Economics, Monograph No. 13. New York: Johon Wiley and Sons, Inc. 1951.
 18

KLOPP, G. The Analysis of the efficiency of production system with multiple inputs and outputs.
University of Illinois at Chicago. Industrial and Systems Engineering College, 1985.

Mc MILLAN, M; Datta, D. The Relative Efficiencies of Canadian Universities: A DEA Perspective.

Canadian Public Policy, v. 24 (4), p. 485-511, 1988.

MALMQUIST, S. Index numbers and indifference surfaces. Trabajos de Estatistica 1953. 4, 209–42.

OLIVEIRA, C. E. M; TURRIONI, J. B. Avaliação de Desempenho de Instituições Federais de Ensino

Superior Através da Analise por Envoltória de Dados (DEA). In ENGEP 2005.

PIRES, J. S. Dal Bem. Proposta de Modelo de Orçamento Baseado em Desempenho Acadêmico para
as Universidades Públicas do Estado do Paraná. Tese (Tese em Engenharia de Produção) – Programa
de Pós-Graduação em Engenharia de Produção, Universidade Federal de Santa Catariana, Florianópolis,
2001.

RHODES, E. L.; SOUTHWICK, L. Determinants of efficiency in public and privates universities.

Working Paper, School of Environmental and Public Affairs. Indiana University. Blomington IN 47407,
USA. 1986.

SAMPAIO DE SOUZA, M. DA C; STOSIC, B. D. Technical Efficiency of the Brazilian

Municipalities: Correcting Non-Parametric Frontier Measurements for Outliers. Journal of
Productivity Analysis, vol. 24, n. 2, p. 157-181, 2005.

SARRICO, C. S. Data envelopment analysis and university selection. Journal of the Operational
Research Society, v. 48, p. 1163-1177, 1997.

SCHWARTZMAN, J. As Teorias Da Universidade Brasileira. Disponível em:

http://www.schwartzman.org.br/simon/teorias.htm. Acesso em 11 Out. 2008.

SCHWARTZMAN, J. Um sistema de indicadores para as universidades brasileiras. In:

SGUISSARDI, V. (org.), Avaliação Universitária em Questão. Campinas, Autores Associados, 1997.

SENGUPTA, J. K. Persistence of dynamic efficiency in Farrell models. Applied Economics, vol. 29, p.
665–671, 1997.

SENGUPTA, J. K. A dynamic efficiency model using data envelopment analysis. International Journal
of Production Economics, vol. 62, p. 209-218, 1999.

SENGUPTA, J. K. Nonparametric efficiency analysis under uncertainty using data envelopment

analysis. International Journal of Production Economics, vol. 95, p. 39–49, 2005.

SOUZA, H. R.; RAMOS. F.S. Performance Evaluation in The Public Sector: An Application of
Efficiency Measures to Brazilian Federal Higher Education Institutions. Tenth World Productivity.
p. 430-450, 1997.

SHEPHARD, R. Cost and Production Functions. Princeton University Press, Princeton, New Jersey,
EUA. 1953.
 19

TONE, K.; TSUTSUI, M. A Slacks-based Measure of Efficiency in Data Envelopment Analysis.

Research Reports, Graduate School of Policy Science, Saitama University and subsequently published in
European Journal of Operational Research 130. p.498-509, 2001.

TONE, K; TSUTSUI, M. Dynamic DEA: A slacks-based measure approach. Omega, vol. 38, p.145-
156, 2010.

WONG, Y; BEASLEY, J. Restricting Weight Flexibility in DEA. Journal of Operational Research

Society, v. 41, p. 829-835.