双语教学和收入：教育语言改革外文文献

Contents lists available at ScienceDirect Economics of Education Review journal homepage Bilingual schooling and earnings Evidence from a language in education reform Lorenzo Cappellaria Antonio Di Paolob aDepartment of Economics and Finance Universit Cattolica Milano Largo Gemelli 1 20123 Milano Italy bAQR IREA University of Barcelona Avinguda Diagonal 690 08034 Barcelona Spain A R T I C L E I N F O Keywords Bilingual education Returns to schooling Language in education reform Catalonia JEL classifi cations J24 J31 I28 A B S T R A C T We estimate the wage eff ects of bilingual education for the fi rst time using a reform that introduced bilingualism in Catalan schools Variation across years of schooling and birth cohorts provides identifi cation We fi nd sub stantial eff ects of bilingual education which increases baseline returns to education by about 20 Robustness checks show that eff ects stem from exposure to the language in education reform and are neither a consequence of unobservable determinants of educational attainment nor an artifact of education cohort specifi c trends in wages The reform was mostly benefi cial for individuals of non Catalan background from low parental back ground thus achieving its goal of leveling the playing fi eld in a bilingual society 1 Introduction Language of instruction is a key input in the educational production function While most students still receive education only in their native language bilingual education is becoming increasingly popular among families and is attracting growing attention from educationalists Bilingualism aff ects the quality of education favors the development of cognitive skills and its eff ectiveness in fostering educational outcomes has been found to increase with early exposure Bleakley Costa Clots Figueras Accepted 20 March 2018 Funding from C tedra Pasqual Maragall at Universitat de Barcelona the MEC grant no ECO2013 41022 R the PRIN grant no 2010T8XAXB and the RecerCaixa grant no 2016ACUP00020 is gratefully acknowledged Thanks are due to Frederic Udina and Josep Maria Mart nez at IDESCAT for their assistance with the data Ada Ferrer i Carbonell Ramon Caminal Albert Costa Xavier Vila i Moreno Germ Bel Joan Ramon Borrell and audiences at EUR URV UdG UB IRVAPP Trento SFI Copenhagen IDESCAT EALE SOLE ESPE IWAEE RES SAEe and the Lisbon Workshop on Economics and Econometrics of Education provided useful comments The usual disclaimers apply Corresponding author E mail addresses lorenzo cappellari unicatt it L Cappellari antonio dipaolo ub edu A Di Paolo 1See Aspachs Bracons et al 2008 and Clots Figueras and Masella 2013 Economics of Education Review 64 2018 90 101 Available online 03 April 2018 0272 7757 2018 Elsevier Ltd All rights reserved T Catalan and Spanish are romantic languages but Catalan is closer to English compared with Spanish 2The return of democracy in the late 1970s provided the political framework for a reintroduction of Catalan also as a means of promoting the political autonomy of the region and its internal social cohesion Increasing Catalan profi ciency in the po pulation became an important target for the new regional government because a substantial share of the population was non Catalan after the internal migration fl ows of the 1950s and 1960s especially in the more industrialized areas of the region More in general after forty years of prohibition profi ciency in written Catalan was problematic among prime age and young individuals irrespective of regional origins Cat alan became a compulsory subject in schools immediately after the end of dictatorship but it was only with the 1983 law that a broader reform of the educational system towards bilingualism was put in place The Language Normalization Act LNA of 1983 established Catalan as a medium of instruction in primary and compulsory secondary schools alongside Spanish making the education system eff ectively bilingual This language in education reform aff ected virtually all stu dents enrolled at the time of implementation irrespective of their birth cohort or school of enrolment The comprehensiveness of the reform rules out any possibility of students selection into bilingual schools which naturally occurs whenever two systems monolingual and bi lingual co exist and parents are free to choose the language of edu cation of their children as for example in the Basque Country see Aspachs Bracons et al 2008 From a theoretical point of view there are two main channels through which exposure to bilingualism at school might be rewarded in the labor market First bilingualism can favor the development of cognitive skills in this way increasing the stock of human capital ac cumulated per year of education resulting in higher wage returns Related to this it may increase the quality of education if more skilled teachers are needed for bilingual teaching inducing higher returns to schooling Second bilingual knowledge might increase search effi ciency and increase the rate of arrival of job off ers in a local labor market in which Catalan speaking employers represent a relevant share of overall labor demand leading to better job matches and higher wages Eff ects through better search might also derive from better ac cess to Catalan social networks which could be particularly relevant for individualswithnon Catalanoriginsthatincreasinglyperceived themselves as Catalans after the reform Aspachs Bracons et al 2008 Clots Figueras Artigal 1997 These were targeted to schools whose students predominantly more than 70 came from Spanish speaking families which tended to be settled in neighborhoods where Catalan had very little presence Schools in immersion programmes used Catalan as the only language of instruction during the fi rst years of education and followed a specifi c methodology to stimulate second language L2 i e Catalan acquisi tion Spanish was introduced as additional language of instruction only at a later stage normally grade 3 The LNA also regulated the language regime in post compulsory secondary school However the introduction of Catalan as medium of instruction in upper secondary education was less intense and more gradual compared to compulsory education The reform established that upper secondary schools besides mandatory language courses in Spanish and Catalan had to employ Catalan as medium of instruction in at least two subjects In practice however the choice between Catalan and Spanish was left to teachers and taken on the basis of the linguistic composition in the classroom The language regime of universities was not explicitly regulated by the LNA as universities were already endowed with special autonomy when the language policy was implemented The law only established the right to use any of the two offi cial languages and required Catalan universities to off er Catalan courses to students and teachers with limited knowledge of Catalan 3 Although with varying intensity all the levels of the educational system became bilingual in the year of the reform This means that while cohorts starting primary school after 1983 were fully aff ected by bilingual education cohorts already in school in 1983 were diff eren tially aff ected by bilingual schooling depending upon the grade at tended in that year This creates a partial exposure framework similar to the one exploited in Angrist et al 2008 for the case of Puerto Rico We come back in Section 4 on the practical implementation of the estimator based on partial exposure TheLNApromotedbilingualismamongteaching staff Arenas 1990 Several specialized teachers were hired to guarantee the minimum level of staff profi ciency required by the law These new teachers were allocated on the basis of the linguistic composition of the school in terms of both teachers and students At the same time less Catalan profi cient teachers were given the opportunity to develop adequate skills through special courses promoted by the SEDEC Servei d Ensenyament del Catal Catalan Teaching Service and the Institutes of Education of the three Catalan public universities of the time While in 1978 1979 only 52 of teachers in Catalonia were able to speak and write in Catalan this percentage rose to 87 in 1986 1987 Arenas 1977 19651977 01965 i ii iii i 1 where eiare years of potential exposure of person i siare years of schooling and biis the year of birth The treatment variable eimay diff er from actual exposure because of grade retention and diff erential intensity in the use of Catalan as medium of instruction across schools Therefore our treatment variable can be interpreted as capturing an Intention to Treat ITT eff ect 5 1 Baseline model Following Angrist et al 2008 our baseline specifi cation for esti mating the eff ect of the reform on labor market outcomes is w x s e iiiibi 2 where widenotes labor market outcomes xiis a vector of controls year gender potential experience and its square and bis a year of birth fi xed eff ect Our main outcome of interest is monthly net earnings from employment but we also consider other outcomes such as employment hours and occupation 6 The coeffi cient is the additional return per year of education generated by the reform and parameterizes a DiD estimator across birth years and schooling levels that allows for partial exposure Eq 2 provides consistent estimates of the eff ects of the reform as long as any other trends in labor market outcomes be sides the ones induced by the reform were common across birth years and schooling levels 7 5 2 Accounting for potential confounders Life cycle earnings trends may confound the eff ects of the reform because treated and non treated cohorts are observed at diff erent stages of their labor market trajectories a form of life cycle bias Life cycle eff ects are controlled for in our baseline model through year of birth fi xed eff ects Moreover we exploit the availability of two cross sections and further saturate the specifi cation with age dummies omitting two of these rather than one from the regression to avoid collinearity The use of two cross sections comes with a cost since the second one 2011 refers to a period of slack in the Spanish and Catalan economies which has substantially increased unemployment especially among the young The potential change in selection into employment might aff ect the estimates to check if this is an issue as a sensitivity we estimate a joint model for earnings and employment and correct for selection 8 Table 1 Imputed years of exposure Year of birthYears of schooling 3681112141517 20 1961 6500000000 196600001346 196700012457 196800023568 196900034679 1970001457810 1971002568911 19720136791012 197302478101113 197403589111214 1975146910121315 19762571011131416 1977 823681112141517 4We run sensitivity checks excluding from the analysis self employed and en trepreneurs fi nding that results were robust to this sample selection 5Alternatively we discretized 2011 earnings into the same intervals that categorize earnings in 2006 and applied interval regressions Results were pretty much unaff ected by changing the harmonization procedure 6Besides providing evidence on labor market outcomes other than earnings looking at the eff ect of the reform on working hours is particularly relevant in our case because 2006 earnings are reported in brackets preventing us from deriving hourly earnings measures 7Eq 2 imposes linearity of the return to bilingual schooling Alternatively we al lowed for non linearities through a specifi cation with dummies for years of exposure or through a spline function Results from these models do not indicate any signifi cant variation of the eff ect by years of exposure 8We estimate an endogenous selection model using as exclusion restriction the un employment rate at the province level when the individual was 16 the legal minimum working age We also show that the estimated eff ects are not very diff erent if we estimate the earnings equation separately by year see Table A4 in the Appendix L Cappellari A Di PaoloEconomics of Education Review 64 2018 90 101 94 An additional confounding factor that we take into account is the contemporaneous expansion of education that took place in Spain over the period in which the LNA reform was implemented To account for it following Angrist et al 2008 we augment the specifi cation with a measure of the educational cumulative density function CDF and its square For each survey respondent we use data from the Spanish Census of 2011 and defi ne the educational CDF as the fraction of in dividuals with a lower level of education who were born in the same year of birth The CDF would capture the changing selectivity into schooling levels due to education expansion which might have reduced the average ability of individuals with higher schooling attainments 5 3 Placebo experiments We pay special attention to the existence of spurious relationships between the measure of exposure and the outcome due to education cohort specifi c trends The main assumption upon which validity of the identifi cation strategy rests is the absence of diff erential trends in labor market outcomes across years of birth and years of schooling We ad dress these concerns by means of two diff erent falsifi cation exercises First we run a placebo analysis that uses data on cohorts of individuals born in the same years as those in the baseline estimation sample 1961 1982 but not residing in Catalonia or in any other bilingual region of Spain We use the Spanish component of EU SILC waves 2006 and 2011 and impute reform exposure as if it had been implemented also in those regions Second we go back to ECHVP data and consider cohorts of Catalans born between 1945 and 1965 which were never actually exposed to bilingual education to whom we impute placebo exposure to bilingual schooling i pretending that the reform was rolled out in year t 1963 1970 rather than in 1983 thus generating partial placebo exposure for birth cohorts from 1946 to t 7 and full placebo ex posure for birth cohorts from t 6 to 1965 siftb maxstbifbt ifb 6 1966 0 6 1946 6 01946 i ii iii i 3 We estimate the baseline model on these cohorts using ias the treatment variable Since these cohorts were not exposed to the true reform obtaining a signifi cant eff ect of the placebo variable would suggest that our treatment variable 1 is capturing pre existing edu cation cohort specifi c trends in earnings rather than the impact of bi lingual schooling 5 4 Selection into education There is a fi nal major concern with the derivation of the treatment variable namely that it is constructed on the basis of completed schooling which is a choice variable 9This would not be an issue if the unobservables driving selection into education did not change after the implementation of the LNA reform Conversely if before the reform some individuals with intrinsically high ability were not enrolling into post compulsory education because school programmes were taught only in Spanish and the introduction of bilingualism induced their younger counterparts to continue education then we would be erro neously attributing to the reform what in eff ect is a change in the un observables In order to gauge the relevance of this potential issue we apply the strategy of Clots Figueras and Masella 2013 and defi ne the treatment as years of potential exposure at compulsory education e ifb maxbifb ifb 81977 0 8 1977 19701977 01970 i i ii i 4 As suggested by Clots Figueras and Masella 2013 using compul sory education means that the corresponding exposure variable only depends on year of birth which would free from any selection bias the estimated eff ect of exposure to bilingual schooling Following their intuition as an alternative to our baseline specifi cation we estimate the model substituting total years of exposure with years of exposure at compulsory school Since it varies only by year of birth and not also by years of schooling year of birth fi xed eff ects are not identifi ed in this model and we group them into birth cohort fi xed eff ects c yielding the following equation10 w x s e iiiici 5 Therefore if changing selection into schooling does not represent the main driver of the eff ect of language exposure on earnings the results from the baseline model 2 which exploits the potentially en dogenous variation of ei would be similar to those obtained from Eq 5 which exploits exogenous variation ofe iacross birth cohorts On the contrary fi nding lower or no eff ects of compulsory language exposure would suggest that changes in unobserved individual het erogeneity rather than language exposure generate the eff ects that we instead attribute to the policy change We also run a falsifi cation ex ercise for the compulsory exposure variable by imputinge ito in dividuals from other non bilingual Spanish regions again using the EU SILC data which would capture potential contemporaneous trends across the cohorts 11 6 Results 6 1 Bilingual schooling and labor market outcomes We begin the presentation of results in Table 2 reporting selected parameter estimates for the eff ects of the reform on various labor market outcomes 12Column 1 reports results of the earnings model The DiD estimates show a sizeable positive eff ect of bilingual education on earnings One additional year of bilingual education increases earnings by 1 4 This incremental return to schooling comes on top of a yearly return of 6 5 and represents a proportional increase of about 20 of returns to schooling due to exposure to bilingual education In Columns 2 4 of Table 2 we apply the same specifi cation of Column 1 to labor market outcomes other than earnings Column 2 focuses on employment In Column 3 we consider whether the eff ect observed on monthly earnings could stem from diff erences in hours of work between treated and non treated individuals and use weekly hours as the relevant outcome Such exercise is important because we are unable to construct an hourly wage measure consistently