Technical efficiency and productivity for higher education institutions in Sweden

ABSTRACT

This study investigates technical efficiency and productivity for Swedish higher education institutions (HEIs). One identified problem in previous research concerns adjusting efficiency scores for input quality. This problem is avoided using grades from upper-secondary schools. A second problem concerns heterogeneity with respect to subjects and institutions between HEIs. Using the Swedish national resource allocation system, students are weighted according to subject. For research production, a bibliometric index that allows for differences in publication tradition is used. A third problem when using the data envelopment analysis approach is the lack of statistical inference. Bootstrapping is used to approach this problem. The results indicate an average inefficiency of 12% and a productivity increase of around 1.7% per year.

KEYWORDS:

• Efficiency
• productivity
• data envelopment analysis (DEA)
• bootstrap estimation
• higher education institutions

JEL CLASSIFICATIONS:

• C15
• D24
• I21

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

¹Approximately €5.39 billion.

²OECD (2010).

³Studies at the Swedish undergraduate and advanced level correspond to bachelor and master level studies, respectively.

⁴All HEIs receive the same funding for a student enrolled within a certain disciplinary domain. On average, the funding per HST and HPR is the same. The exact funding for each disciplinary domain and the funding cap for each HEI is presented in the government approval document for each year.

⁵Other possible methodological approaches include deterministic parametric frontier (see, e.g., Bjurek, Hjalmarsson, & Førsund, 1990; Månsson, 2006) and Stochastic Frontier Analysis (see, e.g., Thanassoulis et al., 2011).

⁶The input-based efficiency measure express inefficiency as potential input savings given output.

⁷See, for example, Färe and Grosskopf (1994).

⁸Limitations of the DEA method are discussed in, for example, Dyson et al. (2001) and Mettas, Vargas, and Whybark (2001).

⁹To our knowledge the DEA-bootstrap method has not been applied for measuring efficiency of HEI's. However, see Essid, Ouellette, and Vigeant (2013) for a bootstrap application using DEA on high school efficiency.

¹⁰See also Färe, Grifell-Tatjé, Grosskopf, and Knox Lovell (1997).

¹¹There are also studies that examine efficiency among departments within a university. See for example Köksal and Nalçaci (2006) and Tyagi, Yadav, and Singh (2009).

¹²This index builds on a questionnaire sent out to HEIs where they were asked to rank the influence of the research carried out by other institutions.

¹³Johnes (2006).

¹⁴The study uses Stochastic Frontier Approach, which is a statistical rather than a mathematical approach to study efficiency.

¹⁵There are other studies that use a Stochastic Frontier Analysis in analysing HEIs, however, these are on cost efficiency rather than technical efficiency (see, e.g., Izadi, Johnes, Oskrochi, & Crouchley, 2002; Johnes, 2004; Johnes & Johnes, 2009).

¹⁶The DEA framework has, however, been used on Swedish data analysing efficiency in, for example, public employment offices (see, for example, Andersson, Månsson, & Sund, 2014) and public child care (see Bjurek, Gustafsson, Kjulin, & Kärrby, 2006).

¹⁷For 2005–2007, the number of HEIs is 31 because the Stockholm Institute of Education (SIE) is included. In 2008, the SIE merged with Stockholm University.

¹⁸In addition, 10 minor institutions have been excluded since they only produce education and no research.

¹⁹Our hope was to make a distinction between different types of teaching and research staff, however, the number of observations made this impossible. We have included a second-stage analysis to at least give some information on its importance.

²⁰The staff categories included in this measure are professors, research assistants, senior lecturers, lecturers, other research or teaching staff, visiting staff, part-time fixed-term lecturers, and technical and administrative staff with teaching and/or research duties.

²¹Included in tangible assets are for example buildings, land, machines, inventory, and installations.

²²To the tangible assets we have added the Avtal om läkarutbildning och forskning (ALF) and tandläkarutbildningsavtalet (TUA) compensations that some universities receive because of their involvement in the education of physicians and dentists. ALF is an agreement for funding of medical training and research. TUA is an agreement for funding of dental training and research. Seven HEIs receive these extra resources and they are University of Gothenburg, Karolinska Institute, Linköping University, Lund University, Malmö University College, Umeå University, and Uppsala University. The sum of tangible assets and ALF- and TUA-compensations constitutes the variable other resources. This variable reflects the physical capital in which the HEIs have invested. The variable other resources is denominated in 2008 prices. The consumer price index has been used as deflator.

²³The baseline for the weights is social science and humanities, which is set to 1. Students in other educational fields, for example, science, have according to the Swedish resource allocation system different resource allocation factors that vary between 1 and 14. The adjusted HPR output is defined as the share of students within different educational fields at a specific HEI multiplied by the resource allocation factor. This implies that the HPR output for an HEI with more resource intensive education is weighted higher than an HEI with education within social sciences and humanities. It also implies that the weighting factors can be outside the (0,1) interval. This is intended since a HEI with very high resource allocation factors has to be weighted with a factor that is larger than 1.

²⁴See Kronman, Gunnarsson, and Karlsson (2010).

²⁵The indicator contains nine research fields.

²⁶Data for the bibliometric publication indicator is available up until 2009, which implies that one year is missing from the data needed to construct a moving average for 2008. The average trend for the years 2004 to 2008 is therefore used to impute a bibliometric indicator for 2010, which is then used in the moving average calculation for 2008.

²⁷See Dyson et al. (2001) for a discussion of model dimensions and number of observations.

²⁸All sensitivity analyses are available from the authors on request.

²⁹It is important to note that we do not claim to do any causal analysis since the knowledge of how such an analysis should be performed is still under discussion (see, e.g., Simar & Wilson, 2011).

³⁰Bonaccorsi and Daraio (2007).