The efficiency of German universities–some evidence from nonparametric and parametric methods

Abstract

Due to tight public budget constraints, the efficiency of publicly financed universities in Germany is receiving increasing attention in the academic as well as in the public discourse. Against this background, we analyse the efficiency of 72 public German universities for the years 1998–2003, applying data envelopment and stochastic frontier analysis. Contrary to earlier studies, we account for the faculty composition of universities which proves to be an essential element in the efficiency of higher education. Our main finding is that East German universities have performed better in total factor productivity change compared to those in West Germany. However, when looking at mean efficiency scores over the sample period, West German universities still appear at the top end of relative efficiency outcomes.

Acknowledgements

We thank Martin Beck, Federal Statistical Office of Germany, for providing the data as well as for his helpful remarks. Moreover, we acknowledge support from Christian von Hirschhausen, Helmut Seitz, Marcel Thum and especially Astrid Cullmann and Borge Hess for valuable comments and intensive discussions. We thank the participants of the 1st Halle Efficiency and Productivity Analysis Workshop (HEPAW), June 20–21 (2006), especially Tim Coelli and Matthias Staat for very helpful remarks.

Notes

¹ University finance is a major responsibility of the German states, which account for about 90% of spending on higher education. The federal level contributes about 10% to university finance while the local governments are not involved at all.

² German reunification complicates pre-1990 and post-1990 comparisons. In addition, in 1997 a major change in the university statistics was introduced that makes it virtually impossible to compare data from the period before 1997 to data after 1997.

³ Worthington (2001) provides an extensive review on the efficiency analyses of universities.

⁴ To some degree, accounting for faculty composition is possible within the DEA by distinguishing between arts/sciences in outputs (e.g. publications or graduates) and/or inputs (e.g. students), see Warning (2004). However, there is a limit to extending the number of inputs/outputs; introducing additional inputs/outputs reduces the number of benchmark universities and, as a result, an excessive number of universities will be indicated as efficient.

⁵ Previous investigations also specified the number of publications, in some cases weighted by journal ranks or the number of pages. However, due to data availability we did not include publications in our investigation. The number of students has been used ambiguously as an output of teaching activities or as an input into teaching production.

⁶ See Banker et al . (2004) for a detailed presentation of the model.

⁷ See Coelli et al . (2005) for a detailed presentation of the Malmquist index.

⁸ Note that all three universities of Berlin are considered here as universities located in East Germany.

⁹ On the local level, Germany is divided into 434 regional authorities. Yet, a distinction has to be made since only about 75% of them can be characterized as districts (‘Kreise’), comprising rural areas as well as villages and smaller cities. In contrast, the remaining quarter can be referred to as larger cities (‘Kreisfreie Städte’). This institutional peculiarity makes local GDP per capita a poor proxy for the location effect that we want to test because local GDP per capita is systematically biased downwards in cities that include surrounding areas. Thus, we choose regional GDP per capita at the level of ‘Raumordnungsregionen’, which is supplied by the BBR (2006).

¹⁰ Simar and Wilson (2007) indicate that results from the second stage, i.e. the regression of productive efficiency on environmental variables, might be subject to serial correlation within the efficiency estimates. They propose a double bootstrapping procedure that improves statistical efficiency in the second-stage regression. However, in our study the second-stage regression is accompanied by the estimates of the Battese and Coelli (1995) model, which can be considered as a sensitivity check to the results from the Tobit regression (Section V).

¹¹ We choose to estimate a cost function compared to a distance function as we prefer to interpret the coefficients–especially the coefficients from the faculty controls–in the intuitive context of a cost function.

¹² Unfortunately, we are not able to include a proxy for incoming students’ educational background. This information is impossible to obtain in Germany as there is no general admission test procedure.

¹³ Truncated at zero to display inefficient performances ‘above’ the estimated cost function, see also Coelli (1996).

¹⁴ Universities with medical faculties are associated with a university hospital. The only exception is the University of Bochum, which operates with surrounding hospitals (‘Bochumer Modell’). Thus, the efficiency scores of the University of Bochum have to be interpreted with caution.

¹⁵ A 1% increase in third-party funds per student decreases total costs less third-party funds (divided by the number of students) by 0.21% in universities with engineering faculties. In universities with medical faculties, a 1% increase in third-party funds per student increases total costs less third-party funds (per student) by 0.11%.

¹⁶ Some caution is warranted when interpreting efficiency scores from SFA single-stage models such as Battese and Coelli (1995) since these scores are gross efficiency scores (Coelli et al., 1999). However, an efficiency ranking obtained from our baseline model is significantly correlated with efficiency rankings obtained from a Battese and Coelli (1992) model setup by about 80%.