Available online at www.sciencedirect.

com

ScienceDirect
Procedia - Social and Behavioral Sciences 89 (2013) 895 900

2nd Cyprus International Conference on Educational Research, (CY-ICER 2013)

Analysis of the Determinants of Universities Efficiency in Turkey:

Application of the Data Envelopment Analysis and Panel Tobit Model
Sibel Selima*,Sibel Aybarc Bursalioglub

Celal Bayar University,Department of Econometrics, Manisa, Turkey

Celal Bayar University, Ahmetli Vocational High School, Manisa, Turkey

Abstract
This paper builds on a two- stage Data Envelopment Analysis to determine factors on the efficiency of
universities in Turkey in 2006-2010. The first stage is concerned with data envelopment analysis to measure
the efficiency of the universities. This is followed by factors that affect the efficiency of the universities. The
results of the model demonstrate that, the effect of project allocations found to be insignificant. Additionally,
number of students per academic has a positive effect on relative efficiency of universities in Turkey. It is seen
the highest increase at doctorate level. Employment and number of publications, as expected, have positive
effect on efficiency. Because the central government budget appropriations has a negative effect on efficiency,
the higher education institutions in Turkey may lead to the search for alternative financing such as Triple Spiral
model in co-operation with the private sector, R&D support, project support, counselling, within the framework
of university-industry-government.
2013 Published by Elsevier Ltd.
2013 The Authors. Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
Selection and or peer-review under responsibility of Assoc. Prof. Dr. Zehra znar, Ataturk Teacher Training
Selection and/or peer-review under responsibility of Prof. Dr. Huseyin Uzunboylu, Near East University, Faculty of Education,
Academy, North Cyprus
Cyprus

Keywords: Efficiency in higher education, Data envelopment analysis, Panel tobit model, Turkey

1. Introduction
Higher education institutions that shape today's information society are an important actor in
providing economic development and growth, and competitive advantage to countries in the
international arena, as well as in providing prestige and high level of income to individuals. In this
regard, efficiency of higher education institutions in academic and research activities and
investigating the factors that determine the efficiency are also important.
*

Corresponding author: Sibel Selim. Tel: + 90 236 2330657.

E-mail: [email protected].

1877-0428 2013 The Authors. Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
Selection and/or peer-review under responsibility of Prof. Dr. Huseyin Uzunboylu, Near East University, Faculty of Education, Cyprus
doi:10.1016/j.sbspro.2013.08.952

896

Sibel Selim and Sibel Aybarc Bursalioglu / Procedia - Social and Behavioral Sciences 89 (2013) 895 900

All universities and schools of higher education are affiliated to the Council of Higher Education
(CoHE) established in 1981 in Turkey. The Council is an autonomous public juridical body with the
authority and responsibility to administer the activities of all institutions of higher education. It
annexes to itself the following: the Higher Supervisory Board, the Student Selection and Placement
Centre (SSPC) and other sections related to planning, research, development, evaluation, budgets,
investment and coordination. (Girgin, 2006).
With over 3 million students currently enrolled at university or following distance education
courses, Turkey currently has a mass higher education system. Over the last ten years, in order to
facilitate student access throughout the country, the main objective of higher education policy has
been to increase the number of universities (Altnsoy, 2011). Today, there are 188 universities including 103 public universities, 65 private universities, 7 private vocational schools of higher
education and 13 other institutions of higher education.
The paper contributes to the literature public universities in Turkey by estimating the relative
efficiency through the DEA and the factors that affect the relative efficiency through the panel tobit
model with random effects. The research literature on Turkish public universities relative
efficiency is very limited. Most of the studies have been made for the efficiency of a public
universitys departments or the relative efficiency of public universities with one-stage DEA.
However, this study presents a comprehensive analysis for the 51 public universities with two-stage
DEA and provides important findings on the determinants of public universities efficiency with
panel tobit model in Turkey. The employment is the most important variable for the efficiency of a
university. For the first time, the employment is used in this study.
2. The data and variables
The data employed in this study is derived from the CoHE, Employment Institution (EI),
Measuring Selection and Placement Center (MSPC) and The Scientific and Technological Research
Council of Turkey (TBTAK). Based on a 5-year period (2006-2010) and 51 public universities, a
panel dataset was constructed.
Analytical Budget Classification became law in 2004 and started to apply in 2006 in Turkey. So,
we determine analysis period as 2006-2010 in this study. The other constraint of our analysis is
impossible to reach the employment from MSPC and private sector. Therefore, we used
employment data of the EI. The inputs and outputs are identified for university performance
measurement. They are presented in Table 1.
Table 1. Input and output variables for efficiency of the public universities in Turkey
Input Variables
Central government budget appropriations
Own Revenue
Project allocations (TBTAK)
Project allocations (Scientific Research Projects)
The total academic

Output Variables
Number of graduate students per academic
Number of post graduate students per academic
Number of doctorate students per academic
Number of publications
Number of employment

3. Results
3.1. Results of the data envelopment analysis
DEA is used to estimate 51 public universities efficiency scores. This is a non-parametric
technique that considers each public university as a decision making units (DMU) using inputs to
produce outputs (Cooper et al., 2000). DEA mathematical formulation can deal with both constant
returns to scale (CRS) and variable returns (VRS). Both CRS and VRS efficiency can be calculated

Sibel Selim and Sibel Aybarc Bursalioglu / Procedia - Social and Behavioral Sciences 89 (2013) 895 900

for each unit (Sibiano and Agasisti, 2011). BCC model takes into account the effect of VRS within
the analyzed group of DMUs while CCR model takes into account the effect of CRS (Golany and
Roll, 1989). We used VRS formulation to take into account the different relative size of the public
universities in Turkey.
Table 2. Efficiency scores of the 51 public universities in Turkey
University
Abant zzet Baysal University
Adnan Menderes University
Afyon Kocatepe University
Akdeniz University
Ankara University
Atatrk University
Balkesir University
Boazii University
Celal Bayar University
Cumhuriyet University
anakkale Onsekiz Mart
ukurova University
Dicle University
Dokuz Eyll University
Dumlupnar University
Ege University
Erciyes University
Eskisehir Osmangazi University
Frat University
Galatasaray University
Gazi University
Gaziantep University
Gazi Osman Paa University
Gebze Institute of Technology
Hacettepe University
Harran University
nn University
stanbul University
stanbul Teknik University
zmir Institute of Technology
Kafkas University
Kahraman Mara St mam
Karadeniz Teknik University
Krkkale University
Kocaeli University
Marmara University
Mersin University
Mula University
Mustafa Kemal University
Nide University
Onkokuz Mays University
Orta Dou Teknik University
Pamukkale University
Sakarya University
Seluk University
Sleyman Demirel University
Trakya University
Uluda University
Yldz Teknik University
Yznc Yl University
Zonguldak Karaelmas University

2006
0.99
0.57
0.95
0.783
1
0.919
1
0.982
0.812
0.842
1
0.82
0.764
0.686
1
0.762
1
0.662
0.904
1
0.874
0.797
0.823
1
0.966
1
1
0.823
1
1
1
0.945
0.882
1
1
1
0.73
0.746
0.826
1
0.803
1
0.663
0.908
1
0.836
0.923
0.865
0.829
0.896
1

2007
1
0.772
1
0.791
1
0.9
1
0.841
0.873
0.894
1
0.895
0.813
0.864
1
0.901
1
0.828
0.969
1
0.958
1
0.909
1
0.931
0.939
1
0.603
0.84
1
1
0.876
1
0.976
1
1
0.804
0.758
0.85
1
1
1
0.778
0.939
1
0.964
0.894
0.93
0.944
0.937
0.928

2008
0.995
0.841
1
0.803
1
1
1
1
0.844
0.883
0.843
1
0.845
0.864
1
0.881
1
0.963
1
1
0.954
1
1
1
0.906
0.96
0.964
0.894
0.808
1
1
0.913
1
0.971
1
1
0.855
0.808
0.9
1
0.956
1
0.803
1
1
1
0.981
1
0.821
0.847
0.979

2009
0.954
0.819
1
0.794
1
0.991
1
0.929
0.879
0.902
0.838
1
0.959
0.842
1
0.91
1
0.762
0.958
1
0.923
0.972
0.892
1
0.864
1
1
0.829
0.759
0.974
1
0.982
0.939
0.891
0.987
1
0.815
0.829
1
1
0.933
1
0.874
0.974
1
1
0.904
0.871
1
0.941
0.916

2010
1
0.928
1
0.83
0.966
0.982
1
0.894
0.944
0.953
0.826
1
0.821
0.885
1
0.923
1
0.772
0.946
1
0.957
0.956
0.916
1
0.946
1
0.91
0.842
0.737
1
1
0.986
0.967
0.905
1
1
0.844
0.83
1
1
0.949
1
0.827
1
1
1
0.924
0.931
0.743
0.908
0.936

The DEA method can be input or output orientated. The output-oriented approach focuses on
how high maximal output can be achieved with the same amount of resources. Tibenszkyn (2007)
states that the output-oriented approach is appropriate for higher education because the principle of
cost minimization is not applied according to the market conditions (Toth, 2009). We used outputoriented approach to determine the output maximization of the public universities in Turkey. As it is

897

898

Sibel Selim and Sibel Aybarc Bursalioglu / Procedia - Social and Behavioral Sciences 89 (2013) 895 900

seen, DEA with BCC Model (VRS surface) is chosen for analyzing 51 Turkish public universities
by DEAP-XP software in this study. The findings of the DEA analysis are given in Table 2. We see
that Turkish public universities are efficiency 37% in 2006, 39% in 2007, 47% in 2008, 35% in
2009, 37% in 2010. Balkesir Univ., Dumlupnar Univ., Erciyes Univ., Galatasaray Univ., Gebze
Institute of Technology, Kafkas Univ., Marmara Univ., Nide Univ., Middle East Technical Univ.,
Seluk Univ. are efficiency during the 2006-2010. This analysis focuses on efficiency. It doesnt
focus on performance. So, a public university that has a good performance cant be efficiency
because of high level of inputs. For example; while stanbul Univ. has the decreasing efficiency
scores year by year, Dumlupnar Univ. is efficient during the analysis period.
The efficiency score ranges all lie in the 0-1 interval. DMUs with equal to 1 are efficient and
they determine the efficiency frontier. The others with less than 1 are inefficient and their
inefficiency is calculated by the distance from the efficiency frontier. We could calculate the
potential improving rate for each DMU. Potential improving rate was presented as an example only
three universities. Some of the results of the potential improving rate demonstrate that, in 2006, to
become to be efficient, while Akdeniz Univ. could increase the number of employment to 27.69%,
Celal Bayar Univ. and K. St Imam Univ. could increase the number of employment to
respectively 23.24% and 40.26%. In 2010, to become to be efficiency, Akdeniz Univ. could
increase the number of doctorate per academic to 100% and Celal Bayar Univ. could increase the
number of graduate students per academic to 6.77%, the number of post graduate per academic to
5.66%, number of doctorate per academic to 200%, publication per academic to 6.15%, number of
employment to 5.96% in 2010.
3.2. Results of the panel tobit models with random effects
The DEA efficiency scores estimated in the first part of our study all lie in the 0-1 interval. Panel
tobit model is used in the second part in this study.Hausman test result in Table 3 indicates that
random effects panel model is appropriate. LR test provides a formal test for the pooled Tobit
estimator against the random effects panel estimator. The result of the LR test indicates that the
random effects panel estimator is important. An important test for heteroscedasticity is the Levene,
Brown and Forsythe heteroscedasticity test. The result of the test shows that our models do not
suffer from a heteroscedasticity problem. Therefore the null hypothesis of homoscedasticity is not
rejected. Modified Bhargava et al. Durbin-Watson and Baltagi-Wu LBI autocorrelation and
Friedman's test of cross sectional independence tests indicate no autocorrelation and cross-section
correlation. The Wald test statistics reject the null hypothesis that the parameters in the regression
equation are jointly equal to zero.
Table 3. Results of the panel tobit models with random effects
Independent Variables
Central Government Budget Appropriations
Own Revenue
Project Allocations (TUBITAK)
Project Allocations (Scientific Research Projects)
Total number of academic
Number of graduate students per academic
Number of post graduate students per academic
Number of doctorate students per academic
Employment
Number of publications

Coef.
-0.226
-0.005
-0.019
-0.012
0.000023
0.003
0.012
0.618
0.165
0.000313

Std. Err.
0.052
0.033
0.014
0.014
0.000011
0.001
0.006
0.101
0.029
0.000043

Prob.

-4.360
-0.150
-1.330
-0.860
2.130
2.170
2.040
6.110
5.790
7.360

0.000***
0.880
0.185
0.392
0.033**
0.030**
0.041**
0.000***
0.000***
0.000***

Marginal
effect
-0.226
-0.005
-0.019
-0.012
0.000
0.003
0.012
0.618
0.165
0.000

Sibel Selim and Sibel Aybarc Bursalioglu / Procedia - Social and Behavioral Sciences 89 (2013) 895 900

Independent Variables
Constant
Likelihood-ratio (LR) test (2(01))
Wald (2(10))
Rho ()
Log likelihood
Number of obs

Coef.
2.711
24.31
171.20
0.480
378.453
255

Std. Err.
0.328

t
8.260

Prob.

Marginal
effect

0.000***
0.000
0.000

Hausman test : 2(9) = 11.73

Prob> 2 = 0.1209
Levene, Brown and Forsythe Heteroscedasticity Test : W50 = 0.84710338 df(50, 204) Pr > F = 0.75302822
Autocorrelation : Modified Bhargava et al. Durbin-Watson = 1.9895783 Baltagi-Wu LBI = 2.1636916
Cross-sectional independence : Friedman's test of cross sectional independence = 21.788, Pr = 0.9998
***p<.01, **p<.05, *p<.10.

The marginal effect results of the model in Table 3 demonstrate that, the effect of project
allocations and own revenue found to be insignificant. Additionally, number of students per
academic has a positive and important effect on relative efficiency of universities in Turkey.
Because the central government budget appropriations has a negative effect on efficiency, the
higher education institutions in Turkey may lead to the search for alternative financing such as
Triple Spiral model in co-operation with the private sector, R&D support, project support,
counseling, within the framework of university-industry-government. Employment and number of
publications have positive effect on efficiency. Tobit model with random effects results estimated
using the STATA MP10 software.
4. Conclusion
This paper builds on a two-stage DEA total efficiency approach to determine impact factors on
51 public universities efficiency in Turkey in 2006-2010. DEA is chosen for analyzing 51 Turkish
public universities in this study. This analysis indicate that Turkish public universities are efficiency
37% in 2006, 39% in 2007, 47% in 2008, 35% in 2009, 37% in 2010. Ordinary Least Square is not
an appropriate method to determine factors on the efficiency of public universities in Turkey. The
observed dependent variable may either be zero or positive. The data are censored in the lower tail
of the distribution. Therefore panel tobit model is used in this study. The marginal effect results of
the panel tobit model with random effects demonstrate that, employment and number of
publications, as expected, have positive effect on efficiency. Because the central government budget
appropriations has a negative effect on efficiency, the higher education institutions in Turkey tend to
the search for alternative financing such as Triple Spiral model in co-operation with the private
sector, R&D support, project support, within the framework of university-industry-government.
References
Altnsoy, S. (2011). A review of university facilities in Turkey, CELE Exchange 2011/6, OECD.
Cooper, W.W., Seiford, L.M. & Tone, K. (2000). Data envelopment analysis. USA: Kluwer
Academic Publishers.
Golany, B., Roll, Y. (1989). An application procedure for DEA, OMEGA, 17( 3), 237-250.
Girgin, M. C. (2006). History of higher education provision for the deaf in Turkey and current
applications at the Anadolu University. The Turkish Online Journal of Educational Technology
TOJET, ISSN: 1303-6521, 5(3), 6-11.
Sibiano, P. & Agasisti, T. (2011), Efficiency of public spending in education: A challenge among
Italian regions. Investigaciones de Economa de la Educacin 6, 503-516.

899

900

Sibel Selim and Sibel Aybarc Bursalioglu / Procedia - Social and Behavioral Sciences 89 (2013) 895 900

Toth, R. (2009). Using DEA to evaluate efficiency of higher education. Applied Studies in
Agribusiness and Commerce Conference Papers.