The effects of decentralisation on educational outcomes: the portuguese municipalities’ case

The effects of decentralisation on educational outcomes: The Portuguese municipalities’ case The effects of decentralisation on educational outcomes: The Portuguese municipalities’ case Beatriz Costa Azevedo UMinho | 2023 May 2023 Beatriz Costa Azevedo Beatriz Costa Azevedo Master’s Dissertation Master in Economics Work done under the supervision of Professora Doutora Linda Veiga Professor Doutor João Cerejeira The effects of decentralisation on educational outcomes: The Portuguese municipalities’ case May 2023 Copyright and Terms of Use for Third Party Work This dissertation reports on academic work that can be used by third parties as long as the internationally accepted standards and good practices are respected concerning copyright and related rights. This work can thereafter be used under the terms established in the license below. Readers needing authorization conditions not provided for in the indicated licensing should contact the author through the RepositóriUM of the University of Minho. License granted to users of this work: Attribution-NonCommercial-NoDerivatives CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/ i Acknowledgements The development of this dissertation relied on the contribution of several people, to whom I express my sincere appreciation. First of all, I would like to express my gratitude to my supervisor, Professor Linda Veiga, for all the support, availability, sharing of knowledge and experiences, the opportunities offered, the helpful meetings, and all the feedback provided during the completion of this dissertation. Thank you for helping me to improve my skills and to discover my interest in economics research. To Professor João Cerejeira, co-supervisor of this dissertation, I gratefully acknowledge his availability, help and, above all, the transmission of econometric knowledge indispensable for this research. To Professor Francisco Veiga, I appreciate the time devoted to this work and the suggestions provided. To the Directorate General of Local Authorities (DGAL), I acknowledge the availability of the solicited data and the readiness to send them. To Statistics Portugal (INE), I appreciate the prompt response to the clarifications requested. I would also like to thank all those who, in one way or another, crossed my academic path or personal life and left their mark, including all my teachers. A thank you to all my friends and particularly to Joana for the mutual support during the development of our dissertations. A special and grateful acknowledgement to Rui for appearing in my life, accompanying me during these months, and for all the support, love and understanding. Finally, although undoubtedly not less important, I sincerely appreciate my parents for all the dedication, security and love they have given me over the years. I am very grateful to you for supporting me unconditionally in all my life journeys and giving me the opportunity to pursue my studies. Without you, nothing would make sense. ii Statement of Integrity I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. University of Minho, Braga, may 2023 Beatriz Costa Azevedo iii Os efeitos da descentralização nos resultados educativos: O caso dos municípios Portugueses Resumo Nas últimas décadas, Portugal experienciou uma progressiva descentralização das funções do governo central, sendo a educação um dos setores mais descentralizados. Estudos anteriores concluíram que esta crescente atribuição de poderes apresenta vários impactos, por vezes contraditórios, apesar de não ter sido realizada nenhuma análise semelhante para o caso de Portugal. Assim, o principal objetivo desta dissertação prende-se com a avaliação dos impactos da descentralização nos resultados educacionais dos municípios Portugueses, focando-se nos contratos de execução assinados em 2009 e em 2015. Este trabalho assenta na construção de uma base de dados nova, com informação sobre os 278 municipíos do Continente para o período entre 2004 e 2019. A estimação de um modelo base e de um modelo flexível, usando o método das diferenças-em-diferenças, demonstrou que os efeitos destas novas responsabilidades na qualidade e no acesso à educação foram pouco expressivos, não existindo variações significativas ao longo dos anos. Contudo, a extensão para um enquadramento com múltiplos períodos de tratamento permitiu a diferenciação dos municípios descentralizados de acordo com o primeiro ano em que seria esperado experienciarem efeitos. Os resultados obtidos após esta consideração demonstraram que a descentralização melhorou as taxas de retenção e de escolarização, sobretudo ao nível do ensino básico. Vários testes foram aplicados para garantir a robustez dos resultados. A abordagem empírica escolhida e as particularidades do processo de descentralização em Portugal podem ajudar a explicar os resultados obtidos. Palavras-chave Contratos; Descentralização; Educação; Municípios Portugueses. iv The effects of decentralisation on educational outcomes: The Portuguese municipalities’ case Abstract Over the last decades, Portugal has experienced a progressive decentralisation of central government functions, education being one of the most decentralised sectors. Previous studies have found that this increased attribution of powers presents various and, sometimes, contradictory impacts, even though no similar analysis was performed for Portugal. Therefore, the main goal of this dissertation is to assess the impacts of decentralisation on the educational outcomes of Portuguese municipalities, focusing on the execution contracts signed in 2009 and 2015. This research relies on a newly built database, encompassing information on the 278 mainland municipalities from 2004 to 2019. Estimating baseline and flexible models using a difference-in-differences approach indicates that the new responsibilities promoted little changes in education access and quality, not existing significant variations throughout the years. Nonetheless, the extension to a multiple time periods framework allowed the differentiation of decentralised municipalities according to the expected starting year of effects. The results obtained after this consideration suggest that decentralisation improved retention and schooling rates, especially at the basic education level. Several tests were applied to prove the robustness of the results. The empirical methodology followed and the particularities of the decentralisation process in Portugal may help explain the results obtained. Keywords Contracts; Decentralisation; Education; Portuguese Municipalities. v Contents 1 Introduction...................................... 1 2 LiteratureReview................................... 2 2.1 The concept of decentralisation . . . . . . . . . . . . . . . . . . . . . . . . 2 2.2 The impacts of decentralisation on education . . . . . . . . . . . . . . . . . 3 3 PortugueseCase ................................... 8 3.1 Portuguese educational system . . . . . . . . . . . . . . . . . . . . . . . . 8 3.2 Decentralisation of the Portuguese educational system . . . . . . . . . . . . 9 4 HypothesesandDataSources............................. 12 4.1 Hypotheses ................................. 12 4.2 Datasources................................. 13 5 EconometricModels ................................. 18 5.1 Empiricalframework............................. 18 5.2 Baselinemodel................................ 24 5.3 Flexiblemodel................................ 26 5.4 DD with multiple time periods . . . . . . . . . . . . . . . . . . . . . . . . 26 6 EmpiricalResults................................... 29 6.1 Baselinemodel................................ 29 6.2 Flexiblemodel................................ 33 6.3 Multiple time periods approach . . . . . . . . . . . . . . . . . . . . . . . . 41 6.4 Robustnesstests............................... 46 7 DiscussionandConclusion .............................. 47 Bibliography ........................................ 51 AppendixA......................................... 52 AppendixB......................................... 64 AppendixC......................................... 66 vi and their intensity experienced by the different groups of municipalities over time. Nonetheless, the lack of significance for some indicators and the negative impacts found in specific variables might reflect the decentralisation process structure and the small capacity of the transferred functions to impact students’ outcomes. This dissertation provides two different types of contributions. First, it sheds light on the impacts that two successive reforms had on education in Portugal and on the adequacy of the set of competencies transferred, an analysis that had ever been done before. Second, it demonstrates that the precise definition of the setup and the extension of typically-used approaches to frameworks taking into account the specificities of decentralisation, such as the different starting years of effects, might provide different and significant results. The remainder of the dissertation is organised as follows: section 2presents a review of the previous literature on the topic; section 3describes the institutional background, mainly regarding the organisation of the Portuguese educational system and the details on the several decentralisation reforms of the last years; section 4starts with the definition of the hypotheses under study, followed by the description of the data used, its sources and computation processes; section 5explains the empirical methodologies to be used, with the derivation of the adequate models; section 6depicts the results obtained; section 7ends with possible explanations for it and the main conclusions. 2 Literature Review 2.1 The concept of decentralisation In recent decades, the world has witnessed increased pressure to decentralise government activities. Although decentralisation was first observed in industrialised countries (Oates, 1999), it soon spread across the developing world (E. Ahmad & Brosio, 2006). Nevertheless, the decentralised governance systems and the motivations behind the transition are considerably diverse between countries, which further challenges its analysis (J. Ahmad et al., 2006; OECD, 2019; Veiga et al., 2015). Even though it has been deeply studied over time, the scope of decentralisation may vary significantly and several definitions exist. The broader concept of decentralisation encompasses several interdependent dimensions, including fiscal, administrative and political ones (J. Ahmad et al., 2006; OECD, 2019; Veiga et al., 2015). Given its complexity, measuring the decentralisation level is not an easy task since there is no single indicator capable of embracing all its dimensions (Martinez-Vazquez et al., 2017), and relying 2 exclusively on fiscal indicators1might provide a distorted interpretation of reality (OECD, 2019). There are several arguments in favour and against decentralisation2. In a decentralised context, it has been argued that mobile consumers have the opportunity to move to the communities that best satisfy their preferences (Tiebout, 1956). Other arguments commonly used in favour of decentralisation include more efficient provision of public goods and services due to the higher proximity of subnational governments to local populations, enhanced regional productivity due to the increased competition between local governments, and higher political participation (Oates, 1999). Nonetheless, those arguing against decentralisation have pointed out negative impacts, such as non–internalisation of spillovers in the presence of externalities (Oates, 1999), loss of economies of scale, growth of inequalities, and additional costs due to the creation of new administrations and local elections (Veiga et al., 2015). According to J. Ahmad et al. (2006), the attempt to improve the delivery of essential services, including education and health, is one of the main motivations behind the most decentralisation processes worldwide, given the failures of central governments in providing those services and the mismatch between local preferences and centralised decisions. On the other hand, social protection is often the least decentralised function due to the belief that central governments deal more efficiently with redistribution than local governments (Dafflon, 2006; Oates, 1999; Veiga et al., 2015). Despite the variation in degree of decentralisation across countries, education is one of the most commonly identified areas in which subnational governments are considered essential and has merited significant attention from previous literature. 2.2 The impacts of decentralisation on education In the education framework, a conflict between desirable outcomes has been observed, with several societies arguing over the need to decentralise public schools without neglecting the assurance of minimum quality standards at a national level (Dafflon, 2006). In the past years, several studies have analysed the impacts of decentralisation on education, with the majority finding evidence of positive effects (Veiga et al., 2015). Nonetheless, the degree of decentralisation differs across those analyses, ranging from the attribution of powers in terms of education finance and expenditures (Barankay & Lockwood, 2007; Faguet & 1Commonly used fiscal indicators include tax and spending autonomy, revenue and spending shares of local government, transfer dependency, tax and revenue decentralisation ratios, subnational allocation of resources, among others (Borrett et al., 2021; International Monetary Fund, 2020; Lledó et al., 2020; OECD, 2020). 2For a complete review of positive and negative impacts associated with greater degrees of decentralisation, particularly concerning its fiscal dimension, see E. Ahmad and Brosio (2006) and Veiga et al. (2015). 3 Sánchez, 2008; Kyriacou & Roca-Sagalés, 2019) to a higher degree of autonomy, such as the decentralisation of school administration (Elacqua et al., 2021; Galiani et al., 2008; Hanushek et al., 2013; Salinas & Solé-Ollé, 2018), which may help explain the eventual differences in the findings (Guerra & Lastra-Anadón, 2019). Some studies focused on specific countries while others performed cross–country analyses3. Concerning education access, there is evidence that decentralisation improves public schools’ enrolment rates (Elacqua et al., 2021; Faguet & Sánchez, 2008; Guerra & Lastra-Anadón, 2019) and reduces the rates of early school dropouts (Salinas & Solé-Ollé, 2018). Regarding education quality, there is evidence of the positive impact that decentralisation has, in general terms, on students’ performance. Notably, of an improvement in the classifications obtained in national exams (Elacqua et al., 2021; Galiani et al., 2008) and the percentage of students attaining the university-required entrance levels (Barankay & Lockwood, 2007). International comparisons have shown improved PISA test scores, although this result held only for developed countries (Hanushek et al., 2013). Additionally, Elacqua et al. (2021) have found that decentralisation improved teachers’ quality and that hiring high-quality teachers could partially explain the better student outcomes in decentralised municipalities. However, decentralisation may also harm education quality due to the congestion prompted by the positive effect on education access (Guerra & Lastra-Anadón, 2019). Some studies have highlighted differences in results within the same analysis: the impacts of decentralisation benefit more males (Barankay & Lockwood, 2007) and non-poor students (Galiani et al., 2008). The latter may result from the higher ability of non-poor families to move to areas with better education quality and the lower ability of poor individuals to hold politicians accountable for their resource allocation decisions (J. Ahmad et al., 2006). Moreover, several authors have identified a particular connection between the magnitude of the effects and the attributes of subnational governments: decentralisation has more substantial impacts in communities with greater levels of local revenues (Salinas & Solé-Ollé, 2018), as well as in those that are more assertive and prioritise costly and visible policies (Guerra & Lastra-Anadón, 2019). In addition, decentralisation also impacts the governance of local authorities directly. There is evidence of incentives for service delivery improvement when local governments cannot depend only on central transfers and need to raise their own revenues (J. Ahmad et al., 2006). Furthermore, local governments appear to become more responsive to local needs (Faguet & Sánchez, 2008) and more effective as the perceived quality of public services increases (Kyriacou & Roca-Sagalés, 2019). As argued by Elacqua 3Among others, see Barankay and Lockwood (2007), Faguet and Sánchez (2008), and Salinas and Solé-Ollé (2018) for country studies and Guerra and Lastra-Anadón (2019) and Hanushek et al. (2013) for cross–country analyses. 4 et al. (2021), the impacts of decentralisation on education seem to be more closely related to the better allocation of resources than to its amount. Another fundamental aspect of decentralisation, particularly in the scope of education, concerns the timing of its impacts: the positive impacts may not be observed in the short run due to the need for a period of policy consolidation (Elacqua et al., 2021). Given the contradictory findings described above, identifying the desired levels of decentralisation on educational services takes time and effort. Table 1 presents a summary of the previous literature’s main findings concerning the impacts of decentralisation on education outcomes. Table 1: Main findings of previous literature Reference Dependent Variable Sample Methodology Main Conclusions Barankay and Lockwood, 2007 Share of 19-year-old students obtaining university entry qualification 26 Swiss cantons (1982–2000) Fixed effects with clustered standard errors Decentralisation associated with greater educational attainment Faguet and Sánchez, 2008 Public investment by sector (Bolivia) and annual change in public schools enrolment rates (Colombia) Colombian (1994-2004) and Bolivian (1987-1993) municipalities Tobit estimations and principal component analysis (Bolivia); Two-Stage Least Squares panel estimations (Colombia) Decentralisation of education finance increased enrolment rates (Colombia) and government responsiveness to local needs (Bolivia) 5 Table 1: Main findings of previous literature, continued Reference Dependent Variable Sample Methodology Main Conclusions Galiani et al., 2008 Test scores (school and province level) Argentine public schools (1994-1999) DD and Generalized Least Squares methods Decentralisation positively impacted students’ results, with these gains not benefiting the poor Hanushek et al., 2013 Students’ achievement (PISA test scores) 42 countries (2000-2009) Panel estimation with country-fixed effects More autonomy negatively impacted students’ achievement in developing and developed countries Salinas and Solé-Ollé, 2018 Dropout rates in secondary education 17 Spanish regions (1977-1991) DD method and event-study analysis Decentralisation significantly impacted the early school dropouts rate; Stronger results in regions with more revenues 6 Table 1: Main findings of previous literature, continued Reference Dependent Variable Sample Methodology Main Conclusions Guerra and Lastra-Anadón, 2019 PISA test scores and enrolment rates (OECD); Graduation and enrolment rates (Spain) OECD countries (2000-2012) and Spanish regions (1980-1999) DD and Synthetic Controls methods Decentralisation positively impacted education access but affected its quality negatively (OECD and Spain); Stronger effects in assertive regions (Spain) Kyriacou and Roca-Sagalés, 2019 Government effectiveness (quality of public services) 30 European countries (1996–2015) Ordinary Least Squares with panel corrected and robust standard errors Decentralisation of education expenditures increased the perceived quality of this public service Elacqua et al., 2021 Test scores, school enrolment and teachers’ quality Colombia (1996-2015) DD and regression discontinuity methodologies Decentralisation improved student achievement, school enrolment and teachers’ quality 7 3 Portuguese Case 3.1 Portuguese educational system According to Eurydice (2022), the Portuguese educational system encompasses distinct levels, including pre-primary education, which is optional for all children aged three to six years4. Basic education is mandatory and lasts nine years, divided into three different cycles. The first cycle lasts four years for students aged six to nine years old, the second cycle lasts two years for students aged ten to twelve years old, and the third cycle lasts three years for students aged twelve to fourteen years old. Secondary education is also compulsory, lasts three years for students aged fifteen to eighteen years old, and is divided into five separate courses (Science-humanities courses5; Vocational courses; Specialised artistic courses; Own-school-curriculum courses; Apprenticeship courses). Generally, Portuguese students finish mandatory education at the age of eighteen6. National exams are taken by students in the final year of the third cycle of basic education and in the two last years of secondary education7. All students take Portuguese and Mathematics exams at the end of basic education8. In contrast, the final exams of secondary education cover several areas, depending on the specific course in which the student is enroled9. Several substantial changes in the structure of the national exams regarding secondary education were introduced during the production of this research. However, since the focus of this analysis is the 2004-2019 period, and the changes will only produce effects from the 2023/2024 academic year onwards, they were not considered when choosing the variables used10. It is essential to mention that there may be remarkable dissimilarities in the educational sector of 4Despite being optional, Law n.º 65/2015, July 3rd established the universality of pre-primary education for all children over four years old. For children under three years old, education is focused on childcare, not considered a level of the Portuguese education system (Eurydice, 2022). 5The Science-humanities courses are subdivided into the following courses: Science & Technology, Socio-economic Science, Languages & Humanities and Visual Arts (Eurydice, 2022). Since most Portuguese students choose one of these science-humanities courses, they were the focus of this analysis. 6Although compulsory education used to correspond to only nine years, Law n.º 85/2009, August 27th, established the new regime of mandatory education, which now corresponds to 12 years. 7National exams for Portuguese and Mathematics used to be carried out in the last year of the first cycle of basic education, but after several setbacks in their introduction, these exams ceased in 2015, and there is a substantial lack of data concerning their results. 8Non–native students can take the Portuguese Non–Native Language and the Portuguese Second Language exams, but these are relatively uncommon compared to standard tests and, therefore, will not be considered in the proposed analysis. 9In the last two years of secondary education, students enroled in Science-humanities courses must take the Portuguese exam and three other exams, depending on their courses (one of them is performed in the same year as the Portuguese exam, and the other two are carried in the previous year) (IAVE I.P., 2022). 10 In February 2023, the Minister for Education, João Costa, and the Minister for Science, Technology and Higher Education, Elvira Fortunato, announced that, starting in the academic year 2023/2024, students will need to take three national exams to complete secondary education, one of which must be the Portuguese exam, while students can choose the other two according to the specific requirements for accessing higher education (MCTES and ME, 2023). 8 the Portuguese Autonomous Regions of Madeira and Azores. Even though local governments and their functions may be similar to those in the mainland, Madeira and Azores have regional governments with considerable autonomy in decision-making. Therefore, the proposed analysis will focus on the 278 mainland municipalities. 3.2 Decentralisation of the Portuguese educational system The decentralisation of the Portuguese educational system began in 1984, with the transfer of competencies related to school transport and social action in pre-primary and basic education. These transferences were regulated by the Decree-Law n.º 299/1984, September 5th and the Decree-Law n.º 399-A/1984, December 28th, respectively. Over the years, several legal regulations progressively enlarged the responsibilities of local governments to implement the principles of local autonomy and administrative decentralisation. In 1999, Law n.º 159/99, September 14th transferred a significant set of competencies to local governments, including responsibility for maintaining school buildings, providing school transport, and organising complementary activities, among others. These competencies were mainly related to pre-primary and basic education. Additional responsibilities were transferred to local governments in 2003 with the publication of Decree-Law n.º 7/2003, January 15th and Law n.º 41/2003, August 22th. Those diplomas regulated the transfer of the new competencies and the functioning of municipal councils, created in the scope of education, and approved the educational letter. Later, the specific regime of the non-teaching staff and the legal regime concerning the collective transport of students from and to schools were defined by the Decree-Law n.º 184/2004, July 29th and by the Law n.º 13/2006, April 17th. In the following years, some Portuguese municipalities received additional competencies in education through contracts signed with the central government. These agreements occurred around 2009 and 2015 and implemented the decentralisation foreseen in the previous legislation. Since those contracts are the basis of this empirical analysis, their details are provided in the following subsections. More recently, Law n.º 50/2018, August 16th, promoted a new decentralisation reform in Portugal, covering various domains. A set of unique competencies was transferred to municipal entities, focusing on the second and third cycles of basic education, and on secondary education. These new responsibilities include, among others, managing school canteen meals, elaborating the education letter and the plan of school transport, developing school social action, and maintaining and preserving the pre-primary, basic and secondary education buildings. Given the complexity of the decentralisation process in education and the unforeseen COVID–19 crisis, 9 Portuguese municipalities were able to postpone the transfer of new competencies from January 1st, 2021, to March 31st, 2022, as defined in the Decree-Law n.º 56/2020, August 20th. By the end of 2021, only 161 out of the 278 Portuguese mainland municipalities had implemented their new education competencies (DGAL, 2022b)11. Nonetheless, all the 278 mainland municipalities were exercising competencies in the field of education by July 2022 (DGAL, 2022a). The contracts of 2009 In 2008 and 2009, the Ministry of Education (ME) and 113 municipalities signed execution contracts12. These contracts were of voluntary signature for municipalities and executed the transference of competencies in the scope of education, following Law n.º 2/2007, January 15th, which approved the new regime of local finances, and Decree-Law n.º 144/2008, July 28th, which determined the process of transferring competencies to local municipalities. As stipulated in the contracts, the new responsibilities were related to the non–teaching staff in preprimary and basic education schools, the curricular enrichment activities in the first cycle of basic education, and the school estate management relative to the second and third cycles of basic education13. The agreements also defined the monetary amounts to be transferred to municipalities to cover the additional costs, as well as the start date of those transfers. These values were specific to each municipality, depending on its characteristics, such as the number of non-teaching staff that would be transferred and the number of students enroled in schools targeted for decentralisation. For the majority of the contracts signed, the competencies and the respective amounts started to be transferred in January 2009, while for other municipalities, the assumption of the new responsibilities started afterwards, at the latest in January 201014. Nonetheless, as soon as the municipalities started receiving the new competencies, there were several 11 The exercise of additional educational competencies by municipalities was carried out through three different channels: Programa Aproximar Educação (14 municipalities), Contracts of Execution (51 municipalities) and Decree-Law n.º 21/2019, January 30th (96 municipalities) (DGAL, 2022b). 12 113 contracts were signed between the ME and Portuguese mainland municipalities at the end of 2008 or during 2009 to define the transfer conditions. All of them were published in Diário da República (DR) in 2009 (INCM, 2022). One example of those contracts is available in Appendix A, which also contains information on the municipalities that signed the contracts, including the dates on which the signature took place, the competencies were transferred, and the effects started to be experienced. 13 For some municipalities, the execution contracts also defined two additional responsibilities, concerning the management of secondary schools that also encompassed the third cycle of basic education and students’ residencies. Since the number of municipalities receiving these functions was significantly small compared to the 113 that signed the contracts, no particular attention was given to these cases. 14 The transfer of new competencies took place in January 2009 for 90 municipalities, while the others received the additional responsibilities in March 2009 (5 municipalities), May 2009 (1 municipalities), October 2009 (6 municipalities) or in January 2010 (10 municipalities). There was also a case of a municipality which started to assume functions right in October 2008 ( Freixo de Espada à Cinta ). 10 complaints about the insufficiency of the funds transferred to face the new responsibilities. At the beginning of 2010, the National Association of Portuguese Municipalities (ANMP) surveyed the municipalities participating in this reform and suggested that the transfer conditions should be more precisely determined. Together with the ME, it was decided not to sign execution contracts with other municipalities until there was a meticulous evaluation of the situation resulting from the already signed contracts15 (ANMP, 2010). The contracts of 2015 In 2015, the Ministry of Education and Science, the Presidency of the Council of Ministers, and fifteen Portuguese municipalities signed inter-administrative contracts to delegate new competencies16. The signature of those contracts followed the Law n.º 75/2013, September 12th, which established the legal regime of local governments and defined the transference of competencies from the central administration through inter–administrative contracts, and the Decree-Law n.º 30/2015, February 12th, which established the regime under which this transfer would take place. These contracts were part of a pilot project named Programa Aproximar Educação (PAE) , which aimed to promote the efficiency of educational resources and to contribute to human and community development by covering areas such as educational policies and administration, curriculum development, pedagogical and administrative organisation, resource management and school–community relation. These areas are described at the beginning of each contract (INCM, 2022). Unlike the agreements signed in 2009/2010, these contracts did not target a specific level of education. Due to the importance of decentralisation in Portugal, this pilot project encompassed a limited number of municipalities to promote a gradual approach (Secretário de Estado da Administração Local, 2014). Even though each of them had to consent to the contract signature, the choice of the group to integrate the programme was the central government’s responsibility. The primary objective of this selection process was to achieve a group of municipalities with a notable degree of territorial, political and sociodemographic variability. It took into account not only the strong will of the mayors but also the high commitment demonstrated by municipalities in the past, both regarding the educational mission and the management of public resources (Secretário de Estado da Administração Local, 2014). 15 After checking all the contracts published in DR between 2008 and 2015, it was possible to find that three additional municipalities - Vimioso , Entroncamento and Vidigueira - signed similar contracts in 2011 and 2012. These three municipalities were excluded from the analysis. 16 In total, fifteen municipalities signed these contracts, which were then published in DR. The contracts signed in 2015 and retrieved from INCM (2022) are in the format of protected PDF, not being possible, thus, to include an example in the Appendix. An example of those contracts may be observed in https://dre.pt/dre/detalhe/contrato/552-2015-69879439. 11 an estimate of the annual values for the period of analysis. The number of residents and the average monthly earnings were included in logarithmic terms to facilitate the interpretation of results. All financial variables, such as local revenues and average monthly earnings, were considered in real terms at 2022 prices to ensure the correct comparison over the years. 5 Econometric Models 5.1 Empirical framework As previously mentioned, the central hypothesis posits that the decentralisation of powers in education has improved education-related indicators, such as enrolment rates and national exam grades. A DD approach was used to estimate these impacts, incorporating municipal and year-fixed effects for the 278 mainland municipalities between 2004 and 2019. Nevertheless, for the particular case of the first reform, the estimations considered only the period between 2004 and 2015 to avoid the very likely overlap of effects with the second decentralisation moment36. The use of DD framework ensures that the time-unvarying characteristics of municipalities, which could be related to their choice and educational outcomes, as well as the time trends, do not confound the obtained results (Guerra & Lastra-Anadón, 2019; Salinas & Solé-Ollé, 2018). Nevertheless, the proposed analysis may not be straightforward since other factors could influence the educational outcomes and the decision to participate in this type of contract. One commonly used argument relies on the idea that the municipalities which agree to receive more powers already have significantly deeper concerns about education. Therefore, self-selection problems may affect the empirical analysis. Even though the Portuguese case did not benefit from an arbitrary choice over the municipalities facing decentralisation as in Elacqua et al. (2021), each municipality did not entirely determine that choice. Several particularities in the celebration of these contracts enable the comparison of outcomes between municipalities that signed them and those that did not: • Decentralisation occurred in two different moments and was uncommon for all municipalities; • The transfer of responsibilities was part of a broader package encompassing other functions, indicating that educational sector-specific features did not determine this process; 36 Regarding the second reform, avoiding the simultaneity of effects is not so direct due to the temporal proximity to the first moment. Therefore, the empirical analysis relied on additional tests to prove the robustness of results, as described in the following section. 18 • In the case of the 2009 reform, the signature of contracts was not mandatory. Moreover, even though the intention was for all municipalities to face the same decentralisation, it did not happen due to complaints about insufficient funds (ANMP, 2010). Therefore, self-selection for this reform may not have occurred; • For the 2015 reform, although each municipality had to agree to sign the contract, it was up to the central government to determine which group should integrate the decentralisation programme. So, while it is reasonable to expect that those who agreed had more significant concerns regarding education, it was not their intention that led to the decentralisation; • There is a substantial variety of features among the municipalities that signed the contracts, namely in terms of location, dimension and other demographic indicators, suggesting that the similar municipality–specific effects were not the determining factor of selection37. All these particularities around the signature of contracts provide the necessary groundwork for applying the DD framework. Nonetheless, the parallel trends assumption must hold, meaning that the trends of decentralised and non-decentralised municipalities should be similar before the signature of contracts. Although there is no particular test to confirm the validity of the parallel trends assumption, a visual inspection can shed light on the behaviour of each outcome trend38.Figure 1 displays the paths followed by the averages of the educational outcomes in municipalities that signed the contracts and those that did not for the period before 2010. As observed, there seems to be an identical path for retention and schooling rates, independently of the study cycle. However, the percentage of students enroled in public schools appears to behave differently in some years of the pre-intervention period. Figure 2 depicts similar information for the second decentralisation reform for the period before 2016. Once again, the trends of decentralised municipalities are identical to those of non-decentralised, particularly in the case of transition/retention rates and average classifications in national exams. Nevertheless, some differences arise in the public-private schools’ student ratio, the pre-schooling rate, and some years of the other educationrelated rates. Additional tests are typically used to assess the validity of the parallel trends assumption, known as ”placebo tests”. These tests were applied in this specific case and evaluated the statistical significance 37 The working memorandum of the 2015 programme clearly stated the goal of achieving a group of municipalities with significant demographic, political and territorial diversity (Secretário de Estado da Administração Local, 2014). Moreover, the descriptive statistics presented in Table 17 and Table 18 of Appendix B demonstrate this considerable variety of characteristics within each group of municipalities. 38 Given the third hypothesis about the impacts of decentralisation on expenses and compensations received, similar graphs for those variables can be found in Appendix C. 19 of the decentralisation dummy variable for the periods before the reforms took place. If no significant differences exist between the municipalities that signed the contracts and those that did not, their behaviour should be identical in the pre-intervention period. Therefore, the dummy variables should not be statistically significant before 2010 or 2016. The discussion of those results is presented in the following section. Despite the checks presented above, it could still be argued that, for some variables, there might be a violation of the parallel trends assumption, which could question the validity of the estimation results. To be extra cautious and account for this possibility, the models also controlled for regional-specific trends39. Figure 1: Trends in the educational outcomes before the 1st reform 39 The inclusion of municipal and time fixed-effects in the models implies the need for a certain degree of within-variation. Therefore, the trends were included at the regional level instead of considering 278 municipal-specific trends to avoid having many variables and the resulting lack of variation. 20 Figure 1: Trends in the educational outcomes before the 1st reform (cont.) 21 Figure 2: Trends in the educational outcomes before the 2nd reform 22 Figure 2: Trends in the educational outcomes before the 2nd reform (cont.) 23 5.2 Baseline model After addressing the specificities of DD framework, particularly the assumption of common parallel trends, the baseline model for assessing decentralisation effects was derived. The model is as follows: Yit =α+β1Decentralised_Ait +β2Xit−1+µi+λt+θit+εit.(1) i= 1, ..., 27840 t= 2004, ..., 201941 where Yit corresponds to a given outcome variable in municipality iin year t. For the 2009’s reform, the dependent variables are the percentage of students enroled in public schools, as well as the schooling and retention rates. When focusing on the 2015 contracts, the average classification in national exams42 and the transition/completion rate of secondary education are also included in the group of dependent variables. Decentralised_Ait is a dummy variable that equals one for the municipalities that signed the contracts from the year the competencies were assumed until 2019 and zero otherwise. Since the first reform’s effects started between January 2009 and January 2010, this dummy equalled one from 2010 onwards for the decentralised municipalities. This decision ensures that the dummy variable encloses all contracts and that the time required to carry the new responsibilities entirely is considered43. The additional competencies appointed in the second reform were transferred slightly before the 2015/2016 school year began. Accordingly, the decentralisation dummy variable equalled one from 2016 onwards for those municipalities that signed the contracts44.β1is the coefficient of interest, representing the effect of decentralisation after the signature of the contract. 40 For the particular case of 2009, the number of municipalities corresponded to 275 due to excluding the three municipalities that only signed the contracts in 2011 and 2012. The sample included then the municipalities of Vimioso , Vidigueira and Entroncamento for a robustness check. 41 For the first reform, the analysis covers only the period between 2004 and 2015 to avoid overlapping effects, while it encompasses the entire period in the case of the second reform. 42 For space-saving purposes, the results concern only the average by study cycle, that is, the average classification of all the exams carried out in the third cycle of basic education or during secondary education. Nonetheless, the results for the average of each exam are, in general terms, according to the results of the cycles’ averages. 43 The academic year starts in September, so the municipalities decentralised in January 2009 received the new competencies in the middle of the academic period. Hence, it is probable that they have only experienced effects in the 2009/2010 school year, as it occurred with those assuming the new responsibilities later. Given the conversion of academic periods into civil years described in section 4, 2010 should be considered in the analysis. As formerly mentioned, the three municipalities that underwent effects only after 2011 were excluded from the study. 44 This consideration was also based on the conversion of school periods into civil years, explained in section 4. 24 Xit is a vector of control variables that may also impact the educational outcomes and is included in lagged terms45. That vector includes the following variables, which may control for other municipal-specific features46: • Population: Represents the number of people living in a given municipality in a specific year. This variable is included to control for the size of the municipalities and is introduced in logarithmic terms; • Average monthly earnings and unemployment rate: These variables control for the economic background of each municipality. The average monthly earnings were adjusted to real terms (at 2022 prices) and were also included in logarithmic terms; • Percentage of residents with higher education: This variable serves as a proxy for the educational attainment of the population in each municipality. Finally, µiand λtrepresent the municipal and year fixed-effects, respectively, while θitare the regionalspecific time trends. Considering the previously mentioned hypothesis about the behaviour of educational expenses and compensations received after decentralisation, a reduced form of the same baseline model was estimated. The dependent variables of this new version encompassed the educational expenses and the compensations received per education cycle47. These values were introduced in real terms (at 2022 prices) and represented the amount per student. Moreover, this reduced form did not include the vector of control variables since demographic and municipal-specific features are not expected to influence the monetary amounts spent and received, nor the number of students, given that the computation of the values already considered them. In this model, the years represented by tstarted only in 2007, the first year for which this financial data is available. 45 The lagged terms are considered because it is highly probable that the conditions involving students and their parents’ realities may take some time to impact educational outcomes and the private-public school choice. One specific example is the proxy for the unemployment rate, which corresponds to the rate registered in December. Therefore, one might expect that the unemployment proxied by this specific rate will only likely affect outcomes and choices in the following year. 46 For testing purposes, this vector of control variables also included other indicators such as the crime rate (collected from INE), own revenues per capita and its percentage in total revenues (both retrieved from DGAL website), the expenses in education and compensations received by the cycle of education (directly provided by DGAL), as well as the number of students per teacher and per computer with an internet connection (both from INE). Nonetheless, these variables did not turn out statistically significant in the estimations in which they were included, and there is no evidence that they impact educational outcomes. 47 Data was available for the pre-primary and the first cycle of basic education, as well as for total values. 25 5.3 Flexible model Apart from estimating the impact of decentralisation in general terms for the period after the reform, it is also interesting to understand if those same effects were constant or modified over the years. With that goal in mind, this empirical research also employed a flexible model. Relying on the same assumptions as the baseline DD framework, this flexible model allowed testing the hypothesis of parallel trends and enabled the examination of whether the reform had different effects over the years48. The derived model was the following: Yit =α+ 2019 ∑ t=2004 β1tDecentralised_Ait +β2Xit−1+µi+λt+θit+εit.(2) i= 1, ..., 27849 t= 2004, ..., 201950 As previously noted, the coefficient of the decentralisation dummy variables, β1t, should not be, on the one hand, statistically significant for all the pre-treatment years to prove that the parallel trends assumption is verified. On the other hand, β1tshould be statistically significant after 2010 or 2016 if decentralisation did impact educational outcomes. Differences in the coefficients’ values and significance for the years after decentralisation indicate that the effects and their intensity may have varied over time. The remaining components of the flexible model were defined as in the baseline DD model. 5.4 DD with multiple time periods The general DD approach considers a setup with two different periods and two groups, which must display a similar trend in the pre-intervention period. However, that is often not the case, with several analyses focusing on a multiple-period framework with significantly different groups that might prevent the common trends assumption from holding. The use of the DD extended for multiple time periods as presented by Callaway and Sant’Anna (2021) allows the estimation of those same impacts for cases in which there are more than two periods, the units receive treatment at a different time, and the common trends assumption does not hold unconditionally. 48 This approach followed the similar one used by Elacqua et al. (2021) in the study of the decentralisation reform which took place in 2002, involving some Colombian municipalities. 49 For the particular case of 2009, the number of municipalities corresponded to 275 due to excluding the three municipalities that only signed the contracts in 2011 and 2012. The sample included then the municipalities of Vimioso , Vidigueira and Entroncamento for a robustness check. 50 For the first reform, the analysis covers only the period between 2004 and 2015 to avoid overlapping effects, while it encompasses the entire period in the case of the second reform. 26 When looking at the contracts of the first decentralisation reform in analysis, two groups of municipalities may be distinguished: those which signed contracts in September 2008 and those which only signed in the middle of 2009. Since the transference of competencies might take time to produce effects on educational outcomes and, in the majority of the cases, the contracts clearly stated that the transference of some functions and funds would only happen on the first day of the following year, the two groups may be seen as potentially experiencing effects in different times. There are several assumptions on which this method relies. Therefore, before adopting this new approach, it was essential to understand them and guarantee they were adaptable to this case. As presented by Callaway and Sant’Anna (2021), the assumptions are the following: • After receiving treatment, which must not happen in the first period, each unit continues to be treated in the following periods. That is precisely the case of the presented analysis since those municipalities signing contracts in 2008/2009 remained treated afterwards, and there is available data for the pre-treatment period; • A panel data should be used51 and an anticipation behaviour towards treatment is generally not allowed, even though this might happen in those cases where its horizon is clear. This research relies exclusively on a panel dataset and, although one could argue that municipalities may know about the contracts before signing them, that knowledge was acquired, at most, in the previous year, not before52; • The parallel trends assumption should also hold in this framework, even though it might be conditional on covariates represented by X. Such premise is fundamental in cases where those covariates are differently distributed across groups and may present particular outcome trends over time. As previously stated, there might be doubts about verifying the unconditional parallel trends assumption in this case since there might be considerable differences between those municipalities that signed and those that did not, particularly in terms of specific features represented by the control variables in X. Therefore, given that all conditions seem to hold in the specific case of this research, the DD extended to a multiple time periods framework was also used to assess decentralisation impacts concerning the 51 Nonetheless, the authors also show that results hold for the cases with repeated cross-section data. 52 The transference of competencies in the scope of education was approved by the Decree-Law n.º 144/2008 and so in the same or the previous year to the signature of execution contracts. 27 assumed additional responsibilities started to receive higher compensations and spend more on education, contrasting with the lower values observed before the reform. In addition, these results might indicate that those municipalities which spent less on education and received lower amounts of compensation before decentralisation were the ones which adopted the reform. Therefore, the selection of municipalities to participate in the reform might not have been entirely random. There seems to be also an increase in the percentage of students enroled in the second cycle of public education, even though it was only verified in 2013 and 2014. Furthermore, the results of the flexible model do not provide evidence of considerable variations in the impacts and their intensity over time, even though some fluctuations are observed over time. That is especially true in the case of municipal expenses and compensations received. For those variables, the coefficients’ magnitude appears to decrease after 2011, indicating a reduction in the difference in amounts registered between decentralised and non-decentralised municipalities. As described, the model was estimated between 2004 and 2015 due to the possibility of overlapping effects. Nonetheless, to check the robustness of results and identify the likeliness of impacts simultaneity, the same model was estimated for the entire analysis period until 2019. Those results are presented in Table 19,Table 20 and Table 21 of Appendix D. Even though no significant differences are spotted in most educational outcomes, the results for municipal expenses appear to differ. When including in the analysis the years after 2015, the coefficients of expenses and compensations are statistical significant and negative from 2015 onwards. Remarkably, those values suggest that decentralised municipalities have spent less on education after 2015 than expected in the case of not signing the contracts. Those results are contrary to the expectations and might evidence the likely overlapping of effects with the second reform since the negative coefficients are only significant after 2015. Regarding the second decentralisation moment, the analysis is not so straightforward since the estimation of pre-treatment coefficients for the dummy variable representing 2015’s contracts is also likely to be confounded by the annual effects of the first reform. Therefore, the results depicted in Table 11,Table 12 and Table 13 might be a potential consequence of that overlap. Along with the sporadic significance observed for some dependent variables, there seem to be many significant coefficients in the periods before treatment for specific indicators. That is the case of education expenses and the percentage of public school enrolment regarding pre-primary education, the schooling rate and the average of national exams in secondary education as well as the retention rates of the second and third cycles of basic education. 34 Table 8: Effects of the 1st reform in municipal accounts - Flexible Model VARIABLES Expenditures (ps) Compensations Received (ps) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year=2007) 2.740 -111.2 22.48 -193.0*** -247.5*** -87.33*** (0.0224) (-0.915) (0.329) (-2.797) (-5.345) (-4.764) Decentralised(Year=2008) 36.04 -143.7 14.25 -110.4 -277.4*** -89.53*** (0.320) (-1.227) (0.216) (-1.498) (-6.093) (-4.535) Decentralised(Year=2009) 171.6 150.6 126.4* -102.2 58.46 8.666 (1.365) (1.208) (1.703) (-1.506) (0.962) (0.435) Decentralised(Year=2010) 181.7* 107.0 93.10*** -60.70 134.3* 31.75* (1.768) (1.398) (2.771) (-0.786) (1.832) (1.743) Decentralised(Year=2011) 198.4** 219.1*** 114.9*** -100.1 169.4*** 39.70** (2.317) (2.895) (3.573) (-1.478) (2.889) (2.181) Decentralised(Year=2012) 109.5 132.2** 69.17*** -54.97 125.6*** 29.00** (1.474) (2.224) (3.009) (-0.956) (2.716) (2.188) Decentralised(Year=2013) 437.1 98.81 132.6 19.69 65.66 21.73* (1.283) (1.574) (1.586) (0.308) (1.618) (1.767) Decentralised(Year=2014) 4.432 58.50 28.47* -54.39 28.70 0.341 (0.0896) (1.234) (1.748) (-1.071) (0.830) (0.0274) Observations 2,475 2,475 2,475 2,475 2,475 2,475 Number of municipality_id 275 275 275 275 275 275 Adjusted R-squared 0.024 0.097 0.031 0.130 0.162 0.115 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 275 municipalities, but consider only the period 2004 - 2015. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 35 Table 9: Effects of the 1st reform on educational outcomes - Flexible Model VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic Decentralised(Year=2005) 0.321 0.0571 -0.376 -1.874 -3.348* (0.853) (0.0756) (-0.468) (-0.999) (-1.757) Decentralised(Year=2006) 0.140 0.384 -0.00627 -0.301 -2.653 (0.405) (0.472) (-0.00647) (-0.157) (-1.409) Decentralised(Year=2007) -0.0934 0.490 -0.486 0.328 -2.981* (-0.279) (0.750) (-0.631) (0.189) (-1.707) Decentralised(Year=2008) -0.218 -0.0703 -0.782 -0.539 -3.370* (-0.614) (-0.109) (-1.075) (-0.291) (-1.792) Decentralised(Year=2009) -0.322 -0.171 -0.333 1.007 -4.938* (-0.939) (-0.276) (-0.489) (0.601) (-1.891) Decentralised(Year=2010) -0.396 -0.745 0.275 0.792 -3.052 (-1.333) (-1.174) (0.407) (0.509) (-1.021) Decentralised(Year=2011) -0.131 -0.320 -0.467 1.625 -1.802 (-0.437) (-0.563) (-0.666) (1.004) (-0.803) Decentralised(Year=2012) -0.400 0.398 -0.585 2.735 -1.316 (-1.407) (0.614) (-0.825) (1.544) (-0.913) Decentralised(Year=2013) -0.400 0.398 -0.585 2.735 -1.316 (-1.407) (0.614) (-0.825) (1.544) (-0.913) Decentralised(Year=2014) 0.275 0.491 -0.565 -0.835 0.303 (0.882) (0.891) (-0.975) (-0.949) (0.497) Log(Population)t−1-3.512** -3.340 4.885 -5.337 -16.64 (-2.056) (-0.757) (1.094) (-0.322) (-1.224) Log(Month.Earnings)t−12.230* 0.212 -0.0908 -11.68 7.960 (1.655) (0.0828) (-0.0317) (-1.618) (0.841) %Unemploy.t−10.0222 0.0643 -0.176 -0.0142 -0.110 (0.414) (0.587) (-1.275) (-0.0428) (-0.334) %Popula.Higher.Educ.t−1-0.0229 -0.368** -0.156 -1.082* 0.285 (-0.415) (-2.390) (-0.811) (-1.886) (0.626) Observations 3,025 3,025 3,025 3,025 3,025 Number of municipality_id 275 275 275 275 275 Adjusted R-squared 0.165 0.331 0.334 0.260 0.338 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 275 municipalities, but consider only the period 2004 - 2015. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 36 Table 10: Effects of the 1st reform on educational outcomes - Flexible Model (cont.) VARIABLES Public School Enrolment Rates (level of education) Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Decentralised(Year=2005) 2.314 0.333 1.239 -0.331 (1.597) (0.396) (0.735) (-0.241) Decentralised(Year=2006) 2.300* 0.644 1.125 -0.542 (1.661) (0.817) (0.656) (-0.348) Decentralised(Year=2007) 1.501 0.865 1.499 -0.868 (1.161) (1.164) (0.878) (-0.657) Decentralised(Year=2008) 1.297 0.892 1.418 -0.222 (1.091) (1.321) (0.892) (-0.185) Decentralised(Year=2009) -0.112 0.881 0.562 -1.319 (-0.103) (1.376) (0.357) (-0.618) Decentralised(Year=2010) 0.253 0.284 0.507 -0.852 (0.267) (0.519) (0.354) (-0.437) Decentralised(Year=2011) -0.451 0.415 1.012 -1.574 (-0.537) (0.813) (0.904) (-0.936) Decentralised(Year=2012) -0.271 0.113 1.389 -0.788 (-0.372) (0.269) (1.449) (-0.587) Decentralised(Year=2013) -0.963 0.249 2.093** 1.271 (-1.490) (0.744) (2.266) (1.268) Decentralised(Year=2014) -0.401 0.101 1.333*** 0.0999 (-0.867) (0.451) (2.700) (0.167) Log(Population)t−19.588 1.580 -17.01 -3.234 (0.972) (0.225) (-1.481) (-0.251) Log(Month.Earnings)t−1-9.503* -7.368** -4.079 -6.663 (-1.686) (-2.220) (-0.509) (-0.961) %Unemploy.t−10.327 0.118 -0.116 0.0584 (1.615) (0.861) (-0.458) (0.215) %Popula.Higher.Educ.t−10.0388 -0.251* -0.0461 -0.0928 (0.122) (-1.699) (-0.218) (-0.356) Observations 2,749 1,092 1,122 1,623 Number of municipality_id 255 130 151 243 Adjusted R-squared 0.068 0.058 0.076 0.202 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations consider only the 2004 - 2015 period and encompass 275 municipalities, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 37 Table 11: Effects of the 2nd reform in municipal accounts - Flexible Model VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year=2007) 567.7*** -173.1 23.31 -180.9 -317.1** -116.5 (3.468) (-0.989) (0.318) (-0.628) (-2.245) (-1.617) Decentralised(Year=2008) 515.8*** -246.3 -16.19 -185.2 -298.7** -115.0 (2.956) (-1.465) (-0.226) (-0.671) (-2.037) (-1.564) Decentralised(Year=2009) 358.8** -76.16 -3.074 -3.411 -91.62 -46.47 (2.449) (-0.462) (-0.0529) (-0.0109) (-0.521) (-0.643) Decentralised(Year=2010) 406.7*** -149.3 -10.89 -44.85 -125.7 -57.20 (3.046) (-0.876) (-0.192) (-0.136) (-0.702) (-0.763) Decentralised(Year=2011) 451.1*** -174.2 3.476 -157.6 -208.8 -89.96 (3.319) (-1.248) (0.0718) (-0.567) (-1.306) (-1.331) Decentralised(Year=2012) 407.0*** -131.1 10.69 -178.3 -194.6 -88.03 (3.287) (-0.983) (0.203) (-0.625) (-1.220) (-1.304) Decentralised(Year=2013) 251.1 -173.7 -24.60 -310.3 -227.3 -110.8 (1.321) (-1.334) (-0.413) (-1.009) (-1.584) (-1.618) Decentralised(Year=2014) 417.0*** -127.2 33.00 -282.8 -247.6* -111.1 (3.580) (-1.135) (0.772) (-0.943) (-1.758) (-1.441) Decentralised(Year=2015) 403.6*** -39.32 54.50 -256.6 -77.68 -53.79 (3.455) (-0.295) (1.152) (-0.997) (-0.519) (-0.843) Decentralised(Year=2016) 382.6*** 6.626 68.08** -191.1 128.9 24.27 (3.682) (0.0853) (2.094) (-0.909) (1.301) (0.495) Decentralised(Year=2017) 192.4** 3.835 30.87 -151.5 54.50* 0.513 (2.296) (0.0622) (0.943) (-0.774) (1.665) (0.0141) Decentralised(Year=2018) 136.9** -6.421 8.420 -49.43 -47.74 -22.94 (2.074) (-0.126) (0.379) (-0.297) (-1.437) (-0.800) Observations 3,611 3,614 3,614 3,611 3,614 3,614 Number of municipality_id 278 278 278 278 278 278 Adjusted R-squared 0.069 0.066 0.059 0.111 0.096 0.049 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 278 municipalities. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 38 Table 12: Effects of the 2nd reform on educational outcomes - Flexible Model VARIABLES Retention Rates Transition Rate Average Exam Classifications Schooling Rate (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Secondary 3rd Cycle (Basic) Secondary Pre-Primary Decentralised(Year=2005) 0.238 -2.079** 0.986 2.960 -0.140 (0.471) (-2.100) (0.659) (0.962) (-0.0382) Decentralised(Year=2006) 0.250 -0.711 -2.071 3.222* -3.088 (0.573) (-0.815) (-1.624) (1.699) (-0.616) Decentralised(Year=2007) -0.494 -1.178 -2.514** 0.604 2.013 (-1.033) (-1.195) (-2.022) (0.277) (0.587) Decentralised(Year=2008) 0.591 -0.237 -1.537 1.067 -0.00911 0.661 -1.861 (0.956) (-0.249) (-1.201) (0.704) (-0.185) (0.302) (-0.533) Decentralised(Year=2009) 0.645 -0.232 0.139 0.805 0.0111 3.394*** -0.104 (1.443) (-0.326) (0.130) (0.526) (0.251) (2.615) (-0.0256) Decentralised(Year=2010) 0.0756 0.484 3.780** 0.414 0.0175 3.325* -3.937 (0.240) (0.637) (2.209) (0.312) (0.384) (1.749) (-1.274) Decentralised(Year=2011) 0.170 -0.929 -0.782 -0.302 0.0532 3.485*** -3.179 (0.375) (-1.209) (-0.853) (-0.167) (1.149) (3.377) (-1.044) Decentralised(Year=2012) -0.0812 -1.162 -0.0410 -0.246 0.0511 2.897* 1.123 (-0.205) (-1.355) (-0.0325) (-0.155) (1.136) (1.873) (0.187) Decentralised(Year=2013) -0.214 -2.304** -0.382 1.738 0.0357 1.711 2.211 (-0.499) (-2.465) (-0.459) (1.296) (0.895) (1.208) (0.369) Decentralised(Year=2014) -0.00145 -1.250 0.601 0.645 0.0408 3.536** -1.810 (-0.00316) (-1.116) (0.441) (0.297) (0.699) (2.126) (-0.436) Decentralised(Year=2015) -0.616 -1.313* 0.0295 -0.245 -0.00886 2.944** 1.766 (-1.078) (-1.719) (0.0309) (-0.249) (-0.393) (2.104) (0.636) Decentralised(Year=2016) -0.485 -1.373** -1.031 1.308 0.0142 0.590 0.870 (-1.285) (-2.399) (-0.971) (1.079) (0.280) (0.470) (0.456) Decentralised(Year=2017) -0.00602 -1.045 1.017 1.226 0.0164 1.763* 3.099* (-0.0204) (-1.061) (1.016) (1.218) (0.331) (1.659) (1.811) Decentralised(Year=2018) -0.113 -0.681 -0.198 0.260 -0.0178 1.112 0.812 (-0.429) (-0.725) (-0.234) (0.219) (-0.564) (1.366) (0.791) Log(Population)t−1-3.978*** -6.137** -1.287 -6.791 -0.205 -11.28 28.34** (-2.676) (-1.998) (-0.413) (-1.421) (-1.400) (-1.548) (2.312) Log(Month.Earnings)t−11.367 -0.337 1.911 2.777 0.232*** 9.363* -3.132 (1.330) (-0.132) (0.853) (1.046) (2.600) (1.710) (-0.523) %Unemploy.t−10.00862 0.0929 -0.0585 -0.0888 -0.00629 -0.197 -0.750*** (0.220) (1.132) (-0.611) (-0.671) (-1.615) (-1.169) (-2.669) %Popula.Higher.Educ.t−10.0307 -0.121 0.231** -0.242 -0.000654 -0.515*** -0.510 (0.884) (-1.392) (2.462) (-1.643) (-0.148) (-3.069) (-1.230) Observations 4,113 4,088 4,152 3,789 3,317 2,005 4,170 Number of municipality_id 278 278 278 262 277 218 278 Adjusted R-squared 0.226 0.391 0.536 0.596 0.580 0.587 0.234 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass all 278 mainland municipalities, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 39 Table 13: Effects of the 2nd reform on educational outcomes - Flexible Model (cont.) VARIABLES Schooling Rates Public School Enrolment Rates (level of education) Basic Secondary Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Secondary Decentralised(Year=2005) -2.741 -9.294 -4.073* -2.362 -5.385 -7.261 -4.468 (-0.757) (-0.931) (-1.820) (-1.576) (-1.221) (-1.619) (-0.808) Decentralised(Year=2006) -2.456 -4.659 -3.330* -2.463 -5.958 -7.152 -4.042 (-0.639) (-0.487) (-1.911) (-1.437) (-1.371) (-1.513) (-0.782) Decentralised(Year=2007) -0.995 -10.49 -5.040** -2.568 -4.804 -7.588 -4.764 (-0.278) (-1.091) (-1.982) (-1.492) (-1.067) (-1.624) (-0.874) Decentralised(Year=2008) -2.669 -8.701 -4.957*** -2.265 -5.009 -7.262 -5.483 (-0.696) (-1.010) (-2.758) (-1.379) (-1.149) (-1.580) (-1.086) Decentralised(Year=2009) 1.251 -19.58* -2.875** -1.411 -2.756 0.128 -1.162 (0.248) (-1.811) (-1.991) (-1.028) (-0.609) (0.0297) (-0.244) Decentralised(Year=2010) -1.666 -14.32 -4.142*** -1.099 -2.334 1.310 -1.859 (-0.356) (-1.649) (-2.894) (-1.284) (-0.481) (0.297) (-0.408) Decentralised(Year=2011) -0.468 -18.31** -3.565*** -0.582 -2.565 -2.060 -4.125 (-0.106) (-2.294) (-2.985) (-0.888) (-0.526) (-0.462) (-0.977) Decentralised(Year=2012) 1.417 -8.896 -2.329** -0.217 -4.950 -2.183 -2.692 (0.401) (-1.360) (-2.465) (-0.367) (-1.095) (-0.511) (-0.597) Decentralised(Year=2013) 2.296 -11.42 -2.221*** -0.105 -5.905 -6.180 -6.485 (0.901) (-1.524) (-2.643) (-0.158) (-1.336) (-1.293) (-1.618) Decentralised(Year=2014) 2.428 -9.005 -2.200* 0.391 -4.511 -5.345 -6.823 (0.997) (-1.312) (-1.795) (0.405) (-0.979) (-1.151) (-1.413) Decentralised(Year=2015) 2.405 -6.185 -1.569 0.233 -5.103 -5.583 -6.801 (1.076) (-1.045) (-1.604) (0.389) (-1.176) (-1.243) (-1.425) Decentralised(Year=2016) 3.207 -2.334 -1.811* 0.808 -5.343 -6.443 -6.213 (1.229) (-0.524) (-1.937) (1.296) (-1.274) (-1.486) (-1.289) Decentralised(Year=2017) 2.383 2.343 -0.815 1.021 -3.350 -5.442 -5.401 (0.832) (0.450) (-0.803) (1.220) (-1.113) (-1.403) (-1.386) Decentralised(Year=2018) 1.064 0.122 -0.373 1.636 -1.900 -3.785* -3.373 (0.903) (0.0385) (-0.427) (1.230) (-1.199) (-1.864) (-1.216) Log(Population)t−1-13.46 -33.04 13.92 -6.384 -15.62* -2.234 -22.90** (-1.031) (-1.205) (1.528) (-1.111) (-1.816) (-0.263) (-2.180) Log(Month.Earnings)t−15.478 19.96 -9.479** -2.448 -3.457 -14.35 -15.96 (0.587) (1.165) (-2.019) (-0.815) (-0.573) (-1.277) (-1.305) %Unemploy.t−1-0.0861 -0.502 0.0898 0.185 -0.257 -0.0769 0.323 (-0.293) (-0.833) (0.432) (1.289) (-0.914) (-0.275) (1.123) %Popula.Higher.Educ.t−1-0.166 1.255 0.202 -0.290** -0.211 -0.159 -0.208 (-0.339) (0.984) (0.807) (-2.239) (-1.025) (-0.722) (-0.816) Observations 4,170 3,948 3,754 1,510 1,534 2,144 2,292 Number of municipality_id 278 275 257 139 158 246 244 Adjusted R-squared 0.360 0.304 0.106 0.122 0.129 0.197 0.068 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass all 278 mainland municipalities, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 40 6.3 Multiple time periods approach The results of the extension of the DD approach to a multiple time periods framework may be observed in Table 14 and Table 15. The results obtained with this extension are interesting and concern the development of further analyses. Those include the isolation of effects by each group of municipalities, and the estimation of the overall ATT by group and over the years, derived with an event study. The computation of the overall ATT demonstrates that the retention rate in the second cycle of basic education is lower than it would be had decentralisation not happened, as detailed in Table 14. This variation corresponded to a 1.41 percentage points decrease in general terms, but the isolation of effects by different groups did not provide any statistical evidence of differences in the impacts faced. This table also shows evidence of decentralisation effects on the schooling rate of pre-primary education, which was 4.71 percentage points higher after decentralisation. When isolating the effects by group, it is observable that the 2010 Group experienced a considerable variation of 13.22 percentage points in its pre-primary schooling rate, for which the ATT of the 2009 Group is not statistically significant. The 2010 Group seems also to have faced a decrease of 3.77 percentage points in the retention rate of the third cycle of basic education, even though the same did not verify in the case of the 2009 Group and general terms. Regarding the percentage of students enroled in public education, the results displayed in Table 15 demonstrate no evidence of the effects prompted by decentralisation on those enrolment rates in general terms. Nonetheless, the computation of the ATT by each treated group suggests that those rates increased after the reform, but only on one group: the 2009 Group registered higher rates in the case of the second and the third cycles of basic education, while the 2010 Group faced an increase in that rate only in the first cycle of basic education. Additional results from this approach may be found in Table 34 to Table 39, available in Appendix E. Starting with the estimations for each group of municipalities, it is observable that, for the 2009 Group , there is statistical evidence of the effects that decentralisation had on the retention rate in the second cycle of basic education. As portrayed in Table 34, the coefficients resulting from the successive comparisons of two different years started to be significant in 2011, two years after this group started to experience the effects. As hypothesised, the retention rates are lower after 2011 than if no decentralisation has happened. The variations in this indicator correspond to a more than one percentage point difference, increasing to a variation of about three percentage points in periods further away from decentralisation. Moreover, the 2009 Group appear to have also experienced an increase in the percentage of students enroled in public schools regarding the second cycle of basic education, as observable in Table 35. As expected, there is evidence of an increase in this indicator when the years after decentralisation are used 41 to compare the outcomes, with those variations ranging from seven to fifteen percentage points in the periods after 2010. Nonetheless, the same table also shows that the coefficients related to this rate in the third cycle of basic education are statistically significant when the periods before treatment are used to compute the ATT. Such results suggest differences between municipalities regarding their public enrolment rates even before decentralisation. Therefore, the estimation of this group’s ATT for the following years in this specific indicator may be biased. For the municipalities that first experienced effects in 2010, there is also statistical evidence of decentralisation impacts on retention rates, in addition to the effects verified on the pre-primary schooling rate. Table 36 demonstrates that retention rates are lower than they would be without decentralisation. Those variations range from one to six percentage points differences, depending on the year post-reform used to compute the coefficient and the level of education considered. The effects faced by the 2010 Group appear thus to be more intense than the ones experienced by the 2009 Group , in addition to the largest number of significant coefficients obtained. In turn, there was also a considerable increase in the schooling rate of pre-primary education after the signing of contracts. This effect corresponded to an about fifteen percentage points variation in years further away from decentralisation, as observable in the same table. In addition, Table 37 shows that the coefficients associated with the percentage of students enroled in the first and second cycles of public education are statistically significant when computed with pretreatment periods. Hence, as before, there seem to be differences between municipalities before decentralisation in terms of enrolment in public schools, which may bias the post-decentralisation results56. The extension of the DD to a multiple time periods framework also allows the computation of the overall ATT by periods before and after treatment through an event study. The results of that analysis are represented in Table 38 and Table 39 of Appendix E. As observable, the ATT is statistically significant in almost all periods after decentralisation in the case of the retention rates registered in the second cycle of basic education, which appears to have decreased in the years following the reform. Moreover, the coefficients associated with the pre-primary education schooling rate are statistically significant and positive in most post-treatment periods. As before, the statistical significance of the coefficients related to the ratio of public school enrolment regarding the second cycle of basic education in the periods before decentralisation renders interpreting the post-reform values impossible due to the likelihood of bias issues. 56 In addition, the hypothesis stating that the pre-treatment trends are equal to zero in the statistical test presented by Callaway and Sant’Anna (2021) is rejected in the case of the public enrolment ratio in the second cycle of basic education. These results suggest that this variable’s coefficients may be biased, even if the regional-specific trends are included in the regressions. This conclusion follows the observed in section 5. 42 The graphs displayed in Figure 3 facilitate visualising those results. Each graph corresponds to a specific educational outcome and depicts the evolution of the ATT by periods before and after the signature of contracts. Table 14: Average Treatment Effect on Treated VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic ATT -0.320 -1.414* -0.446 4.713* 2.470 (-0.85) (-1.80) (-0.48) (1.68) (0.73) ATT by group 2009 Group -0.246 -1.403 0.242 2.950 2.048 (-0.58) (-1.64) (0.24) (1.00) (0.58) 2010 Group -0.675 -1.465 -3.766** 13.221** 4.504 (-1.48) (-0.89) (-1.98) (2.37) (0.64) Observations 3,025 3,025 3,025 3,025 3,025 Number of municipality_id 275 275 275 275 275 Notes: The estimations encompass 275 municipalities, excluding the three municipalities that signed contracts after 2011, and cover the 2004-2015 period. The control group considers all municipalities that did not sign a contract in 2009 or 2010. All regressions include a vector of control variables and regional-specific trends. Z-statistics, based on standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 43 Faguet, J. P., & Sánchez, F. (2008). Decentralization’s Effects on Educational Outcomes in Bolivia and Colombia. World Development , 36 (7), 1294–1316. https://doi.org/10.1016/j.worlddev.2007. 06.021 Galiani, S., Gertler, P., & Schargrodsky, E. (2008). School decentralization: Helping the good get better, but leaving the poor behind. Journal of Public Economics , 92 (10), 2106–2120. https://doi.org/ 10.1016/j.jpubeco.2008.05.004 Goldhaber, D. (1996). Public and Private High Schools: Is School Choice an Answer to the Productivity Problem? Economics of Education Review , 15 (2), 93–109. https://doi.org/10.1016/02727757(95)00042-9 Guerra, S. C., & Lastra-Anadón, C. X. (2019). The quality-access tradeoff in decentralizing public services: Evidence from education in the OECD and Spain. Journal of Comparative Economics , 47 (2), 295– 316. https://doi.org/10.1016/j.jce.2018.12.004 Hanushek, E. A., Link, S., & Woessmann, L. (2013). Does school autonomy make sense everywhere? Panel estimates from PISA. Journal of Development Economics , 104 , 212–232. https://doi.org/10. 1016/j.jdeveco.2012.08.002 Hoxby, C. M. (2003). School Choice and School Productivity. Could School Choice Be a Tide that Lifts All Boats? In C. M. Hoxby (Ed.), The Economics of School Choice (pp. 287–341). University of Chicago Press. http://www.nber.org/chapters/c10091 IAVE I.P. (2022). Provas e Exames: Calendário . Retrieved October 25, 2022, from https://iave.pt/provase-exames/calendario/ IEFP. (2022). Estatísticas Mensais por Concelhos . Retrieved November 24, 2022, from https://www.iefp. pt/estatisticas INCM. (2022). Diário da República Eletrónico . Retrieved October 15, 2022, from https://dre.pt/dre INE. (2022a). Base de Dados - Educação, formação e aprendizagem . Retrieved November 8, 2022, from https://www.ine.pt/xportal/xmain?xpid=INE&xpgid=ine_bdc_tree&contexto=bd&selTab=tab2 INE. (2022b). Base de Dados - População . Retrieved November 8, 2022, from https://www.ine.pt/ xportal/xmain?xpid=INE&xpgid=ine_bdc_tree&contexto=bd&selTab=tab2 INE. (2022c). Sistema de Metainformação - Conceitos . Retrieved November 24, 2022, from https://smi. ine.pt/Conceito/Detalhes/5086 International Monetary Fund. (2020). Fiscal Decentralization Dataset . Retrieved September 13, 2022, from https://data.imf.org/?sk=1C28EBFB-62B3-4B0C-AED3-048EEEBB684F Kyriacou, A. P., & Roca-Sagalés, O. (2019). Local Decentralization and the Quality of Public Services in Europe. Social Indicators Research , 145 , 755–776. https://doi.org/10.1007/s11205-01902113-z Lledó, V., Gbohoui, W., Ncuti, C., Hu, C., & Xiang, Y. (2020). The IMF Fiscal Decentralization Dataset . International Monetary Fund. Martinez-Vazquez, J., Lago-Peñas, S., & Sacchi, A. (2017). The Impact of Fiscal Decentralization: A Survey. Journal of Economic Surveys , 31 (4), 1095–1129. https://doi.org/10.1111/joes.12182 MCTES and ME. (2023). Alterações à conclusão do ensino secundário e acesso ao ensino superior . Ministério da Ciência, Tecnologia e Ensino Superior and Ministério da Educação. Retrieved March 14, 2023, from https://wwwcdn.dges.gov.pt/sites/default/files/sumario_executivo_revisao_ acesso_ao_ensino_superior_final_.pdf Oates, W. E. (1999). An Essay on Fiscal Federalism. Journal of Economic Literature , 37 (3), 1120–1149. https://www.jstor.org/stable/2564874 OECD. (2019). Making Decentralisation Work: A Handbook for Policy-Makers . OECD Publishing. Paris, France. https://doi.org/10.1787/g2g9faa7-en OECD. (2020). OECD Fiscal Decentralisation Database . Retrieved September 13, 2022, from https:// www.oecd.org/tax/federalism/fiscal-decentralisation-database/ 50 Salinas, P., & Solé-Ollé, A. (2018). Partial fiscal decentralization reforms and educational outcomes: A difference-in-differences analysis for Spain. Journal of Urban Economics , 107 , 31–46. https:// doi.org/10.1016/j.jue.2018.08.003 Secretário de Estado da Administração Local. (2014). PAE Programa Aproximar Educação - Descentralização de Competências na Área da Educação: Contrato de Educação e Formação Municipal - Memorando de Trabalho . Governo de Portugal. https://educar.files.wordpress.com/2014/12/ memorandotrabalhopae.pdf Tiebout, C. M. (1956). A Pure Theory of Local Expenditures. Journal of Political Economy , 64 (5), 416–424. https://www.jstor.org/stable/1826343 United Nations Conference on Trade and Development. (2021). Handbook of Statistics 2021 . United Nations. https://doi.org/10.18356/9789210010610 Veiga, L. G., Kurian, M., & Ardakanian, R. (2015). Intergovernmental Fiscal Relations: Questions of Accountability and Autonomy . Springer. https://doi.org/10.1007/978-3-319-06296-9 51 Appendix A Table 16: List of contracts signed by Portuguese municipalities 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Arcos de Valdevez 239/2009 16/09/2008 01/2009 Melgaço 249/2009 16/09/2008 01/2009 Monção 250/2009 16/09/2008 01/2009 Paredes de Coura 255/2009 16/09/2008 01/2009 Ponte da Barca 256/2009 16/09/2008 01/2009 Ponte de Lima 335/2009 16/02/2009 03/2009 Valença 262/2009 16/09/2008 01/2009 Viana do Castelo 269/2009 16/09/2008 01/2009 Vila Nova de Cerveira 264/2009 16/09/2008 01/2009 Amares 336/2009 16/02/2009 03/2009 Braga 242/2009 16/09/2008 01/2009 Terras de Bouro 260/2009 16/09/2008 01/2009 Cabeceiras de Basto 267/2009 16/09/2008 01/2009 Fafe 202/2009 16/09/2008 01/2009 Guimarães 204/2009 16/09/2008 01/2009 Vila Nova de Famalicão 562/2015 18/05/2015 07/2015 52 Table 16: List of contracts signed by Portuguese municipalities, continued 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Vizela 266/2009 16/09/2008 01/2009 Espinho 245/2009 16/09/2008 01/2009 Gondomar 247/2009 16/09/2008 01/2009 Maia 554/2015 18/05/2015 07/2015 Matosinhos 205/2009 16/09/2008 01/2009 555/2015 9/06/2015 08/2015 Oliveira de Azeméis 559/2015 18/05/2015 07/2015 Paredes 254/2009 16/09/2008 01/2009 Santo Tirso 230/2009 16/09/2008 01/2009 Trofa 208/2009 16/09/2008 01/2009 Vila do Conde 209/2009 16/09/2008 01/2009 Montalegre 207/2009 16/09/2008 01/2009 Baião 241/2009 16/09/2008 01/2009 Cinfães 244/2009 16/09/2008 01/2009 Felgueiras 203/2009 16/09/2008 01/2009 Lousada 248/2009 16/09/2008 01/2009 Paços de Ferreira 253/2009 16/09/2008 01/2009 Resende 257/2009 16/09/2008 01/2009 Armamar 240/2009 17/09/2008 01/2009 Carrazeda de Ansiães 243/2009 16/09/2008 01/2009 Freixo de Espada à Cinta 246/2009 16/09/2008 10/2008 Murça 252/2009 16/09/2008 01/2009 53 Table 16: List of contracts signed by Portuguese municipalities, continued 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Peso da Régua 338/2009 16/09/2008 01/2009 Sabrosa 339/2009 16/09/2008 01/2009 Santa Marta de Penaguião 268/2009 16/09/2008 01/2009 Tabuaço 258/2009 16/09/2008 01/2009 Tarouca 259/2009 16/09/2008 01/2009 Torre de Moncorvo 261/2009 16/09/2008 01/2009 Vila Nova de Foz Côa 265/2009 16/09/2008 01/2009 Mirandela 206/2009 16/09/2008 01/2009 Vila Flor 263/2009 16/09/2008 01/2009 Vimioso 259/2012 19/04/2012 09/2012 Alenquer 186/2009 16/09/2008 01/2009 Arruda dos Vinhos 190/2009 16/09/2008 01/2009 Lourinhã 195/2009 16/09/2008 01/2009 Nazaré 471/2009 24/09/2009 01/2010 Óbidos 197/2009 16/09/2008 01/2009 557/2015 18/05/2015 07/2015 Águeda 169/2009 16/09/2008 01/2009 549/2015 29/06/2015 08/2015 Ílhavo 470/2009 31/08/2009 01/2010 Oliveira do Bairro 472/2009 31/08/2009 01/2010 560/2015 18/05/2015 07/2015 Góis 469/2009 31/08/2009 01/2010 Mealhada 173/2009 16/09/2008 01/2009 556/2015 1/07/2015 09/2015 Mira 175/2009 16/09/2008 01/2009 54 Table 16: List of contracts signed by Portuguese municipalities, continued 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Mortágua 176/2009 16/09/2008 01/2009 Batalha 551/2015 18/05/2015 07/2015 Porto de Mós 179/2009 16/09/2008 03/2009 Castelo Branco 171/2009 16/09/2008 01/2009 Vila Velha de Ródão 185/2009 16/09/2008 01/2009 Entroncamento 25/2012 12/10/2011 01/2012 Ourém 473/2009 23/09/2009 01/2010 Sardoal 200/2009 01/04/2009 05/2009 Sertã 181/2009 16/09/2008 01/2009 Tomar 367/2009 23/09/2009 01/2010 Torres Novas 166/2009 16/09/2008 03/2009 Vila de Rei 184/2009 16/09/2008 01/2009 563/2015 18/05/2015 07/2015 Vila Nova da Barquinha 201/2009 16/09/2008 01/2009 Celorico da Beira 467/2009 31/08/2009 01/2010 Mêda 174/2009 16/09/2008 03/2009 Amadora 189/2009 16/09/2008 01/2009 550/2015 1/06/2015 08/2015 Cascais 552/2015 18/05/2015 07/2015 Loures 194/2009 16/09/2008 01/2009 Mafra 365/2009 09/09/2009 10/2009 Montijo 196/2009 16/09/2008 01/2009 Odivelas 366/2009 23/09/2009 01/2010 Oeiras 558/2015 17/07/2015 09/2015 Sintra 486/2009 21/09/2009 01/2010 55 Table 16: List of contracts signed by Portuguese municipalities, continued 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Grândola 221/2009 16/09/2008 01/2009 Sines 228/2009 16/09/2008 01/2009 Alvito 211/2009 16/09/2008 01/2009 Cuba 216/2009 16/09/2008 01/2009 Ferreira do Alentejo 219/2009 16/09/2008 01/2009 Ourique 224/2009 16/09/2008 01/2009 Vidigueira 690/2011 19/01/2011 03/2011 Almeirim 187/2009 16/09/2008 01/2009 Alpiarça 188/2009 16/09/2008 01/2009 Azambuja 191/2009 16/09/2008 01/2009 Cartaxo 192/2009 16/09/2008 01/2009 Coruche 468/2009 24/09/2009 01/2010 Golegã 193/2009 16/09/2008 01/2009 Rio Maior 198/2009 16/09/2008 01/2009 Santarém 199/2009 16/09/2008 01/2009 Arronches 212/2009 16/09/2008 01/2009 Campo Maior 214/2009 16/09/2008 01/2009 Crato 215/2009 16/09/2008 01/2009 553/2015 30/06/2015 08/2015 Gavião 220/2009 16/09/2008 01/2009 Nisa 223/2009 16/09/2008 01/2009 Ponte de Sor 225/2009 16/09/2008 01/2009 Sousel 561/2015 18/05/2015 07/2015 Alandroal 210/2009 16/09/2008 01/2009 Borba 213/2009 16/09/2008 01/2009 Estremoz 217/2009 16/09/2008 01/2009 Évora 218/2009 16/09/2008 01/2009 56 Table 16: List of contracts signed by Portuguese municipalities, continued 2009 Contracts 2015 Contracts Municipality Contract Number Celebration Date Effects Date Contract Number Celebration Date Effects Date Mourão 222/2009 16/09/2008 01/2009 Portel 226/2009 16/09/2008 01/2009 Reguengos de Monsaraz 227/2009 16/09/2008 01/2009 Albufeira 170/2009 16/09/2008 01/2009 Alcoutim 474/2009 22/09/2009 10/2009 Faro 172/2009 16/09/2008 01/2009 Lagos 475/2009 24/09/2009 10/2009 Loulé 476/2009 24/09/2009 10/2009 Monchique 251/2009 16/09/2008 01/2009 Olhão 177/2009 16/09/2008 01/2009 Portimão 178/2009 16/09/2008 01/2009 São Brás de Alportel 180/2009 16/09/2008 01/2009 Silves 182/2009 16/09/2008 01/2009 Tavira 183/2009 16/09/2008 01/2009 Vila do Bispo 477/2009 22/09/2009 10/2009 Vila Real de Santo António 478/2009 24/09/2009 10/2009 Example of 2009 Contract 57 Appendix C Figure 4: Trends in municipal accounts before the 1st reform 66 Figure 5: Trends in municipal accounts before the 2nd reform 67 Appendix D Table 19: Effects of the 1st reform in municipal accounts - Flexible Model (2004 - 2019) VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year=2007) -61.94 -292.4** -68.05 -224.7** -300.3*** -119.2*** (-0.422) (-2.217) (-0.984) (-2.477) (-5.066) (-4.399) Decentralised(Year=2008) -20.91 -324.9** -74.49 -139.0 -334.1*** -122.7*** (-0.156) (-2.564) (-1.114) (-1.502) (-5.495) (-4.277) Decentralised(Year=2009) 122.3 -30.68 39.48 -127.8 -2.156 -25.80 (0.809) (-0.235) (0.525) (-1.429) (-0.0306) (-0.920) Decentralised(Year=2010) 140.2 -74.31 7.950 -83.23 69.73 -4.023 (1.030) (-0.808) (0.208) (-0.835) (0.817) (-0.141) Decentralised(Year=2011) 164.7 37.75 31.58 -119.6 100.9 2.627 (1.303) (0.420) (0.798) (-1.380) (1.409) (0.0922) Decentralised(Year=2012) 83.41 -49.17 -12.38 -71.44 53.09 -9.380 (0.708) (-0.633) (-0.371) (-0.857) (0.881) (-0.367) Decentralised(Year=2013) 418.7 -82.63 52.82 6.249 -10.77 -17.96 (1.208) (-0.981) (0.617) (0.0688) (-0.185) (-0.686) Decentralised(Year=2014) -6.193 -123.0* -49.51 -64.79 -51.68 -40.66 (-0.0596) (-1.707) (-1.581) (-0.814) (-0.902) (-1.530) Decentralised(Year=2015) -2.902 -181.5*** -76.18** -7.379 -84.34 -42.31 (-0.0313) (-2.922) (-2.586) (-0.103) (-1.412) (-1.508) Decentralised(Year=2016) 12.18 -145.2*** -48.91* -141.5** -88.21 -56.17** (0.133) (-2.666) (-1.944) (-2.143) (-1.432) (-1.971) Decentralised(Year=2017) -61.34 -117.7*** -43.85** -55.78 -99.22** -47.92** (-0.782) (-2.613) (-2.055) (-0.944) (-2.561) (-2.413) Decentralised(Year=2018) -59.64 -123.1*** -47.55*** 16.56 -42.09 -17.30 (-0.811) (-3.476) (-2.611) (0.291) (-1.620) (-1.437) Observations 3,572 3,575 3,575 3,572 3,575 3,575 Number of municipality_id 275 275 275 275 275 275 Adjusted R-squared 0.074 0.079 0.066 0.114 0.126 0.070 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 275 municipalities and cover the entire period of analysis. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 68 Table 20: Effects of the 1st reform on educational outcomes - Flexible Model (2004 - 2019) VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic Decentralised(Year=2005) 0.577* -0.192 -0.149 -1.390 -1.813 (1.737) (-0.284) (-0.203) (-0.732) (-0.914) Decentralised(Year=2006) 0.407 0.138 0.210 0.102 -1.137 (1.218) (0.178) (0.218) (0.0524) (-0.587) Decentralised(Year=2007) 0.185 0.249 -0.282 0.649 -1.482 (0.598) (0.359) (-0.389) (0.362) (-0.817) Decentralised(Year=2008) 0.0729 -0.303 -0.585 -0.406 -1.878 (0.208) (-0.518) (-0.882) (-0.209) (-0.942) Decentralised(Year=2009) -0.0222 -0.402 -0.141 1.104 -3.475 (-0.0768) (-0.748) (-0.221) (0.600) (-1.287) Decentralised(Year=2010) -0.0879 -0.977* 0.454 0.975 -1.629 (-0.287) (-1.662) (0.705) (0.626) (-0.534) Decentralised(Year=2011) 0.193 -0.544 -0.309 1.786 -0.395 (0.648) (-0.994) (-0.461) (0.961) (-0.171) Decentralised(Year=2012) -0.0660 0.176 -0.435 2.930 0.0563 (-0.225) (0.281) (-0.609) (1.509) (0.0355) Decentralised(Year=2013) 0.110 -0.999 0.904 1.114 0.588 (0.363) (-1.447) (1.430) (0.588) (0.493) Decentralised(Year=2014) 0.628** 0.279 -0.418 -0.897 1.635 (2.110) (0.451) (-0.687) (-0.509) (1.539) Decentralised(Year=2015) 0.358 -0.212 0.149 -0.298 1.320 (1.014) (-0.353) (0.260) (-0.179) (1.393) Decentralised(Year=2016) 0.265 -0.207 -0.0375 0.759 1.540 (0.959) (-0.418) (-0.0701) (0.517) (1.597) Decentralised(Year=2017) 0.369 0.427 0.161 1.035 0.920 (1.400) (0.825) (0.322) (0.809) (1.192) Decentralised(Year=2018) 0.481* 1.016** -0.0581 1.174 0.641 (1.903) (2.336) (-0.113) (1.391) (1.268) Log(Population)t−1-4.194*** -7.750*** -2.077 23.98** -17.24 (-2.780) (-2.622) (-0.678) (2.045) (-1.269) Log(Month.Earnings)t−11.582 -0.331 1.745 -4.932 5.363 (1.553) (-0.134) (0.781) (-0.820) (0.598) %Unemploy.t−10.00953 0.0863 -0.0757 -0.769*** -0.0930 (0.239) (1.052) (-0.763) (-2.691) (-0.318) %Popula.Higher.Educ.t−10.0319 -0.0979 0.251*** -0.474 -0.0323 (0.913) (-1.167) (2.604) (-1.095) (-0.0651) Observations 4,070 4,043 4,107 4,125 4,125 Number of municipality_id 275 275 275 275 275 Adjusted R-squared 0.229 0.400 0.536 0.253 0.362 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 275 municipalities and cover the entire period of analysis. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 69 Table 21: Effects of the 1st reform on educational outcomes - Flexible Model (2004 - 2019) (cont.) VARIABLES Public School Enrolment Rates (level of education) Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Decentralised(Year=2005) 2.334 0.797 1.546 0.441 (1.479) (1.031) (0.664) (0.187) Decentralised(Year=2006) 2.311 1.145 1.431 0.550 (1.542) (1.561) (0.615) (0.248) Decentralised(Year=2007) 1.495 1.366* 1.891 0.329 (1.031) (1.944) (0.782) (0.154) Decentralised(Year=2008) 1.243 1.446** 1.847 0.704 (0.933) (2.175) (0.798) (0.357) Decentralised(Year=2009) -0.186 1.471** 0.607 -0.389 (-0.147) (2.310) (0.274) (-0.159) Decentralised(Year=2010) 0.235 0.901 0.803 0.169 (0.211) (1.530) (0.383) (0.0724) Decentralised(Year=2011) -0.471 1.072* 1.343 -0.588 (-0.454) (1.869) (0.738) (-0.267) Decentralised(Year=2012) -0.286 0.806 1.933 0.269 (-0.302) (1.509) (1.210) (0.142) Decentralised(Year=2013) -0.987 1.094* 2.928* 2.421 (-1.061) (1.903) (1.725) (1.326) Decentralised(Year=2014) -0.507 0.923* 2.316 1.238 (-0.597) (1.737) (1.429) (0.736) Decentralised(Year=2015) -0.252 0.810* 1.182 1.159 (-0.327) (1.764) (0.676) (0.743) Decentralised(Year=2016) 0.254 -0.0817 1.496 1.585 (0.348) (-0.225) (0.926) (1.104) Decentralised(Year=2017) -0.0347 -0.488 1.900* 0.320 (-0.0559) (-1.394) (1.677) (0.223) Decentralised(Year=2018) -0.133 -0.495 -0.113 1.274 (-0.239) (-1.595) (-0.146) (1.067) Log(Population)t−115.52* -2.965 -14.04 -0.792 (1.661) (-0.446) (-1.449) (-0.0903) Log(Month.Earnings)t−1-10.21** -4.904* -4.537 -15.90 (-2.141) (-1.693) (-0.783) (-1.445) %Unemploy.t−10.110 0.0564 -0.366 -0.0979 (0.531) (0.416) (-1.346) (-0.353) %Popula.Higher.Educ.t−10.196 -0.257** -0.208 -0.139 (0.764) (-2.184) (-1.025) (-0.629) Observations 3,724 1,495 1,525 2,121 Number of municipality_id 255 138 156 243 Adjusted R-squared 0.111 0.115 0.127 0.190 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations cover the entire period of analysis and encompass 275 municipalities, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 70 Table 22: Effects of the 1st reform in municipal accounts - All municipalities VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year>=2010) 59.34 123.9 15.99 112.0** 200.3*** 67.36*** (0.661) (1.340) (0.292) (2.471) (5.896) (4.665) Observations 3,611 3,614 3,614 3,611 3,614 3,614 Number of municipality_id 278 278 278 278 278 278 Adjusted R-squared 0.070 0.071 0.062 0.114 0.106 0.059 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 278 municipalities, including those that only signed contracts after 2011. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 23: Effects of the 1st reform on educational outcomes - All municipalities VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic Decentralised(Year>=2010) 0.00205 -0.0544 0.244 0.314 2.289* (0.0130) (-0.149) (0.597) (0.247) (1.747) Log(Population)t−1-4.081*** -6.063* -1.384 28.55** -15.03 (-2.734) (-1.968) (-0.446) (2.311) (-1.117) Log(Month.Earnings)t−11.463 -0.250 1.690 -3.396 5.325 (1.436) (-0.0994) (0.756) (-0.570) (0.591) %Unemploy.t−10.00984 0.0919 -0.0622 -0.761*** -0.112 (0.251) (1.124) (-0.649) (-2.740) (-0.390) %Popula.Higher.Educ.t−10.0270 -0.123 0.245** -0.482 -0.0595 (0.772) (-1.413) (2.544) (-1.153) (-0.121) Observations 4,113 4,088 4,152 4,170 4,170 Number of municipality_id 278 278 278 278 278 Adjested R-squared 0.227 0.391 0.534 0.234 0.362 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass all 278 municipalities, including those that only signed contracts after 2011. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 24: Effects of the 1st reform on educational outcomes - All municipalities (cont.) VARIABLES Public School Enrolment Rates (level of education) Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Decentralised(Year>=2010) -1.583 -0.816 0.00974 0.330 (-1.617) (-1.519) (0.00751) (0.347) Log(Population)t−116.81* -2.755 -14.06 -2.399 (1.822) (-0.437) (-1.481) (-0.279) Log(Month.Earnings)t−1-10.64** -4.928* -4.544 -15.34 (-2.272) (-1.672) (-0.792) (-1.414) %Unemploy.t−10.0818 0.157 -0.285 -0.119 (0.397) (1.137) (-0.983) (-0.419) %Popula.Higher.Educ.t−10.178 -0.278** -0.204 -0.144 (0.703) (-2.255) (-0.996) (-0.651) Observations 3,754 1,510 1,534 2,144 Number of municipality_id 257 139 158 246 Adjusted R-squared 0.109 0.101 0.126 0.189 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass all 278 municipalities, including those that only signed contracts after 2011, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 71 Table 25: Effects of the 1st reform in municipal accounts - Excluding late adopters VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year>=2010) 122.3 221.0*** 71.11*** 86.07* 176.0*** 55.70*** (1.468) (3.993) (3.024) (1.721) (4.585) (3.524) Observations 3,289 3,289 3,289 3,289 3,289 3,289 Number of municipality_id 253 253 253 253 253 253 Adjusted R-squared 0.075 0.104 0.095 0.104 0.136 0.067 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 266 municipalities and the decentralisation variable considers only the 91 that started experiencing effects at the beginning of 2009. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 26: Effects of the 1st reform on educational outcomes - Excluding late adopters VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic Decentralised(Year>=2010) -0.0560 -0.377 0.00247 1.124 3.071* (-0.319) (-0.945) (0.00550) (0.847) (1.968) Log(Population)t−1-2.742** -6.559** -1.671 21.96* -12.00 (-2.393) (-2.023) (-0.528) (1.690) (-0.826) Log(Month.Earnings)t−11.470 -0.470 0.909 -2.865 2.535 (1.403) (-0.188) (0.397) (-0.487) (0.273) %Unemploy.t−10.0115 0.0433 -0.0548 -0.660** -0.103 (0.285) (0.537) (-0.538) (-2.381) (-0.337) %Popula.Higher.Educ.t−10.0391 -0.0959 0.275*** -0.406 -0.0787 (1.140) (-1.105) (2.817) (-0.915) (-0.162) Observations 3,742 3,717 3,779 3,795 3,795 Number of municipality_id 253 253 253 253 253 Adjusted R-squared 0.225 0.398 0.537 0.257 0.357 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 266 municipalities and the decentralisation variable considers only the 91 that started experiencing effects at the beginning of 2009. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 27: Effects of the 1st reform on educational outcomes - Excluding late adopters (cont.) VARIABLES Public School Enrolment Rates (level of education) Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Decentralised(Year>=2010) -1.709 -0.870 -0.712 0.185 (-1.565) (-1.653) (-0.945) (0.230) Log(Population)t−110.08 -4.441 -10.76 5.242 (1.001) (-0.571) (-1.015) (0.597) Log(Month.Earnings)t−1-10.69** -5.201 -4.188 -13.36 (-2.225) (-1.550) (-0.717) (-1.169) %Unemploy.t−10.0558 0.194 -0.0735 0.0290 (0.268) (1.390) (-0.287) (0.112) %Popula.Higher.Educ.t−10.232 -0.273** -0.201 -0.145 (0.896) (-2.185) (-0.936) (-0.643) Observations 3,416 1,350 1,365 1,930 Number of municipality_id 234 125 143 224 Adjusted R-squared 0.099 0.119 0.145 0.191 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 266 municipalities, but some regressions may include a smaller number due to missing data. The decentralisation variable considers only the 91 municipalities that started experiencing effects at the beginning of 2009. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 72 Table 28: Effects of the 1st reform in municipal accounts - Time Period 2004 - 2019 VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year>=2010) 63.81 131.1 17.40 108.8** 200.7*** 66.32*** (0.696) (1.387) (0.310) (2.350) (5.780) (4.484) Observations 3,572 3,575 3,575 3,572 3,575 3,575 Number of municipality_id 275 275 275 275 275 275 Adjusted R-squared 0.070 0.072 0.062 0.112 0.107 0.057 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 275 municipalities and cover the entire period of analysis. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 29: Effects of the 1st reform on educational outcomes - Time Period 2004 - 2019 VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic Decentralised(Year>=2010) -0.0224 -0.0538 0.236 0.869 2.386* (-0.141) (-0.149) (0.576) (0.737) (1.796) Log(Population)t−1-4.166*** -7.440** -2.131 23.79** -16.80 (-2.776) (-2.545) (-0.701) (2.038) (-1.239) Log(Month.Earnings)t−11.521 -0.491 1.645 -4.811 5.154 (1.483) (-0.197) (0.737) (-0.810) (0.569) %Unemploy.t−10.00756 0.0695 -0.0712 -0.749*** -0.0968 (0.191) (0.863) (-0.735) (-2.682) (-0.332) %Popula.Higher.Educ.t−10.0304 -0.108 0.253*** -0.467 -0.0462 (0.872) (-1.301) (2.640) (-1.087) (-0.0927) Observations 4,070 4,043 4,107 4,125 4,125 Number of municipality_id 275 275 275 275 275 Adjested R-squared 0.228 0.398 0.536 0.253 0.363 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 275 municipalities and cover the entire period of analysis. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 30: Effects of the 1st reform on educational outcomes - Time Period 2004 - 2019 (cont.) VARIABLES Public School Enrolment Rates (level of education) Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Decentralised(Year>=2010) -1.636* -0.712 0.0601 0.411 (-1.657) (-1.359) (0.0460) (0.429) Log(Population)t−115.20 -3.365 -13.92 -0.460 (1.644) (-0.509) (-1.460) (-0.0528) Log(Month.Earnings)t−1-10.42** -4.635 -4.611 -16.04 (-2.211) (-1.543) (-0.804) (-1.472) %Unemploy.t−10.0985 0.122 -0.297 -0.0959 (0.475) (0.911) (-1.025) (-0.338) %Popula.Higher.Educ.t−10.204 -0.251** -0.197 -0.146 (0.797) (-2.113) (-0.963) (-0.660) Observations 3,724 1,495 1,525 2,121 Number of municipality_id 255 138 156 243 Adjusted R-squared 0.111 0.105 0.127 0.192 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations cover the entire period of analysis and encompass 275 municipalities, but some regressions may include a smaller number due to missing data. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 73 Table 31: Effects of the 2nd reform in municipal accounts - Including 1st reform effects VARIABLES Expenditures (per student) Compensations Received (per student) (level of education) Pre-Primary 1st Cycle (Basic) Total Pre-Primary 1st Cycle (Basic) Total Decentralised(Year>=2010) 68.54 128.5 17.07 107.4** 196.5*** 64.72*** (0.748) (1.356) (0.303) (2.330) (5.618) (4.378) Decentralised(Year>=2016) -246.7** 133.6 17.67 72.30 219.1 83.53* (-2.141) (1.093) (0.403) (0.440) (1.625) (1.650) Observations 3,572 3,575 3,575 3,572 3,575 3,575 Number of municipality_id 275 275 275 275 275 275 Adjusted R-squared 0.071 0.073 0.061 0.112 0.112 0.063 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. All dependent variables are in real euros (at 2022 prices) per student. The estimations encompass 275 municipalities and include two dummy variables to represent the effects of the two decentralisation moments. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 32: Effects of the 2nd reform on educational outcomes - Including 1st reform effects VARIABLES Retention Rates Transition Rate Average Exam Classifications Schooling Rate (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Secondary 3rd Cycle (Basic) Secondary Pre-Primary Decentralised(Year>=2010) -0.0194 -0.0573 0.234 -0.354 -0.00873 -0.939 0.843 (-0.123) (-0.159) (0.572) (-0.652) (-0.547) (-1.440) (0.710) Decentralised(Year>=2016) -0.188 0.321 0.138 -0.202 -0.0221 -1.978*** 1.685 (-0.717) (0.925) (0.224) (-0.279) (-1.032) (-2.814) (0.686) Log(Population)t−1-4.125*** -7.509** -2.161 -5.811 -0.193 -8.724 23.41** (-2.736) (-2.564) (-0.706) (-1.235) (-1.311) (-1.202) (2.003) Log(Month.Earnings)t−11.497 -0.453 1.661 2.639 0.232*** 8.604 -4.611 (1.457) (-0.180) (0.744) (1.001) (2.596) (1.577) (-0.772) %Unemploy.t−10.00664 0.0711 -0.0706 -0.103 -0.00509 -0.183 -0.741*** (0.168) (0.882) (-0.730) (-0.772) (-1.300) (-1.104) (-2.642) %Popula.Higher.Educ.t−10.0316 -0.110 0.252*** -0.273* -0.000689 -0.534*** -0.477 (0.906) (-1.322) (2.626) (-1.841) (-0.158) (-3.219) (-1.120) Observations 4,070 4,043 4,107 3,763 3,281 1,994 4,125 Number of municipality_id 275 275 275 260 274 217 275 Adjusted R-squared 0.228 0.398 0.536 0.600 0.582 0.588 0.253 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 275 municipalities, but some regressions may include a smaller number due to missing data. Two dummy variables are included to represent the effects of the two decentralisation moments. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. Table 33: Effects of the 2nd reform on educational outcomes - Including 1st reform effects (cont.) VARIABLES Schooling Rates Public School Enrolment Rates (level of education) Basic Secondary Pre-Primary 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Secondary Decentralised(Year>=2010) 2.363* 0.189 -1.685* -0.679 0.0454 0.412 -0.883 (1.783) (0.0773) (-1.711) (-1.328) (0.0346) (0.427) (-0.883) Decentralised(Year>=2016) 1.475 10.93* 2.726** 1.958* 1.683 -0.0349 0.848 (0.437) (1.670) (2.527) (1.657) (0.702) (-0.0177) (0.302) Log(Population)t−1-17.13 -28.82 14.57 -4.822 -15.39* -0.437 -23.53** (-1.277) (-1.030) (1.576) (-0.763) (-1.712) (-0.0505) (-2.194) Log(Month.Earnings)t−15.328 19.56 -10.04** -3.517 -3.676 -16.06 -15.42 (0.585) (1.152) (-2.138) (-1.196) (-0.625) (-1.450) (-1.269) %Unemploy.t−1-0.0904 -0.435 0.110 0.128 -0.288 -0.0962 0.397 (-0.309) (-0.723) (0.531) (0.959) (-0.998) (-0.340) (1.395) %Popula.Higher.Educ.t−1-0.0556 1.239 0.188 -0.261** -0.205 -0.146 -0.246 (-0.112) (0.977) (0.749) (-2.172) (-1.003) (-0.662) (-0.946) Observations 4,125 3,916 3,724 1,495 1,525 2,121 2,267 Number of municipality_id 275 272 255 138 156 243 242 Adjusted R-squared 0.363 0.310 0.114 0.121 0.129 0.192 0.076 Notes: All regressions include municipal and year-fixed effects, and regional-specific trends. The estimations encompass 275 municipalities, but some regressions may include a smaller number due to missing data. Two dummy variables are included to represent the effects of the two decentralisation moments. T-statistics, based on robust standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 74 Appendix E Table 34: Effects of the 1st reform on educational outcomes - 2009 Group VARIABLES Retention Rates Schooling Rates (level of education) 1st Cycle (Basic) 2nd Cycle (Basic) 3rd Cycle (Basic) Pre-Primary Basic t_2005_2006 -0.339 -0.0113 -0.0745 1.736 0.788 (-0.923) (-0.0176) (-0.0747) (1.261) (0.963) t_2006_2007 -0.366 0.217 -0.210 0.839 -0.811 (-1.128) (0.367) (-0.209) (0.724) (-0.925) t_2007_2008 -0.223 -0.716 -0.852 -1.501 -0.606 (-0.644) (-1.149) (-1.161) (-1.191) (-0.435) t_2008_2009 -0.159 -0.291 0.438 1.610 -1.371 (-0.508) (-0.581) (0.674) (1.254) (-0.694) t_2008_2010 0.147 -0.271 -1.622** 0.985 1.724 (-0.713) (-2.145) (0.981) (0.868) (0.323) t_2008_2011 -0.138 -1.627** -0.164 1.394 1.922 (-0.331) (-2.040) (-0.156) (0.537) (0.549) t_2008_2012 -0.486 -0.477 -0.744 5.875 1.528 (-0.957) (-0.454) (-0.598) (1.411) (0.388) t_2008_2013 -0.370 -3.043** 1.294 6.346 2.386 (-0.560) (-2.187) (0.771) (1.236) (0.494) t_2008_2014 0.0619 0.0715 -0.506 -0.487 2.148 (0.121) (-0.382) (-0.331) (0.482) (0.908) t_2008_2015 -0.371 -2.255** 0.370 1.550 3.877 (-0.627) (-1.984) (0.287) (0.397) (0.751) Observations 3,025 3,025 3,025 3,025 3,025 Number of municipality_id 275 275 275 275 275 Notes: The estimations encompass 275 municipalities, excluding the three municipalities that signed contracts after 2011, and cover the 2004-2015 period. The control group considers all municipalities that did not sign a contract in 2009 or 2010. All regressions include a vector of control variables and regional-specific trends. Z-statistics, based on standard errors clustered by each municipality, are depicted in parentheses. Significance level: *** p<0.01, ** p<0.05, * p.<0.10. 75