Cost structure, productivity and efficiency of the Italian public higher education industry 2001–2011

Abstract

In this paper, I analyze the cost structure of the Italian higher education system for the decade between 2001 and 2011, by means of a stochastic translog cost function. I suggest that the judgment about the optimal configuration of the sector is strongly dependent upon the policy priorities set by decision makers. When assuming that the universities’ output is the number of students, scale economies are exhausted, and marginal costs are relatively low; when considering graduates as outputs instead, there is opportunity for increasing the scale of operations. Inefficiency affects production in a sensible manner, especially when assuming that the target output is the number of graduates. Moreover, efficiency contributes to explaining a relevant portion of the productivity increases in the period. No significant scope economies between teaching and research emerge, suggesting that a higher degree of universities’ specialization can be a direction for improving the sector’s efficiency and productivity.

Keywords:

• higher education industry
• costs
• efficiency
• productivity
• stochastic frontier

JEL Classifications:

• L38
• I21
• I23
• I28
• L32

Acknowledgments

I am grateful to the Teachers College (Columbia University) and the Institute for Education and Social Policy (New York University), which hosted me during the writing of this paper’s first draft (Summer 2013). I also thank H2CU College Italia, which provided me with accommodation services during this period. I acknowledge the assistance of M. Lezzi in preparing the dataset and in conducting some preliminary statistical elaborations. Some colleagues provided very useful comments, and I am indebted to them: G. Johnes, K. de Witte, C. Haelermans, E. Sneyers. Any errors are obviously my sole responsibility.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. The relative weaknesses and strengths of DEA and SFA are described in several academic manuals: the interested reader can refer to Fried, Lovell, and Schmidt (2008).

2. To the extent that inputs are well reflected into costs, the two concepts of cost efficiency and technical efficiency (net of the influence of prices) are equivalent – in this setting, there are not differentials due to allocative efficiency. However, the approach used in this paper is different as we use the number of students and graduates as alternative outputs, and never as inputs.

3. The consequences of these misspecifications are described and discussed by Gronberg, Jansen, and Taylor (2011) with an application to the case of public schools in Texas.

4. There is a relevant debate in the literature about the functional form that should be preferred for the cost function; the interested reader should refer to Cohn and Cooper (2004). Therefore, the translog specification has several advantages due to its flexibility: see the theoretical discussion on this point in Caves, Christensen, and Tretheway (1980).

5. Actually, the translog complete specification of the cost function should be the following:

(3a)

where p are inputs’ prices. However, in the Italian context, the prices for higher institutions are homogenous, as the Ministry of education centrally determines staff salaries, and capital expenditures are not included in this empirical exercise, and are assumed to be captures by C_it (see the hypotheses in this section).

6. As the translog specification does not tolerate zero values in the data-set, I substituted the zeros (universities without medical, scientific or other students) with 0.001.

7. The Italian universities still used at that time financial accounting techniques (the introduction of accrual accounting happened in 2013); as a consequence, costs should actually be intended as expenditures, and no information is available about capital stock. This is a further reason to consider equation (3(3) ) net of the component for capital expenditures.

8. The most recent theoretical literature highlights the importance of considering, among outputs, the so-called ‘third mission’ of universities, namely the technology transfer initiative, and the knowledge services produced for the territory and society at large. Unfortunately, the availability of good indicators for measuring such a dimension is still very limited; to the best of the author’s knowledge, the only study that considers such indicators in empirical studies about universities’ efficiency is Johnes, Johnes, and Thanassoulis (2008), Therefore, no university-level data are at our disposal for the Italian context, so the third mission is not considered in this paper.

9. A similar approach has been used by Agasisti and Haelermans (2013) when comparing the efficiency of Italian and Dutch universities.

10. I calculated the correlations between the number of students and the number of graduates per student (calculated as #graduates/#enrolments), by year and subject, and they are indeed negative.

11. Technically, all the students are ‘regular’ until they do not exceed the legal duration of a course; so a student in her second year of a bachelor-level course, attending courses of the first year, is still considered as ‘regular’.

12. Also, I did not consider the number of PhD students, because the mixed nature of their tasks (attending but also giving classes, participation in research) makes it difficult to classify them more as an input (captured by costs) or as an output. In other words, they receive didactical services – for instance for the courses they attend – as well as using research facilities, so can be considered as outputs in the sense used in this paper; nevertheless, they also give classes and produce research, so they are inputs at the same time.

13. The dependent variable (‘recurrent costs’) also includes research grants. In this sense, the estimation of the efficiency differentials is driven by variation in variables related to teaching. However, the inclusion of the (monetary) variable about the amount of grants (thousands €) among outputs is necessary for estimating marginal and average costs; otherwise, the unit costs for teaching activities would be overestimated.

14. A specific feature of defining the research output this way is that the indicator is partially able to capture the third mission, via the component of grants obtained from industries and applied research purposes, albeit in this paper I cannot explicitly separate the two components.

15. It is important to note that, given the non-linearity of the coefficients, there is no chance to estimate directly the elasticity from single coefficients of the cost function, although both costs and outputs enter the function as logarithms.

16. In Italy, the number of students who can be admitted to Medical Studies is strictly regulated by a national law, with two consequences: the number of students in this area is not unconstrained and dependent on students’ choices, but on national regulations; second, no major changes intervened in policy-making in this area, so there were no fluctuations in the number of medicine students in the period under study.

17. Therefore, it can also well be the case that the dummy captures some price differentials, for inputs whose prices are not regulated nationally – for example, the rent of buildings, etc. If it is the case, this will not affect the reliability of my results, and even it reinforces the reliability.

18. Although it is not the focus of this paper, it is interesting to speculate about why the numbers are so different between Italy/Spain and Australia/England. One possible explanation is that a higher level of inefficiency characterized the former group, so that it was easier for these universities to obtain efficiency gains than for their Australian and English counterparts. While I do not have data available for testing this hypothesis, Agasisti and Johnes (2009) showed that English universities are, on average, significantly more efficient than Italian ones.

19. I do not specifically discuss the model where the number of regular graduates is the output; while the general patterns are substantially similar, the number of regular graduates was very low in early 2000s, and the resulting unit costs implausibly high at that time.

20. This remark is particularly relevant for those universities that have no medical schools, and for those who are specialized in scientific fields (i.e. Politecnici).

21. A noteworthy exception is that of unreasonable high MC for graduates in medicine (the cause resides in the very low relative number of graduates in this discipline). A further confirmation of this anomaly is that the cost function is not able to provide meaningful estimates of elasticity of size for medical students and graduates for the very small universities (25th percentile; note that elasticity is computed to be negative). In addition, nonlinearities in the costs for medical education can be part of the explanation of this phenomenon.