Governance reforms in HE

Many scholars have underlined the apparent convergence towards the “steering at a distance” mode in HE in recent decades (Braun and Merrien Reference Braun and Merrien1999; Huisman Reference Huisman2009; Paradeise et al. Reference Paradeise, Reale, Bleiklie and Ferlie2009; Shattock Reference Shattock2014). This governance arrangement is characterised by mixing the following instruments together: financial incentives to pursue specific outputs and outcomes in teaching and research, student loans, accreditation, ex post evaluation conducted by public agencies, contract benchmarking and provisions by the law for greater institutional autonomy (Gornitzka et al. Reference Gornitzka, Kogan and Amaral2005; Lazzaretti and Tavoletti Reference Lazzaretti and Tavoletti2006; Maassen and Olsen Reference Maassen and Olsen2007; Trakman Reference Trakman2008; Capano Reference Capano2011; Capano and Turri Reference Capano and Turri2017). However, this convergence certainly works to support general principles (more institutional autonomy, more evaluation and more competition), while the concrete ways through which the policies are generated seem to be quite diverse (Capano and Pritoni Reference Capano and Pritoni2019). Nonetheless, this diversity, at least in terms of the concrete composition of the adopted policy instruments, has never been clarified.

Furthermore, it should be noted that in terms of policy performance, the effects of these governance shifts have not been clearly assessed. In fact, the literature on HESs’ performance has focussed mainly on a few aspects as key determinants of policy success (or failure): the mechanism of funding (Liefner Reference Liefner2003), the total amount of public funding (Winter-Ebmer and Wirz Reference Winter-Ebmer and Wirz2002; Horeau et al. Reference Horeau, Ritzen and Marconi2013; Williams et al. Reference Williams, de Rassenfosse, Jensen and Marginson2013), full institutional autonomy (Aghion and Dewatripont Reference Aghion, Dewatripont, Hoxby, Mas-Colell and Sapir2008), partisan dynamics (Ansell Reference Ansell2008), stratification (Willemse and De Beer Reference Willemse and De Beer2012) and the type of loan system adopted (Flannery and O’Donoghue Reference Flannery and O’Donoghue2011). However, this research strategy simply assesses whether certain variables have the power to influence the probability of the outcome changing as expected on average at the population level, regardless of interactions. These types of explanatory results seem weak and generally risk remaining fairly superficial. The salient point here is that focusing on a single dimension to assess the performance capacity of a governance arrangement is quite misleading. For example, the effects of shifting from a centralised governance system to one in which universities are more autonomous cannot be analysed without contextualising the change within its specific configuration, that is, by considering the other relevant instrumental dimensions (e.g. how universities are funded, the system of degree accreditation, whether a national research evaluation assessment is present, etc.).

Overall, despite the significant governance shifts in HE, there is currently a lack of knowledge on both the actual nature of these changes in terms of existing combinations of adopted instruments and the policy performance of the new governance arrangements.

Governance arrangements as a set of policy instruments

Policies are steered by specific governance arrangements composed of rules, instruments, actors, interactions and values (Howlett Reference Howlett2009, Reference Howlett2011; Tollefson et al. Reference Tollefson, Zito and Gale2012; Capano et al. Reference Capano, Howlett and Ramesh2015). The implicit assumption of the governance literature is that different governance modes or arrangements deliver different results in terms of policy outcomes. However, empirical evidence on this issue has been lacking, especially because the main analytical focus in public policy has been the process of changing governance arrangements in terms of their content with respect to the actors involved, the distribution of power, and the adoption of “new” policy instruments. Thus, there has not been enough focus on the results of these governance shifts in mainstream public policy.

However, there is an increasing awareness that pure types of governance arrangements do not actually work; instead, the main principles of coordination (hierarchy, market and network) are combined in various ways. All governance arrangements are essentially hybrids and are characterised as working through policy mixes, that is, policy instruments belonging to “different” instrument categories or pertaining to different policy paradigms/beliefs/systems/ideologies (Howlett Reference Howlett2005; Ring and Schroter-Schlaack Reference Ring and Schroter-Schlaack2010; Capano et al. Reference Capano, Rayner and Zito2012). Thus, an existing set of adopted policy instruments can be conceptualised as specific portfolios, settings and combinations of different types of policy instruments associated with different working logics (Jordan et al. Reference Jordan, Benson, Zito and Wurzel2012; Schaffrin et al. Reference Schaffrin, Sewerin and Seubert2014; Howlett and Del Rio Reference Howlett and Del Rio2015).

However, how can the content of these policy mixes be described and understood, and how can their policy performance be assessed? In an attempt to fill these theoretical and empirical gaps, we adopt a bottom-up perspective by focussing on the basic unit of any governance mode – the policy instruments that can be adopted – and their possible combinations. This approach seems quite realistic; policy instruments are the operational, performance-related dimensions of governance arrangements.

Accordingly, we operationalise systemic governance arrangements in terms of adopted policy instruments and therefore as specific sets of techniques or means by which governments try to affect the behaviour of policy actors to direct them towards the desired results (Linder and Peters Reference Linder and Peters1990; Vedung Reference Vedung1998; Howlett Reference Howlett2000; Salamon Reference Salamon2002).

There are numerous classifications by which policy tools can be ordered based on different criteria of analytical distinction, from coercion to the type of governmental source adopted (Hood Reference Hood1983; Phidd and Doern Reference Phidd and Doern1983; Ingram and Schneider Reference Ingram and Schneider1990; Vedung Reference Vedung1998; Howlett Reference Howlett2011). All these typologies suggest different families of instruments. Our research framework focuses on the capacity of policy instruments to induce specific behaviours; thus, we need to consider the nature of the instruments and examine the different ways through which they induce action towards the expected result. In conducting this examination, we are inspired by the classical theorisation of Vedung (Reference Vedung1998). When focussing on the nature of substantive policy instruments, Vedung grouped these instruments by the basic inducement on which they relied to foster compliance.

By adopting this perspective, we can delimit four distinct families of substantial policy instruments that have different, nonoverlapping capacities to induce behaviours: regulation, expenditure, taxation and information.Footnote ¹ Each family is associated with a specific inducement. Regulation induces behaviour control; expenditure induces remuneration; taxation – depending on the way it is designed – can engender behaviour control and remuneration; and information offers persuasion. Notably, all four families of tools can be employed by applying different methods of coercion that are dependent on how free individuals are to choose alternatives. Regulation can be quite strong or weak according to the type of behavioural prescriptions provided. Expenditure can lack coercion in the case of subsidies but be very demanding when delivering targeted funding. Taxation can be quite coercive when a general tax increase is established, but it can also have a low degree of coercion when many targeted tax exemptions exist. Information can be coercive when compulsory disclosure is imposed or lack strong coercion when monitoring is applied.

The four families of substantial policy instruments we have decided to consider, as well as the types proposed by other policy instrument classifications, represent very general instrumental principles (which need to take specific forms to be practically applied): they are “families of individual instruments sharing similar characteristics” (Bouwma et al. Reference Bouwma, Liefferink, Apeldoorn and Arts2016, 216).

Thus, according to Salamon (Reference Salamon2002), the “shape” through which the substantial instrument is designed to deliver the expected result is the important factor in terms of policy impact and potential performance. For every type of family of substantial policy instruments, there are different shapes of delivery that offer actual outlets through which those substantial instruments can affect reality. In addition, these instrumental shapes should be considered the basic analytical unit when assessing how policies are made and, thus, which types of governance arrangements actually work in terms of policy performance.

Accordingly, the important factors in detecting the adoption of a regulation instrument are the various forms through which regulations can be designed: for example, by imposing a specific behaviour, enlarging the range of opportunities or establishing specific public organisations. Expenditures can be delivered through grants, subsidies, loans, lump sum transfers and targeted transfers, among other shapes. Taxation can imply fees, user charges and exemptions, among others. Information can take the shape of neutral administrative disclosures, monitoring, diffusion, etc.

Each of these instruments’ shapes carries specific potential effects that cannot be measured alone because they should be considered in relation to the other tools that compose the actual set of adopted policy instruments. Understanding the distinct shapes that various types of substantive policy instruments can take when delivered is essential to grasping how governments change the instrumental side of governance arrangements over time.Footnote ² There are three dimensions in which this distinction is helpful.

The first dimension is descriptive. By focussing on the different shapes of policy instruments, the usual description can acquire a more detailed reconstruction of shifts in governance compared to the usual distinction of more or less market, more or less hierarchy, more or less regulation, more or fewer expenditures, etc.

The second dimension is analytical. Due to the focus on the basic shapes through which substantial instruments are delivered, the policy mix concept can be operationalised in a very effective and realistic way. Thus, the eventual relevant differences can be assessed in terms of policy settings because the content of policy mixes can be grasped in a very detailed manner.

The third dimension is clearly exploratory, with a particular emphasis on the possible explanatory relevance of policy instruments for performance. If policy performance is assumed to be conditioned by – among other factors – the adopted set of policy instruments, a detailed operationalisation of the substantial instruments should uncover which combinations of instruments actually work, which can lead to a better reconceptualisation of the (many) determinants of policy performance and of their interactions.

This perspective seriously considers the suggestions of those scholars who have observed how the actual set of adopted policy instruments is the consequence of a diachronic accumulation. Thus, we need to analyse the full package instead of a single type of policy instrument (Hacker Reference Hacker2004; Pierson Reference Pierson2004; Huitema and Meijerink Reference Huitema and Meijerink2009; Tosun Reference Tosun2013; Schaffrin et al. Reference Schaffrin, Sewerin and Seubert2014).

Therefore, from a theoretical perspective, the decision to focus on the instrumental side of governance arrangements to measure their potential impact on policy performance encourages proceeding by using combinatory logic: when the expected effect, namely, policy performance, is assumed to depend on – among other things – a combination of multiple conditions (i.e. specific settings of policy instruments), then a complex causality principle is at work. For this reason, we decided to turn to a Qualitative Comparative Analysis (QCA).Footnote ³ QCA is a configurational and set-theoretic approach in which relationships of necessity and sufficiency are tested and the idea of causality underpinning the approach is fundamentally characterised by equi-finality,Footnote ⁴ conjunctural causationFootnote ⁵ and asymmetry.Footnote ⁶ Therefore, QCA aims at unravelling multicausal rather than mono-causal explanations, focusses on combinations of conditions rather than on single variables, and does not assume that a unique solution (or equation) accounts for both the occurrence and nonoccurrence of a particular outcome.

Furthermore, by applying this method, we can precisely assess which instruments or combinations of instruments are present when better performance is achieved. By using this method of operationalising and methodologically treating policy instruments, a relevant analytical gap can be filled, producing a more precise design for research on the determinants of policy performance; these determinants are too often based on the direct effects of structural and contextual variables (comprising factors such as the socio-economic situation, social capital and the political situation) or processual variables. Overall, policymakers change policy by choosing specific policy instruments, and thus, thanks to the adopted method, it is possible to better describe the content of these decisions. Showing which combinations of instruments are related to specific policy performances plays an enlightening role in readdressing the research on policy evaluation and improving the analysis of the links between policies and outcomes.

Case selection and timespan

This article is based on a specific data set of policy tools used in 12 Western European countries (Austria, Denmark, England, Finland, France, Greece, Ireland, Italy, the Netherlands, Norway, Portugal and Sweden) between 1995 and 2014.

Regarding country selection, we initially intended to cover all the pre-2004 enlargement EU countries. We decided to exclude Eastern European countries due to the period of transition they experienced after 1989, a time of deep turmoil that mixed the communist legacy, the return to precommunist governance and a kind of acceleration towards marketisation (Dobbins and Knill Reference Dobbins and Knill2009); these characteristics make it quite difficult to assess and code the characteristics of the adopted policy instruments.

However, we were forced to exclude, for different reasons, four additional countries: Luxembourg, due to its small size (one university); Belgium and Germany because of their federal structures and Spain because of its very decentralised regionalism, which has a significant impact on the systemic governance of HE. These 11 pre-2004 enlargement countries reflect all historical types of university governance that have developed in Europe and can therefore offer sufficient differentiation in terms of policy legacy (Clark Reference Clark1983; Braun and Merrien Reference Braun and Merrien1999; Shattock Reference Shattock2014) and the inherited set of policy instruments. We also included a non-EU country, Norway. Thus, all the Nordic countries, which are assumed to have adopted a welfarist approach to HE, could be considered, and it was possible to examine whether this common characteristic influenced the analysed outcome.

In all the selected countries, the HESs have undergone structural changes in the last two decades. Accordingly, we decided to begin our analysis in approximately the mid-1990s to encompass all the major changes that involved HESs over the last 20 years.Footnote ⁷ Obviously, each country presents its own reform “starting point” in the field, which means that some of the countries had already produced relevant legislation by the mid-1990s, while others began much later.

Operationalisation

As already explained in the theoretical section, we assumed that differences in HESs’ teaching performance may be associated with – among other things – differences in the combinations of the adopted policy tools. Among the various possible indicators of HESs’ performance – such as access, academic recruitment and careers, and third missions – we decided to focus only on teaching, which ultimately represents one of the main tasks of every HE institute. Furthermore, in the last 30 years, all governments have been committed to incentivising universities to pay more attention to the socio-economic needs of their own country and to the need to increase the stock of human capital (currently, in many countries and at the EU level, increasing the number of citizens obtaining a tertiary degree is a major policy goal).

The most common indicator of teaching is the percentage (%) of people with a university-level degree.Footnote ⁸ As such, we operationalised teaching performance starting from the percentage of adults between 25 and 34 years old who have a university-level degree. Specifically, because many countries’ education systems changed between the 1990s and the 2000s (possibly because of the Bologna Process), HE programs differ from those that existed 20 years ago. Thus, according to the OECD data, we chose our starting point as the percentage of the 25- to 34-year-old population with a “university-level” education (level 5A in the ISCED 1997 classification) in 1996 compared to the percentage of people of the same age who had either a bachelor’s or master’s degree in 2015 (levels 6 or 7 in the ISCED 2011 classification).Footnote ⁹ These data were downloaded from the OECD archive (see also OECD Reference Organization for Economic1998, 2016)Footnote ¹⁰ and are summarised in Figure 1.

Source: Our elaboration on Organization for Economic Cooperation and Development data.

Figure 1 University-level (ISCED 5A 1997 – ISCED 6 + ISCED 7 2011) attainment of 25- to 34-year-old adults: 1996 and 2015 in comparison.

However, the variation in this performance indicator does not directly represent our outcome because we could not account for two equally relevant considerations: first, improving results is easier when the starting point is a very low value than when it is a higher value, and second, results of university-level education are strongly linked to the structure of tertiary education as a whole. In other words, all else being equal, countries that offer short-cycle tertiary degrees (i.e. first-cycle degrees lasting less than 3 years: level 5B in the ISCED 1997 classification and level 5 in the ISCED 2011 classification) should be rewarded more than countries without them because in the former case, HE institutions are subject to more competition for students (or, at least, they have a smaller catchment area) and, in turn, are less likely to improve their results. Consequently, we modified the data slightly following a two-step process. In the first step, we differentiated countries into three categories: countries below the mean of university degree attainment (25–34 years old) in 1996, countries above the mean but less than 1 Standard Deviation (SD) above the mean, and countries above 1 SD above the mean. Countries in the latter category had an increase in their performance (i.e. in the difference between their level of university degree attainment in 2015 and in 1996) that was equal to +50%, whereas countries in the intermediate category (above the mean but less than 1 SD above the mean) had an increase in performance that was between 0 and +50%, depending on how much they were above the mean. Finally, countries below the mean did not show an increase.

In the second step, we adopted the same differentiation of countries with respect to the ratio between 25- to 34-year-old adults with a short-cycle tertiary degree and those with any tertiary degree in 2015. In this way, we could weigh the extent to which the opportunity to enrol in short-cycle degrees affected the HES as a whole. Again, countries more than 1 SD above the mean had a further increase in their performance that was equal to +50%, countries above the mean but less than 1 SD above the mean had an increase in performance that was between 0 and +50%, depending on how far above the mean they were, and countries below the mean did not have an increase. Table 1 presents both the original OECD data and the country rankings based on our measure of teaching performance from 1996 to 2015.

Table 1 Construction of the outcome (teaching performance between 1996 and 2015)

Source: Our elaboration on Organization for Economic Cooperation and Development data.

The operationalisation of the conditions of the next QCA, namely, the policy instruments, required greater theoretical reflection and greater effort in data gathering, as we will explain below. More precisely, we decided to operationalise the four families of substantial policy tools – regulation, expenditure, taxation and information – while considering a long list of shapes (24 in total), which are presented in Table 2.

Table 2 Classification of policy instruments and their shapes

In this way, we tried to capture all the possible shapes that substantial HE policy instruments can take. We also avoided constructing categories that were too exclusive, which would have made the data collected in different countries difficult to compare.

Finally, it is important to note that there is a one-year lag between conditions and the outcome: conditions are operationalised based on data from 1995 to 2014, while the outcome compares (adjusted) teaching performance in 1996 and 2015. The reason is that we believe that most changes in the policy instruments could have a “quasi-immediate” impact on behaviours, and a one-year time lag carefully takes this process into account.Footnote ¹¹

Data collection, data set construction and coding

By following the lines of our theoretical framework – which focussed on the different combinations of the existing set of adopted policy combinations – we collected, analysed and coded all pieces of national legislation and regulations regarding HE in all 12 countries under analysis from the mid-1990s onwards. Hundreds of official documents and thousands of pages of national legislation were carefully scrutinised and hand-coded in the search for both substantial and procedural policy instruments. The coding procedure proceeded in three steps: first, we identified a list of relevant pieces of legislation in national HE policy, namely, laws, decrees, circulars and ministerial regulations that affected the HES of each country under scrutiny. Second, we reduced every piece of legislation to its main issues. Third, we attributed each of those issues to one of the shapes in which we classified the policy instrument repertoire in HE.

For the first two steps, the research strategy was twofold. With respect to Italy, France and both English-speaking countries – England and Ireland – the analysis was conducted “in house”, meaning that the three authors of this article were responsible for entering the Italian, French, English and Irish pieces of legislation into the data set. Linguistic barriers rendered the selection of regulations and their direct coding impossible for the other eight countries – Austria, Denmark, Finland, Greece, the Netherlands, Norway, Portugal and Sweden. Therefore, we contacted a highly reputable country expert for each case to achieve a perfectly comparable list of pieces of relevant regulation and, in turn, legislative provisions regarding HE for those countries.

The attribution of all the analysed relevant decisions to the appropriate categories (substantial policy instruments and related shapes) was again conducted by the authors. This final step of the coding procedure was developed as follows: first, each issue of each legislative provision in each country was coded separately by each author; second, contradictory cases – that is policy instruments placed in different categories by two or more coders (approximately 15% of the whole sample) – were solved jointly in a subsequent stage.

Policy mixes in Western European HESs: an overview of the adopted policy shapes

While we refer to the Online Supplementary Material (Appendix A) for all technical details concerning our QCA, we first would like to provide a general picture of how the countries under scrutiny intervened in HE between 1995 and 2014. More precisely, Table 3 indicates how often each country in our sample recurred to all the shapes of policy instruments listed in Table 2.

Table 3 Higher education systems’ governance reforms in Western Europe (1995–2014)

With regard to regulation – the most utilised family of policy instruments – Figure 2 reveals that two out of the three most adopted instrumental shapes are concerned with giving more opportunities to universities in terms of both the content of curricula and the institutional governance (which seems quite coherent with the common template pursued by all the examined countries: the “steering at a distance” governance model). However, the most frequently utilised of all the policy instruments is related to assessment, evaluation and accreditation, which proves, from a quantitative point of view, what has been repeatedly argued qualitatively in the literature, namely, that Western European countries largely utilised evaluation tools over the course of the last 20 years (Neave Reference Neave2012; Rosa and Amaral Reference Rosa and Amaral2014).

Figure 2 Regulation: Which instruments are utilised the most (%)?

Figure 3 indicates which shapes were the most frequently used over the other three families of substantial instruments. Apparently, this general picture also confirms the trend towards a common template, but with some relevant specifications.

Figure 3 Other families of substantial policy tools (expenditure, taxation and information): Which instruments are utilised the most (%)?

In fact, expenditure was delivered not only through grants but also through targeted funding and, above all, through performance-based institutional funding. Thus, the most expected expenditure instrument in a pure “steering at a distance” governance model – lump sum – was adopted less frequently than expected. Therefore, governments preferred a more coercive way to allocate public funding (together with an emerging attitude favouring the inclusion of families in paying HE loans) and thus demonstrated the will to maintain a certain degree of control over the behaviour of universities.

Finally, regarding both taxation and information, it emerged that two instrumental shapes were particularly utilised: tax reduction for particular categories of students and transparency.

What can be associated with better teaching performance? A configurational analysis

We return to fuzzy-set QCA (fs-QCA)Footnote ¹² to empirically explore the combinations of policy instruments and related shapes that might contribute to explaining teaching performance in Western European HESs between 1996 and 2015. While we refer to the Online Supplementary Material (Appendix A) for a careful discussion and justification of the thresholds chosen in the process of “calibrating” the sets (both the conditions and the outcome) (Ragin Reference Ragin2008; Schneider and Wagemann Reference Schneider and Wagemann2012), this section presents the main results of our empirical analysis in terms of both the need for and sufficiency of relations between conditions, that is instrumental shapes, and the outcome, namely (adjusted) teaching performance.

The analysis of necessary conditions for improving teaching performance was completed quickly: no condition (or its nonoccurrence) was necessary for the outcome (or for its nonoccurrence).Footnote ¹³ In other words, we could not identify any conditions that needed to be present or absent to observe either a good teaching performance (presence of the outcome) or a bad teaching performance (absence of the outcome) between 1996 and 2015.

However, much more interesting is the analysis of the sufficient conditions for improving teaching performance in HE, which was conducted with a “truth table”. It clearly would not have been feasible to perform an analysis of sufficiency on 12 cases with dozens of different conditions (i.e. instrumental shapes) due to problems of limited diversity of the data (Schneider and Wagemann Reference Schneider and Wagemann2012; Vis Reference Vis2012); thus, we turned to theoretical reflections to hypothesise a set of possible combinations to be tested in our fs-QCA.

First, we began with 45 theoretically relevant potential configurations of conditions that populated the literature on HES performance. These combinations of instrumental shapes build mainly on the literature emphasising the adoption of similar policy instruments to follow the common template of the “steering at a distance” model. The combinations also build on the empirical evidence and contradictions emerging from variable-oriented studies that focussed on the determinants of performance in HE. Regarding the “steering at a distance” literature, we have already emphasised the substantial literature underlying the ways governments have been changing governance in HE by improving institutional autonomy (and its dimensions, such as budgetary autonomy, degree of freedom in curricular content and autonomy in recruiting academic staff), quality assurance, accreditation, teaching and research assessments, monitoring, and varieties of funding mechanisms (Gornitzka et al. Reference Gornitzka, Kogan and Amaral2005; Cheps 2006; Lazzaretti and Tavoletti Reference Lazzaretti and Tavoletti2006; Maassen and Olsen Reference Maassen and Olsen2007; Trakman Reference Trakman2008; Huisman Reference Huisman2009, Shattock Reference Shattock2014, Capano et al. Reference Capano, Regini and Turri2016).

This literature clearly addressed our choice to consider the shapes for each family of substantial policy instruments that seemed to best represent the operative dimensions of the main categories of government intervention. This choice was reinforced by the contributions that focussed on the real effects of performance-based funding on institutional autonomy and on the degree of centralisation of the governance system. This literature produced contrasting empirical evidence and thus suggested a strategic dimension that we should take into consideration. For example, with respect to performance and targeted funding as a cause of high graduation and research rates, the relevant studies showed the weak performative capacity of these instruments. Most of these studies focussed on the United States (US), where many states introduced performance criteria to determine the allocation of extra resources beginning at the end of the 1970s (Volkwein and Tandberg Reference Volkwein and Tandberg2008; Rabovsky Reference Rabovsky2012; Rutherford and Rabovsky Reference Rutherford and Rabovsky2014; Tandberg and Hillman Reference Tandberg and Hillman2014). This evidence was quite contradictory compared to the widespread use of these types of instruments by the governments of our analysed countries. Thus, we considered all targeted expenditure tools as relevant in our combinations.

Regarding institutional autonomy, contradictory evidence has emerged from reputable studies. For example, in their comparison between the EU and the US, Aghion et al. (Reference Aghion, Dewatripont, Hoxby, Mas-Colell and Sapir2010) found that high institutional autonomy (and a competitive environment) was positively correlated with high performance both in educational attainment and in research. By contrast, Braga et al. (Reference Braga, Checchi and Meschi2013) showed that high institutional autonomy negatively impacted the level of educational attainment.

With respect to the level of the centralisation of the governance system, Knott and Payne (Reference Knott and Payne2004) considered systemic centralisation in US states to be high, intermediate, or low depending on the scope of the decision-making powers held by state boards. They tested systemic centralisation as a condition affecting an array of resource and productivity measures, including the size of university revenue and the number of published articles. They concluded that state governance matters and that flagship universities are penalised by centralisation, which reduces their total revenues, research funding and number of published articles. However, centralisation was also found to reduce tuition revenues and – presumably – the cost of enrolment. The worst overall performance occurs under mild centralisation. However, this interpretation contrasts with a qualitative study of five US states conducted by Richardson and Martinez (Reference Richardson and Martinez2009). They argued that universities in centralised systems might perform better than those with decentralised designs with respect to access and graduation rates.

Thus, according to the literature, and assuming that governments have tried to pursue the “steering at a distance” model in accordance with their own national identity, we attempted to combine these shapes in such a way that 45 different policy mixes – 30 consisting of four instrumental shapes belonging to different families of instruments and 15 consisting of five of those same instrumental shapes – were established (see Table A4 in the Online Supplementary Material for those 45 different policy mixes).

Then, we performed 45 different QCAs – one for each of the selected policy mixes – to ascertain the combination of conditions that were more theoretically convincing and empirically robust. To accomplish this, we followed a two-step process. First, we focussed on the values of consistency and the coverage of different (intermediate) final solutions. In fact, in the QCA literature, it is generally argued that “coefficients of consistency and coverage provide important numeric expressions for how well the logical statement contained in the QCA solution term fits the underlying empirical evidence and how much it can explain” (Schneider and Wagemann Reference Schneider and Wagemann2010, 414). Therefore, scholars have reviewed the coefficients to measure the goodness of fit of the tested theoretical models (Ragin Reference Ragin2006). This first step allowed us to realise that two combinations existed with both consistency and coverage values that were much higher than any other, as well as higher (or, at least, equal to) than 0.90, which is a notable value.Footnote ¹⁴ However, limiting our comparison to consistency and coverage alone would be insufficient (Braumoeller Reference Braumoeller2015; Rohlfing Reference Rohlfing2018); it is important to also account for the theoretical plausibility of solutions and the (groups of) countries that are identified by each solution term. For this reason, in our second step, we carefully scrutinised those same two combinations of instrumental shapes, looking for results that appeared to be more theoretically plausible and empirically grounded. In applying these further selection criteria, we realised that one combination was preferable to the other.Footnote ¹⁵

Thus, at the end of the above-described analytical process, the best combination of instrumental shapes that emerged from the QCA for explaining the teaching performance of the analysed sample of countries between 1996 and 2015 consisted of R1 (assessment, evaluation and accreditation); R10 (institutional and administrative governance: more opportunities); R12 (rules on goals in teaching); E5 (performance-based institutional funding); and T3 (service-based student fees) (Table 4).

Table 4 Explaining teaching performance: the best policy mix (intermediate solution)

Note: Solution coverage (proportion of membership explained by all paths identified): 0.908847.

Solution consistency (“how closely a perfect subset relation is approximated”) (Ragin Reference Ragin2008, 44): 0.895641.

Raw coverage: proportion of memberships in the outcome explained by a single path.

Unique coverage: “proportion of memberships in the outcome explained solely by each individual solution term” (Ragin Reference Ragin2008, 86).

Complex solution: R12*~R1*~E5*~T3 + E5*~R10*~R12*~T3 + E5*~R1*~R12*~T3 + R10*R12*E5*T3 (coverage 0.85; consistency 0.95).

Parsimonious solution: R12 + R10*~R1 + E5*~T3 + T3*~R1 + R10*T3 + R1*~R10*~T3 (coverage 0.95; consistency 0.85).

The consistency value of the intermediate solution was impressive (0.90), and the coverage of the (intermediate) solution formula was even better (0.91). There was no “deviant case for coverage” (Schneider and Rohlfing Reference Schneider and Rohlfing2013, 585). As shown in Figure 4, six cases (i.e. Denmark, England, Greece, Ireland, the Netherlands and Sweden) were above the diagonal in the upper-right corner – and thus were “typical cases” – whereas two cases (i.e. Finland and Norway) were “deviant cases in consistency of degree”. Finally, the four cases (namely, Austria, France, Italy and Portugal) in the lower-left quadrant – being a good example of neither the solution terms nor the outcome – did not merit particular attention.

Figure 4 The “best policy mix”: final XY plot.

The above analysis of policy mixes thus means that in our sample of countries, it was possible to identify two distinct paths associated with an increase in teaching performance. First, in the Scandinavian countries – Denmark, Finland, Norway and Sweden – together with the Netherlands, it emerged that the rules on systemic goals in teaching were the prevailing policy instrument, which clearly indicates the aims to be achieved by institutions. Second, in the Anglo-Saxon countries – England and Ireland – together with Greece, improved performance was associated with performance-based institutional funding; moreover, those three countries were also similar in that they did not pay very much attention to service-based student fees.