Inserire qui il logo del proprio ateneo d’appartenenza! Do we see economies of scale in universities? (or: differentiate, not merge at all cost) Andrea Bonaccorsi, University of Pisa Cinzia Daraio, University of Pisa Léopold Simar, Institute of Statistics, UCL Tarmo Raty, VATT Finland Outline • • • • • • Introduction Background Data Methodology Preliminary results Further developments Economies of scale Key issue – Widespread belief among policy makers that increasing returns and critical mass effects are at place in universities – Large debate on assumed European “fragmentation” in the university landscape – Arguments: (a) economies of scale (b) economies of variety (Jacob) However, empirical evidence is ambiguous – Brinkman (1981), Brinkman and Leslie (1986), Cohn et al. (1989), de Groot, McMahon and Volkwein (1991), Nelson and Hevert (1992) and Lloyd, Morgan and Williams (1993) – Verry and Layard (1975), Verry and Davies (1976) and Adams and Griliches (1998) – Narin and Hamilton (1996), Abbott, M., & Doucouliagos, C. (2003) – Bonaccorsi and Daraio (2004,2005), Bonaccorsi, Daraio and Simar (2006, 2007) Important practical policy implications • aggregation of universities (e.g. Australian government in the ‘90s; current debate in UK and other EU countries on critical mass); • aggregation of institutes in large public research organisations (e.g. CNRS in France, CNR in Italy). Background • Most empirical investigations done on a country base, at university level or on specific (but limited) subjects (e.g. Brinkman, P.T,. & Leslie, L.L. (1986), Athanassopoulos, A.D., & Shale, E. (1997), Beasley, J.E. (1990, 1995), Flegg, A.T., Allen, D.O., Field, K., & Thurlow, T.W. (2004), Fandel, G. (2007) ) • Lack of systematic comparisons across countries at discipline level: – Microdata not easily available – Comparability issues are important (Bonaccorsi, Daraio, Lepori, Slipersaeter, 2007) • Multi-output production should be taken into account explicitly • Any sensible efficiency analysis should take into account the discipline-wise structure Data and empirical background Data aqua M E T H Aquameth coverage July 2007 Reasonable sample, July 2007 aqua Data METH Aquameth coverage November 2007, including France (487 universities) NUMBER OF UNIVERSIT. COUNTRY Last year availab. FINLAND 20 FRANCE 93 GERMANY 72 HUNGARY 16 ITALY 79 NETHERLAND 13 NORWAY 4 PORTUGAL 14 SPAIN 48 SWITZERLAND 12 UNITED KINGDOM 116 Methodology: Robust Nonparametric efficiency analysis Advantages of Robust nonparametric techniques vs. conventional production function – No need for functional specification – No assumptions on the elasticity of substitution between inputs – Capture local effects as opposed to estimation of average tendency – Inclusion of external factors in a general way Introducing conditional efficiency: an illustration Qzm = Ratio between Conditional and Unconditional efficiency Region of decreasing pattern of ratios: Z has a negative influence If the ratios =1 then Z has no influence on the efficiency 1 Region of increasing pattern of ratios: Z has a positive influence external factor Z Empirical analysis • 4 countries offer data by discipline: Finland, Italy, Norway and Switzerland (later UK, now also Netherlands and a subsample of Germany) • Limited time span: preliminary analysis on the year 2002 (sensitivity analysis) • Outputs: Number of enrolled students; Number of graduates; Number of publications • First take a look at simple output to input ratios and how they vary – Scatter plots of two ratios show whether they are correlated and there are country-wise patterns • Conjoint production model to measure the impact of the university size on teaching and research efficiency. Engineering and Technology Engineering and Technology Field Publications •In Italian Engineering schools publication and graduate intensities go hand in hand. 1,2 Finland Italy Norway Switzerland 1 0,8 •In other countries the relation appears opposite, but the range in graduates is too small and single units dominate the view 0,6 0,4 0,2 0 0 1 2 3 4 5 6 Graduates •Publications/academic staff •Graduates/academic staff In Italy also contracted employees are counted 7 •If the “outliers” are removed, the figure is quite unique. Medical sciences •Graduation has similar patterns in Finland, Switzerland and Norway • Research and education seems to go hand in hand Publications/academic staff Graduates/academic staff Natural Sciences • No country-wise pattern • Performance gains in joint research and teaching are weak • No clear picture of the overall relation Publications/academic staff Graduates/academic staff SocHum •Publication rate can be at any level, regardless of the student population • Independent of the country •ISI cover just a small portion of the peer reviewed literature in this field. Preliminary Comments • Importance of systematic international analysis discipline-wise: subject mix matters • The usefulness of robust conditional measures to summarize overall effects • Too early to draw any policy conclusions • Next Steps – UK universities, first estimates and then NL and a subsample of G Engineering and Technology – Whole sample (2002) Descriptive S tatisti cs N ACTOT 151 M inimum 8 M aximum 3856 M ean 228,84 Std. Deviat ion 388,844 ENR 151 10 38842 3757,01 4973,711 GRA 151 7 8250 1119,62 1007,630 TSTAFF 151 248 10249 2235,68 1819,737 PUB 151 0 904 102,84 136,407 PUB_AC 151 ,00 5,03 ,6091 ,67971 GRA_AC 151 ,08 68,45 10,3472 11,97422 Valid N (listwise) 151 Social Science and Humanities –Whole sample (2002) Descriptive Statistics N ACTOT 176 Minimum 11 Maximum 2210 Mean 421 Std. Deviation 324 ENR 176 214 89092 10265 11015 GRA 176 9 9654 903 1330 TSTAFF 176 48 10249 2057 1729 PUB 176 0 834 69 111 PUB_AC 176 0 3.301 0.168 0 GRA_AC 176 0.03 16 3 3 Valid N (listwise) 176 Economies of scale ENGTECH 3 2.5 But still most universities are in the region of flat conditional efficiency Qzm 2 Overall positive effect of scale (number of students) 1.5 1 Some weak evidence of diminishing returns is also present 0.5 Mod Conj. PUB TEACH 0.5 1 1.5 ENR ENGTECH 2 2.5 4 Economies of scale ENGTECH 4 3.5 3 Qzm 2.5 2 1.5 1 0.5 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Economies of scale ENGTECH - UK 3.5 3 No evidence whatsoever of scale economies Qzm 2.5 2 1.5 1 0.5 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 values of Z Mod Conj. PUB TEACH ENR Economies of scale ENGTECH - UK 3.5 3 Qzm 2.5 2 1.5 1 0.5 1000 2000 3000 4000 5000 6000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Economies of scale ENGTECH - IT 1.7 1.6 1.5 Inverted U-shaped relation 1.4 Qzm 1.3 1.2 1.1 1 0.9 0.8 0.5 1 1.5 2 values of Z Mod Conj. PUB TEACH 4 x 10 ENR Economies of scale ENGTECH - IT Effect of Z on Order-m frontier 1.6 1.4 Qzm 1.2 1 0.8 0.6 0.4 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Economies of scale SOCHUM 3 2.5 Very small Faculties are sub-optimal Qzm 2 1.5 1 0.5 0.5 1 1.5 2 2.5 3 3.5 4 values of Z Mod Conj. PUB TEACH 4.5 4 x 10 ENR Economies of scale SOCHUM 4 3.5 3 Qzm 2.5 2 1.5 1 0.5 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Economies of scale SOCHUM UK 1.6 1.4 Qzm 1.2 1 0.8 0.6 0.4 0.2 2000 4000 6000 8000 10000 12000 14000 values of Z Mod Conj. PUB TEACH ENR 16000 Economies of scale SOCHUM UK 3 2.5 Qzm 2 1.5 1 0.5 1000 2000 3000 4000 5000 6000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Economies of scale SOCHUM IT 2 1.8 Qzm 1.6 1.4 1.2 1 0.8 1 2 3 4 5 6 7 values of Z Mod Conj. PUB TEACH 8 4 x 10 ENR Economies of scale SOCHUM IT 2 Qzm 1.5 1 0.5 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 values of Z Mod Conj. PUB TEACH TSTAFF UNI Conclusions on scale and efficiency • Economies of scale should not be examined at the level of universities at aggregate level • Differentiated pattern by discipline • Also some country-level differences emerge No empirical support for a generalized policy of pressure on universities to grow or merge Rather, each scientific/ educational field must find its own “optimal” scale Policies of concentration/ merger should be aimed at helping universities to find their own optimal configuration among disciplines, each of which follows differentiated patterns University as a strategic multi-divisional agent The key to strategic behaviour is differentiation Differentiation of European universities in PhD education • PhD education crucial in knowledge society • Internationalization and mobility • Competition • Institutional adaptation • Differentiation and “division of academic labor” as response to enlargement of the market and competition Variable observed Number of graduate students/ Number of undergraduate students (ratio) Institutional differentiation Entropy measure h(pi) = log (1/ pi) H= n pi log (1/ pi) Weitzmann’s diversity V(Z) = max ( V(Z\x) + d (Z\x, x)) x Z Mean sum of squared distance SSD = n i=1 n j=1 (wi – wj/ w^)2 n n MSSD = 1/n2 (wi – wj/ w^)2 i=1 j=1 PhD intensity. Netherlands. Year 1994-2004 0,05 0,045 PhD intensity 0,04 0,035 0,03 0,025 0,02 0,015 0,01 0,005 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Differentiation in PhD intensity. Netherlands. Year 1994-2004 160 Differentiation index 140 120 100 80 60 40 20 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 PhD intensity. Finland. Year 1994-2005 0,135 PhD intensity 0,13 0,125 0,12 0,115 0,11 0,105 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Differentiation in PhD intensity. Finland. Year 1994-2005 Differentiation index 160 140 120 100 80 60 40 20 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 PhD intensity. Switzerland. Year 1994-2003 0,18 0,175 Phd intensity 0,17 0,165 0,16 0,155 0,15 0,145 0,14 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Differentiation in PhD intensity. Switzerland. Year 1994-2003 80 70 Differentiation index 60 50 40 30 20 10 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 PhD intensity. United Kingdom. Year 1996-2003 0,25 PhD intensity 0,2 0,15 0,1 0,05 0 1996 1997 1998 1999 2000 2001 2002 2003 Differentiation in PhDintensity. United Kingdom. Year 1996-2003 10000 9000 8000 Differentiation index 7000 6000 5000 4000 3000 2000 1000 0 1996 1997 1998 1999 2000 2001 2002 2003 PhD intensity. Spain. Year 1994-2002 0,06 PhD intensity 0,05 0,04 0,03 0,02 0,01 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 Differentiation in PhD intensity. Spain. Year 1994-2002 1400 Differentiation index 1200 1000 800 600 400 200 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 PhD intensity. Italy. Year 2001-2005 0,02 0,018 0,016 PhD intensity 0,014 0,012 0,01 0,008 0,006 0,004 0,002 0 2001 2002 2003 2004 2005 Differentiation in PhD intensity. Italy. Year 2001-2005 3500 Differentiation index 3000 2500 2000 1500 1000 500 0 2001 2002 2003 2004 2005 Comparative analysis of PhD intensity 0,25 PhD intensity 0,20 Switzerland Netherlands Finland Spain Italy United Kingd 0,15 0,10 0,05 0,00 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Comparative analysis of differentiation in PhD intensity Normalized differentiation index 0,900 0,800 0,700 0,600 0,500 0,400 0,300 0,200 0,100 0,000 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Switzerland Netherlands Finland Spain Italy United Kingdom Conclusions • Universities must learn to compete in an international environment • To compete, you need a strategy • The name of the game is strategic differentiation • by scale and scope • by subject mix • by main type of education (undergraduate, professional master, research training) • by ambition in research (regional producer of usable knowledge; average research producer; world class research university) • by interactions with stakeholders (proximity vs international; industry vs territory/society) • by funding mix • To have a strategy you need indicators of positioning and of competitive dynamics (not only rankings)