The Tower of Babel in the Classroom? Immigrants and Natives in Italian Schools Rosario M. Ballatore 1 Margherita Fort 2 Andrea Ichino 3 1 Bank of Italy University of Bologna, CESifo and IZA 3 EUI & University of Bologna, CEPR, CESifo and IZA 2 Preliminary May 2014 A. Ichino (EUI) The Tower of Babel May 17th 2014 1 / 28 Overview What are we looking for? We are specifically interested in the causal effect of the number of immigrants in a class on the school performance of natives keeping class size constant e.g. adding one immigrant to a class and reducing by one the number of natives Since class size and class composition are chosen jointly by principals, this should be the parameter of interest for policy makers It has been neglected in the literature We call this parameter the Pure Composition Effect (PCE) A. Ichino (EUI) The Tower of Babel May 17th 2014 2 / 28 Overview This Paper We extend Lazear’s model (2001) of the educational production function to the presence of two types of students (natives & immigrants) allowing for positive externalities in the classroom We use data on Italian primary schools and standardised test scores Identification of the PCE is achieved exploiting discontinuities generated by institutional rules of class formation (Angrist & Lavy, 1999; Angrist & Lang, 2004) We find that the PCE is -1% in language and mathematics test scores, for students in grade 2 (age 7) or grade 5 (age 10) A. Ichino (EUI) The Tower of Babel May 17th 2014 3 / 28 Theory A model to think about the problem Lazear’s educational production function (2001): Consider a class with C students P ∈ [0, 1] is the probability that a student does not ask the teacher for specific attention at the expenses of other students. V̄ is the max performance if P = 1 (no disruption) The actual performance is V = V̄ P C < V̄ The principal maximizes MaxV(C ) : PC − W C At the optimum C ∗ = f (P , W ) + + Optimal class size is a function of students’ characteristics A. Ichino (EUI) The Tower of Babel May 17th 2014 4 / 28 Theory Extensions C = Natives (N ) + Immigrants (I ) Each group of students has its own disruption probability : Pn 6= Pi 6= P φ ∈ R\{0}; e.g. positive externalities are possible The extended educational production function is: φ N φih I Vh = V̄h Pn nh Pi where h ∈ [n, i ] A. Ichino (EUI) The Tower of Babel May 17th 2014 5 / 28 Theory Group and composition effects Using small letters for variables in logs: vh = v̄h + pn φnh N + pi φih I The structural effects of changing class size when natives (immigrants) change keeping immigrants (natives) constant are respectively: βh = ∂vh ∂N = pn φnh γh = ∂vh ∂I = pi φih From these group specific effects we can derive the : Pure Composition Effect (PCE) = δh = A. Ichino (EUI) The Tower of Babel ∂vh ∂I = γh − β h C =c May 17th 2014 6 / 28 Theory Why the PCE is important and how it can be identified Principals jointly choose class size and class composition to reduce the negative externalities coming from disruption This choice is generally based on priors concerning disruption, not on facts Reliable estimates of the PCE would improve this joint choice: the PCE is the relevant policy parameter Existing literature does not realise the importance of separating class size changes from class composition changes We exploit institutional features of class formation in Italy to identify the effect of composition when class size changes for exogenous reasons → Angrist and Lavy (1999), Angrist and Lang (2004) A. Ichino (EUI) The Tower of Babel May 17th 2014 7 / 28 Theory Identification in the literature Across schools: aggregation at higher than school level (Card et al. 2007, Brunello et al. 2011, Jensen and Rasmussen 2011) Within school: variation across classes (Ammermuller et al. 2009, Contini 2011, Ohinata & al 2011, Bossavie 2011) Within school: variation across cohorts (Hoxby 2000, Gould et al. 2009, Tonello 2012, Geay 2013) No evidence on the Pure Composition Effect A. Ichino (EUI) The Tower of Babel May 17th 2014 8 / 28 Institutional framework Italian institutional framework Students can pre-enroll in February for the following academic year Schools decide the number of classes (and hence class size) on the basis of pre-enrollment Rules for class formation (primary schools): No less than 10 students per class No more than 25 students per class A. Ichino (EUI) The Tower of Babel May 17th 2014 9 / 28 Institutional framework Actual &Theoretical Class Size in Italy ( as in Angrist and Lavy, 1999) Grade 2 & Grade 5 15 Class Size 20 25 Language Sample 10 Predicted Actual (avg.) 0 50 100 Enrollment in Building 150 Class size is a discontinuous function of school enrollment by grade A. Ichino (EUI) The Tower of Babel May 17th 2014 10 / 28 Institutional framework Space for immigrants (as in the METCO approach of Angrist and Lang, 2004) Principals set the average class size considering the number of pre-enrolled natives Due to higher geographic mobility and lack of information, immigrants typically do not pre-enroll Within school networks (under the same principal), immigrants are directed towards schools where the predicted number of natives per class is lower: Instructions by the Ministry of Education Rules of class formation create exogenous variability in the average number of natives per class and therefore in “space for immigrants” We use this exogenous variability in the number of natives to identify β, γ (and hence the PCE δ) A. Ichino (EUI) The Tower of Babel May 17th 2014 11 / 28 Institutional framework Space for immigrants based on predicted # of natives 10 15 20 Expected Avg. Number of Natives in Class 25 10 15 20 Expected Number of Natives in Class 25 d) Natives enrollment ∈ [76 − 100] Relative Space for Immigrant in Class, net of controls -.1 0 .1 .2 .3 .4 Relative Space for Immigrant in Class, net of controls -.5 -.4 -.3 -.2 -.1 c) Natives enrollment ∈ [51 − 75] 16 A. Ichino (EUI) b)Natives enrollment ∈ [26 − 50] Relative Space for Immigrant in Class, net of controls -.4 -.2 0 .2 Relative Space for Immigrant in Class, net of controls -.4 -.2 0 .2 Natives enrollment∈ [10 − 25] 18 20 22 Expected Number of Natives in Class 24 19 The Tower of Babel 20 21 22 Expected Number of Natives in Class 23 24 May 17th 2014 12 / 28 Institutional framework Space for immigrants based on predicted # of natives Relative Space for Immigrant in Class, net of controls -.4 -.2 0 .2 .4 a)Natives enrollment∈ [10 − 60] 10 A. Ichino (EUI) 20 30 40 Enrolled Number of Natives The Tower of Babel 50 60 May 17th 2014 13 / 28 Data & estimation Data We use the INVALSI dataset for the universe of Italian primary schools in the academic year 2009-10. For each student in grades 2 and 5 the data set contains information on: network, school, grade and class identifiers class size and class composition at the beginning of the year immigrant status based on citizenship test scores in language and math individual and family background information We focus on school networks with immigrants and more than one school (80% of the students in grade 2 and 5 in 2010 are enrolled in these school networks) A. Ichino (EUI) The Tower of Babel May 17th 2014 14 / 28 Data & estimation Descriptive statistics Mean S.D. p10 p25 p50 p75 p90 Language sample: Class size 19 4.1 13 16 19 22 24 Number of natives 17 4.2 11 14 17 20 22 Number of immigrants 2 2.1 0 .54 1.5 3 5 Enrollment 34 22 13 18 28 44 64 Test score language, Natives 0.67 0.11 0.54 0.60 0.67 0.73 0.80 Test score language, Immigrants 0.54 0.19 0.31 0.42 0.53 0.65 0.81 Test score math, Natives 0.64 0.19 0.39 0.5 0.64 0.78 0.89 Test score math, Immigants 0.55 0.19 0.32 0.41 0.54 0.68 0.82 655,132 students (584,025 natives and 71,107 immigrants) 4,333 networks and 11,955 schools 38,537 classes Descriptives are similar for the math sample A. Ichino (EUI) The Tower of Babel May 17th 2014 15 / 28 Data & estimation Estimation The reduced form educational production function of a representative native is: 0 vikgj = αg + βNkgj + γIkgj + Xikgj λ + eikgj (1) where the observation is a student i in class k of grade g ∈ [2, 5] in school j: β is the effect of one additional native on test scores for natives γ is the effect of one additional immigrant on test scores for natives 0 Xikgj is a vector of individual and class level controls The dependent variable is the log test score of a native student An equivalent equation for an immigrant student can be considered A. Ichino (EUI) The Tower of Babel May 17th 2014 16 / 28 Results Conventional evidence (within school strategy; pooled across grades) Number of natives: β̂ Number of immigrants: γ̂ Pure Composition Effect: δ̂ Observations A. Ichino (EUI) Language Mathematics 0.003*** 0.003*** (0.001) (0.001) -0.001 -0.001 (0.001) (0.001) -0.004*** -0.004*** (0.001) (0.001) 584,025 556,292 School-grade FE YES YES Individual specific controls YES YES Class level controls YES YES Mean score 0.67 0.71 (S.D.) 0.11 0.10 (S.D. Log score) 0.32 0.25 The Tower of Babel May 17th 2014 17 / 28 Results Conventional evidence (within school strategy; language; separate by grade) N. natives: β̂ N. immigrants: γ̂ (Pure) Composition: δ̂ Observations A. Ichino (EUI) 2nd grade 5th grade 0.004*** 0.003*** (0.001) (0.000) -0.000 -0.001 (0.003) (0.001) -0.004*** -0.004*** (0.001) (0.000) 292,166 291,856 School FE YES YES Individual specific controls YES YES Class level controls YES YES Mean score 0.67 0.71 (S.D.) 0.22 0.16 (S.D. Log score) 0.62 0.44 The Tower of Babel May 17th 2014 18 / 28 Results Conventional evidence (within school strategy; math; separate by grade) N. natives: β̂ N. immigrants: γ̂ (Pure) Composition: δ̂ Observations A. Ichino (EUI) 2nd grade 5th grade 0.004*** 0.003*** (0.000) (0.000) -0.000 -0.001 (0.002) (0.001) -0.004*** -0.004*** (0.001) (0.001) 277,420 278,870 School FE YES YES Individual specific controls YES YES Class level controls YES YES Mean score 0.67 0.71 (S.D.) 0.20 0.16 (S.D. Log score) 0.59 0.43 The Tower of Babel May 17th 2014 19 / 28 Results IV strategy the METCO approach, Angrist and Lang (2004) Let Z be the predicted class size based on natives school enrolment (Ejg ) for each school and grade: Zjg = Int Ejg Ejg −1 +1 25 (2) As in Angrist and Lang (2004) we derive our set of instruments by constructing indicator variables from Zjg min Γ ∈ [1(Zjq ≤ Zjg < Zjqmin + 1), ...., 1(Zjqmax ≤ Zjg < Zjqmax + 1)] (3) e.g. 1(1 ≤ Zjg < 2),....,1(25 ≤ Zjg < 26) We control also for school networks fixed effects A. Ichino (EUI) The Tower of Babel May 17th 2014 20 / 28 Results New “Metco type” evidence N. Natives β̂ (pooled across grades) Language Mathematics -0.0018*** -0.0020*** (0.001) (0.001) N. Immigrants γ̂ -0.0144*** -0.0156*** (0.004) (0.004) (Pure) Composition δ̂ -0.0126*** -0.0137*** (0.004) (0.004) Observations 584,023 556,290 Network of schools FE YES YES Grade FE YES YES Individual specific controls YES YES Class level controls YES YES F-test excluded instruments Natives 464.1 440.6 Immigrants 309.7 324.7 A. Ichino (EUI) The Tower of Babel May 17th 2014 21 / 28 Results New “Metco type” evidence (language; separate by grade) N. Natives β̂ 2nd Grade 5th Grade -0.0021** -0.0020*** (0.001) (0.001) N. Immigrants γ̂ -0.0114* -0.0178*** (0.007) (0.005) (Pure) Composition δ̂ -0.0093 -0.0158*** (0.006) (0.004) Observations 292,166 291,856 Network of schools FE YES YES Individual specific controls YES YES Class level controls YES YES Natives 232.4 230.4 Immigrants 45.5 159.3 F-test excluded instruments A. Ichino (EUI) The Tower of Babel May 17th 2014 22 / 28 Results New “Metco type” evidence (math; separate by grade) N. Natives β̂ 2nd Grade 5th Grade -0.0025** -0.0025*** (0.001) (0.001) N. Immigrants γ̂ -0.0132** -0.0196*** (0.007) (0.005) (Pure) Composition δ̂ -0.0107* -0.0171*** (0.006) (0.004) Observations 277,420 278,870 Network of schools FE YES YES Individual specific controls YES YES Class level controls YES YES Natives 233.7 227.3 Immigrants 53.3 159.1 F-test excluded instruments A. Ichino (EUI) The Tower of Babel May 17th 2014 23 / 28 Results Concluding remarks We propose: a an extension of Lazear’s educational production function to emphasize that the effect of immigrants on natives is confounded by adjustments in class size a strategy to identify the pure class composition effect, controlling for class size changes, that exploits rules of class formation Preliminary results: Modest composition effect (≈ −1%) in both language & mathematics - for students enrolled in the 2nd (age 7) and 5th grade (age 10) - the size of the effect is about 15% of the effect of poor parental background IV estimates larger than OLS because in the latter the PCE is confounded by class adjustments A. Ichino (EUI) The Tower of Babel May 17th 2014 24 / 28 Results Thank You It is preliminary Questions and comments are welcome You can contact us at rosariomaria.ballatore@bancaditalia.it margherita.fort@unibo.it andrea.ichino@unibo.it A. Ichino (EUI) The Tower of Babel May 17th 2014 25 / 28 Results Actual and Theoretical Class Size in Italy 2nd Grade 5th Grade Grade 5 Language Sample 15 15 Class Size 20 Class Size 20 25 25 Grade 2 Language Sample 0 50 100 Enrollment in Building Predicted Actual (avg.) 10 10 Predicted Actual (avg.) 150 0 50 100 Enrollment in Building 150 Very similar pattern in grade 2 and 5 Back A. Ichino (EUI) The Tower of Babel May 17th 2014 26 / 28 Results Rules concerning space for immigrants Ministero dell’Istruzione, dell’Università e della Ricerca Dipartimento per l’Istruzione Direzione Generale per gli Ordinamenti del Sistema Nazionale di Istruzione e per l’Autonomia Scolastica - Ufficio Sesto Circolare n. 4 MIURA00DGOS prot. n. 381/R.U.U Roma, 15 gennaio 2009 ... 10.2 Accordi di rete e intese territoriali È opportuno che le istituzioni scolastiche, al fine di evitare la concentrazione in talune scuole di iscrizioni di alunni con cittadinanza non italiana e conseguenti squilibri e disagi della popolazione scolastica, realizzino accordi di rete per una razionale distribuzione terri- toriale delle domande, procedendo, quindi, ad un’equa assegnazione degli alunni alle diverse classi (cfr DPR 31 agosto 1999 n. 394, art. 45, commi 3 e 5). I Direttori Generali degli Uffici scolastici regionali avranno cura di promuovere le iniziative ritenute più opportune al fine di attivare gli accordi di rete. Le scuole, possibilmente con azioni in rete, vorranno, pertanto, sollecitare o assecondare attivamente le iniziative degli Enti locali e/o di altri soggetti a livelli istituzionali per l’adozione di misure di prevenzione, orientamento e controllo circa l’assolvimento dell’obbligo di istruzione. Nelle città e nei grandi centri urbani in cui sono presenti ampie reti di scuole, le iscrizioni di alunni con cittadinanza non italiana anno gestite in maniera partecipata e programmata, in modo che la domanda e l’offerta di servizi scolastici risultano equamente distribuite. ... Back A. Ichino (EUI) The Tower of Babel May 17th 2014 27 / 28 Results First stage Relative Space for Immigrant in Class, net of controls 0 .1 -.2 -.1 .2 Expected Relative Space For Immigrants in Class based on expected # of natives in class & rules for class splitting 10 15 20 Expected Number of Natives in Class 25 Back A. Ichino (EUI) The Tower of Babel May 17th 2014 28 / 28