Première grande recherche longitudinale sur les effets de l’éducation scolaire en Allemagne

Les parcours de formation de 60.000 allemands passés au crible

Norberto (12/02/2009)

"Panel sur l’éducation pour l’Allemagne" (National Educational Panel Study, NEPS)

Description :

Annonce du lancement en Allemagne de la première étude longitudinale sur vaste échelle financée par le Ministère fédérale de l’éducation et de la recherche sur l’évolution tout le long de l’existence des compétences acquises au sein du système d’enseignement.

Politique de la recherche

Recherche pédagogique : les parcours de formation de 60.000 allemands passés au crible

Le 3 février 2009, la Ministre fédérale de l’enseignement et de la recherche, Annette Schavan, a donné à Bamberg le coup d’envoi d’une étude à long terme en recherche pédagogique : pendant plusieurs années, 60.000 Allemands d’âges différents seront suivis pour comprendre le déroulement de leur parcours de formation.

Ce "Panel sur l’éducation pour l’Allemagne" (National Educational Panel Study, NEPS)a pour objectif de mesurer le développement des compétences dans le parcours de chacun des participants du panel et devrait fournir des réponses aux questions-clés de la politique de l’éducation. Ainsi, il sera pour la première fois possible d’observer comment des enfants de même origine et disposant des mêmes compétences se développent dans des institutions éducatives différentes, de déterminer quels facteurs provoquent des évolutions positives, à quels éléments l’échec est lié et comment des cas dits "à risque" peuvent être précocement détectés. "Cette étude sur le long terme va fournir à la recherche pédagogique des données de fond détaillées, qui vont nous aider dans l’application de nos approches pédagogiques", espère Annette Schavan. Il s’agit de l’un des plus grands programmes jamais menés en sciences sociales en Allemagne. D’ici 2014, les coûts du panel devraient s’élever à 60 millions d’euros. 150 chercheurs issus des meilleurs instituts allemands prennent part à cette étude.

Le panel est coordonné par un réseau de recherche dirigé par le Prof. Hans-Peter Blossfeld, sociologue à l’Université de Bamberg. 

Le Ministère fédéral de l’enseignement et de la recherche (BMBF) finance les travaux de recherche à hauteur de 7,5 millions d’euros en 2009. Les moyens mis à disposition de l’étude par le BMBF devraient continuellement croître d’ici 2013 pour atteindre finalement 16 millions d’euros. Le Land de Bavière et l’Université de Bamberg financent eux aussi le projet. L’étude suivra l’origine et les caractéristiques sociales des participants, de même que leurs capacités en lecture, en mathématiques et en sciences naturelles. Cette année, le suivi portera d’abord sur un premier échantillon de 13.000 adultes âgés de 23 à 64 ans. En 2010 suivront des enfants de quatre ans, des écoliers de niveau CM2/sixième et troisième/seconde, ainsi que des étudiants.

La particularité du panel est son ancrage sur le long terme, qui est comparable à un film documentaire. Les mêmes personnes seront interrogées et testées régulièrement sur de longues périodes, afin de comprendre comment les compétences s’épanouissent au cours de la vie, comment elles influencent les décisions au moment des transitions dans le parcours de formation et de quelle manière elles sont marquées par l’origine familiale et les institutions éducatives fréquentées. "C’est exactement là que se situe le progrès, par rapport, par exemple, aux études transversales internationales comme PISA  [1] et PIRLS [2]  qui, telles des photographies, ne peuvent fournir que des instantanés de la situation au moment de l’étude", commente Annette Schavan.

Pour en savoir plus, contacts :www.uni-bamberg.de/neps/

---

Les documents attachés à cet article :

I "Digi-Teachers"; gli insegnanti "digitali"

Uso delle nuove tecnologie dell’informazione da parte degli insegnanti

Norberto (23/06/2010)

Opinioni, problemi, resistenze e risultati degli insegnanti che usano le TIC a scuola

Description :

Resoconto di due relazioni sul tema "Teachers’ Technology Use: Beliefs, Practices, and Expertise" presentate al convegno dell’AREA 2010 a Denver, Colorado,il 3 aprile 2010

Uso delle tecnologie da parte degli insegnanti

 [1]

Le relazioni presentate in questa sessione del convegno dell’AERA 2010 riguardavano indagini svolte per conoscere i comportamenti, i pregiudizi, le opinioni e le pratiche degli insegnanti nei confronti delle TIC.

L’adozione delle TIC nelle scuole (per esempio le lavagne luminose) non è affatto una faccenda ovvia. Dove si è forzato il passaggio imponendo alle scuole attrezzature non richieste o il cui uso non era padroneggiato dagli insegnanti, è capitato di tutto, con sprechi immensi. Non basta spendere per attrezzare le scuole con nuovi apparecchi soltanto per brillare davanti agli elettori e gareggiare per essere etichettati come notabili progressisti che vogliono il bene della popolazione senza tirarsi in dietro quando si tratta di spendere per le scuole. 

Occorre anche saperle utilizzare le TIC o ICT (se ci si serve dell’ acronimo inglese) il che non è evidente, tenuto conto del gran numero di insegnanti nelle scuole formati e abituati a lavorare in un altro modo. Molti insegnanti sono scettici, rifiutano le ICT, sono prevenuti nei confronti delle TIC e le sabotano. Ma la scuola, se vuole restare a galla e sopravvivere non può fare altro che trovare una modalità per modificare i curricoli e adottare in modo efficace le nuove tecnologie.

Si tratta di una sfida impari, difficilissima, ma, come nel calcio, anche le piccole squadre talora riescono a mettere in difficoltà le grandi. I promotori della generalizzazione delle TIC nelle scuole primarie e secondarie sottovalutano in genere le reazioni degli insegnanti, non conoscono attraverso quali vie gli insegnanti si preparano a servirsi delle TIC, o le integrano nella loro pratica, o le rifiutano. Prima di lanciare riforme su vasta scala di generalizzazione delle TIC nelle scuole o di innovazione dell’insegnamento e dei curriculi mediante l’ adozione di nuove tecnologie è indispensabile conoscere le opinioni degli insegnanti, osservare come reagiscono quando hanno a che fare con le TIC nel loro lavoro.

In questa sessione sono state presentate due ricerche: una californiana e una inglese che meritavano attenzione non tanto per i risultati ma per i metodi d’indagine, come del resto ha sottolineato alla fine Natalie Milman , professore associato all’università George Washington (nmilman@gwu.edu ), che aveva l’incarico di commentare le due indagini.

Le due relazioni sono state le seguenti [2]:

Techno-Reform: The Intersection of Teacher Practice and Technology-Enhanced Curriculum Delivery and Assessment. A cura di Juna Z. Snow (University of California - Berkeley, jsnow@innovatedconsulting.com)

Digi-Teachers: Technology and Practice. A cura di Andrew C. Goodwyn (University of Reading - a.c.goodwyn@reading.ac.uk), Carol L. Fuller (University of Reading - c.l.fuller@reading.ac.uk), Aristidis Protopsaltis (University of Reading - aprotopsaltis@gmail.com)

Nella discussione che è seguita alla relazioni, Natalie Milman ha insistito sull’importanza degli strumenti da utilizzare per captare le pratiche e le opinioni degli insegnanti : ne occorrono molti e complementari tra loro.

Nell’indagine di Snow in California è particolarmente apprezzabile il fatto che si sono dedicati ben tre semestri all’osservazione dei comportamenti degli insegnanti alle prese con le TIC, al posto di liquidare le loro opinioni in quattro e quattr’otto con un’intervista stereotipata come succede molto spesso.

Occorre anche prestare attenzione alla storia dell’innovazione in una scuola e allo sviluppo professionale degli insegnanti. Questi fattori devono essere presi in considerazione anche perché variano da scuola a scuola e perché influenzano potentemente le pratiche degli insegnanti. Per i ricercatori occorre evitare di andare alla cieca, seguendo un proprio schema teorico che si riflette nei questionari o nelle domande delle interviste. In questo modo si passa accanto ai problemi, non si afferrano e si raccolgono soltanto informazioni poco rilevanti.

Gli insegnanti digitali (i "digi-teachers")

L’indagine pilotata da Goodwyn in Inghilterra ha il pregio di essersi concentrata sugli “insegnanti digitali” , ossia sugli insegnanti che hanno la passione per l’informatica. Quindi l’indagine è stata condotta su un gruppo specifico di insegnanti, quelli sui quali si fa affidamento per contagiare gli scettici e trascinare le scuole nel mondo delle nuove tecnologie.

Obiettivo della ricerca: capire le motivazioni degli insegnanti identificati come “ insegnanti fuori del comune” per la padronanza delle TIC in classe. La ricerca è stata condotta su insegnanti di scuola elementare. Il problema subito evidenziato nella discussione è stato quello dei criteri e della procedura per identificare i "digi-teachers".
 

Quel che conta non sono le attrezzature: queste sono necessarie ma non sufficienti

L’ipotesi di ricerca è da ritenere: il modo con il quale le ICT sono usate conta di più che non la quantità di attrezzature nelle scuole. Nelle scuole elementari inglesi, c`è una buona attrezzatura per quel che riguarda le nuove tecnologie (lavagne luminose, computer, WiFI, ADSL, tavolette numeriche. ecc) , ma l’ uso appropriato e efficace delle ICT è raro. Quindi non è l’attrezzatura che conta ma la competenza, la bravura, la professionalità degli insegnanti. Da qui l’interesse di andare a caccia dei bravi insegnanti che si servono delle ICT per vedere come se la cavano, quali sono secondo loro le difficoltà da superare per utilizzare al meglio le nuove tecnologie.

Quali sono le buone pratiche con le ICT?

Un altro problema metodologico rilevante è¨quello della definizione delle buone pratiche nell’uso delle ICT. L’osservazione e l’analisi delle pratiche degli “insegnanti digitali”, che fanno un buon uso delle ICT, sono un passaggio obbligato per impostare programmi di grande ampiezza e per costruire una scala delle competenze degli insegnanti da usare nelle riforme che si prefiggono di diffondere l’uso delle ICT nelle scuole.

Nell’indagine inglese sono state contattate 250 scuole. Gli insegnanti-digitali, ritenuti eccellenti sia per l’uso didattico delle ICT, sia per i risultati scolastici dei loro allievi, sono stati in tutto per per tutto 93.

Il 60% degli insegnanti che hanno partecipato all’indagine (54 , di cui 26 di scuola elementare e 28 di scuola media) ritiene che il mondo sociale e culturale degli studenti sia profondamente cambiato rispetto ad alcuni anni fa. I “digital natives” vogliono imparare in un altro modo e vogliono un insegnamento diverso.

Rappresentatività dei "digi-teachers" segnalati dai dirigenti

Il commento di Milman ha messo in evidenza alcuni punti critici: come sono state condotte le interviste con gli insegnanti digitali? I dirigenti che hanno segnalato i docenti esperti hanno ricevuto tutti la stessa griglia di analisi, gli stessi criteri di selezione, le stesse definizioni? Le osservazioni in classe fatte con video erano sufficientemente lunghe?

Questa sessione ha permesso di elencare una serie di questioni alle quali si deve prestare attenzione prima di partire lancia in resta con programmi ambiziosi, spettacolari, di distribuzione di computer nelle scuole e di generalizzazione di applicazioni informatiche per la didattica o per la gestione delle scuole, prima di spendere milioni di euro o di dollari per attrezzare le scuole con strumenti e tecnologie che o non sono utilizzati o sono utilizzati male.

 

---

Les documents attachés à cet article :

La ricerca scientifica sulla scuola a una svolta

"Capire la complessità di un mondo in mutazione e le ripercussioni sulla scuola".

Norberto (22/06/2010)

Convegno annuale dell’AERA, Denver (Colorado), 30 aprile 2010-4 maggio 2010: prima parte

Description :

Con quest’articolo inizia il resoconto del congresso annuale della Società Americana di ricerca sulla scuola(AERA) svoltosi a Denver, (Colorado) dal 30 aprile al 4 maggio 2010 sul tema "Capire la complessità di un mondo in mutazione e le ripercussioni sulla scuola". In questa prima parte si spiega la natura del convegno e si descrivono le ragioni che hanno indotto a scegliere il tema del programma del convegno.

"Capire la complessità di un mondo in mutazione e le ripercussioni sulla scuola"

(Introduzione)

 

Più di 12000 ricercatori nelle scienze dell’educazione provenienti dagli Stati Uniti e da una sessantina di paesi hanno partecipato all’incontro annuale della Società americana di ricerche in educazione (AERA, American Educational Research Association) che si è tenuto a Denver (Colorado) dal 30 aprile al 4 maggio scorsi.

Il programma della manifestazione con la lista di tutte le presentazioni (circa 8000) può essere consultato nel sito www.aera.net

In questa relazione saranno esposte alcune considerazioni sulla base di una scelta soggettiva delle sessioni e delle relazioni presentate nelle sessioni. I criteri di scelta riflettono interessi personali dell’autore di quest’articolo. ma anche del programma della fondazione San Paolo per la scuola, nonché temi e preoccupazioni che concernono la politica scolastica italiana. Questo resoconto ha quindi un taglio necessariamente soggettivo. E’ infatti impossibile per una sola persona farsi un’idea di tutto quanto bolle in pentola in un convegno di questo tipo, poiché è non si riesce a seguire l’integralità delle circa 2000 sessioni. 

Un convegno unico

il convegno annuo dell’AERA è un vero è proprio alveare nel quale succede di tutto ed è veramente difficile essere al corrente di quel che capita durante una giornata. Occorre peraltro una buona esperienza per identificare i momenti scottanti del programma e per non perdere le relazioni più significative dal punto di vista scientifico e politico.

Un incontro di levatura mondiale

ll convegno dell’AERA è uno degli incontri a livello mondiale più rilevanti sulle politiche scolastiche, sulle ricerche scientifiche sulla scuola e le politiche scolastiche e l’istruzione . Una gran parte della comunità scientifica mondiale si incontra in questa occasione per scambiare opinioni, critiche e informazioni sullo sviluppo delle principali ricerche e indagini.

La comunità scientifica invitata quest’anno dalla società americana di ricerche sulla scuola è stata quella cinese. Un paio di anni fa, il paese ospite è stata la Russia. Da questi due inviti si può capire subito il ruolo strategico di questo convegno al quale intervengono tutti i principali sindacati scolastici americani, nonché le autorità scolastiche federali. Per esempio quest’anno un ospite di riguardo è stato il nuovo direttore dell’Istituto federale per le scienze dell’educazione John Q. Easton, che ha presentato la missione che l’amministrazione Obama attribuisce alla ricerca scientifica sulla scuola  [1] . Il suo intervento si può trovare nel sito dell’IES oppure allegato a quest’articolo. 

Quest’anno le associazioni di ricerca sulla scuola di diversi paesi presenti all’incontro erano una decina:

l’associazione australiana , l’associazione inglese , l’associazione canadese, l’ associazione olandese , l’associazione irlandese , il forum fiammingo di ricerca sulla scuola, l’associazione neozelandese , il congresso internazionale sull’efficacia e il miglioramento della scuola.

 

Queste società colgono l’occasione di questo congresso per farsi conoscere e per presentare le indagini scientifiche sulla scuola svolte dai propri membri. Inoltre, il programma internazionale è l’occasione per instaurare forme di cooperazione transnazionale nel settore scolastico.

Le varie sessioni e conferenze del programma offrono l’opportunità di raccogliere indicazioni sulle ricerche scientifiche da svolgere per capire i problemi che ostacolano la scolarizzazione e il modo di funzionare dell’apparato scolastico e dei sistemi scolastici. Nessuna tesi riguardante l’andamento della scuola, nessun aspetto della vita scolastica, statale o privata o paritaria che sia, è trascurato. Tutto è passato al vaglio della ricerca scientifica. Nulla è dato per scontato. Da questo punto di vista si può affermare che quest’incontro è un gigantesco esame annuale di coscienza da un lato dalla comunità scientifica e dei suoi metodi di lavoro nonché un lavoro approfondito della comunità scientifica internazionale che si occupa di scuola.


Il convegno AERA 2010 a Denver, Colorado (Stati Uniti)

Il titolo dell’incontro di quest’anno era "Capire la complessità di un mondo in mutazione e le ripercussioni sulla scuola" [2].

Questo tema deriva dalla consapevolezza diffusa nella comunità scientifica che si occupa di scuola che i problemi curricolari e dell’istruzione non possono essere risolti con un solo metodo e nemmeno possono essere capiti con una ricerca scientifica isolata, per quanto eccellente possa essere. La complessità del mondo scolastico e della società impone alla ricerca scientifica sulla scuola di articolare numerosi parametri e di ricorrere a diversi paradigmi di simulazione e interpretazione. 

Nel mondo della ricerca scientifica sulla scuola succede esattamente quanto accade nella scuola dove non esiste un solo stile d’apprendimento che vada bene per tutti gli allievi. In ambito scientifico si è capito che ormai non basta più osservare gli studenti in un solo contesto per capire le strategie d’apprendimento.

Esiste dunque un forte parallelismo tra l’insegnamento e la ricerca. Nell’ insegnamento è illusorio ritenere che basta ricorrere a una soluzione semplice come quella di rifugiarsi dietro le porte dell’aula, chiusa in modo più o meno ermetico, sperando che ciò facendo si riesca a tenere sotto controllo tutti i fattori che incidono sull’apprendimento e realizzare un insegnamento di qualità. Il mondo esterno penetra da tutte le parti dentro le aule. Gli studenti lo portano con loro. Ciò vale anche per la ricerca scientifica. Una ricerca impostata in modo unilaterale, che riduce al massimo il ventaglio delle ipotesi di lavoro, che opera con un numero ristretto di variabili è poco credibile, non è adeguata all’oggetto di indagine.

Quando la ricerca scientifica si occupa della scuola non deve inoltre scordare che gli insegnanti operano e reagiscono in funzione di valori e criteri che riflettono le tendenze dominanti esistenti nella società piuttosto che i valori degli studenti. Questo vale anche per i ricercatori che hanno la tendenza ad impostare le indagine in funzione delle proprie tesi che sono altrettanto soggettive e pregiudiziali di quelle che compongono le opinioni degli insegnanti.

Analogamente a quanto si verifica nel mondo economico e politico, anche in quello scolastico operano molteplici parametri. La scolarizzazione e l’istruzione sono un fenomeno polivalente, poliedrico, influenzato da una vasta gamma di fattori. Le scelte possibili sono svariate: ci si può quindi facilmente sbagliare come si possono operare scelte giuste in modo del tutto inconsapevole o intuitivo. I criteri di scelta sono suggeriti dalla conoscenza che si acquisisce implicitamente nel mondo personale e familiare in cui si vive, nelle reti sociali e dei pari, nelle organizzazioni scolastiche comunitarie, nella vita associativa, nelle associazioni professionali e sindacali oppure tramite i media. La maggioranza dei parametri che si ritrovano in questa lista sono totalmente ignorati dalle riforme che mirano a migliorare le scuole anche se questi parametri concorrono a costituire il reticolo dell’ecologia sociale, interagiscono tra loro spesso in modo molto impercettibile. Le soluzioni semplicistiche che isolano un fattore per spiegare la realtà complessa della scuola sono illusorie e errate. I metodi di indagine devono evolvere e cambiare. strutturarsi come interfacce dove si articolano approcci multipli, teorie diverse, in uno scambio che non può non essere fecondo. Questo era il tema del congresso il quale invitava i partecipanti e la comunità scientifica che opera nel settore scolastico a diversificare , arricchire, mutare i metodi di lavoro, a rendersi conto che la realtà scolastica è poliedrica e non può essere ridotta a una sola dimensione.

Il concetto di "classe" oppure di "aula" si estende ben oltre i banchi, le lavagne, l’ora di lezione, le valutazioni, i compiti in classe e quelli da svolgere a casa. È giunta l’ora di combinare questo insieme variegato di fattori per capire come funziona la scuola, come nascono e sono prodotti i risultati che la scolarizzazione consegue. Si deve quindi adottare un approccio comprensivo, transdisciplinare, che integri prospettive multiple. Questa esigenza presuppone uno sforzo intellettuale considerevole per connettere tra loro fattori diversi ma ciò è indispensabile se si ha l’ambizione di aiutare gli studenti provenienti da ambienti diversi, con culture diverse, con esperienze sociali e umane diverse, ad apprendere e soprattutto se miriamo a rispondere meglio a quanto si aspettano gli studenti quando si recano a scuola.

Questo impegno ad interconnettere competenze e profili professionali diversi è una sfida e una responsabilità sociale, un impegno etico che trascende qualsiasi remora. Solo in questo modo si riuscirà a sviluppare ambienti d’apprendimento dinamici.

ll convegno dell’AERA di quest’anno aveva per l’appunto l’ambizione di mettere in evidenza i limiti degli approcci tradizionali della ricerca scientifica sulla scuola e di invogliare a intraprendere nuove piste di lavoro.

---

Les documents attachés à cet article :

Dibattito aperto negli USA

La dispersione scolastica cresce o diminuisce?

Norberto (14/08/2009)

Problemi di misura: come la si calcola?

Description :

Come si misura la proporzione dei diplomati alla fine della scuola secondaria superiore e come si misura la dispersione scolastica. Il fenomeno è in crescita oppure sta calando?

Le radici della polemica sulle ciffre

Se il numero dei dropout stagna o cresce, ciò significa che il sistema scolastico e le politiche scolastiche non sono stati in grado di lottare contro la dispersione scolastica e di migliorare il livello medio d’istruzione della popolazione giovanile. Quindi, quando Obama afferma che la dispersione scolastica negli Stati Uniti è triplicata in quarant’anni, critica implicitamente le politiche scolastiche del passato e ciò facendo offende la suscettibilità dei difensori della validità del sistema scolastico statale, in primo luogo dei docenti e dei loro sindacati. Il calcolo del numero dei dropout non è dunque un’operazione banale oltre che essere alquanto delicata dal punto di vista statistico. Per questa ragione la polemica sorta negli USA su questo punto merita di essere seguita e commentata. [1]

Una questione di ricerca per la statistica scolastica

Questa discussione statunitense sulla misura della dispersione scolastica merita di essere segnalata perché è una bella dimostrazione di come funzioni la ricerca scientifica, in questo caso, sulle questioni scolastiche. Per principio, non si deve dare nulla per scontato. Qualsiasi affermazione va verificata, soprattutto se si tratta di statistiche. Orbene, questa reazione di per sé normale non avviene quasi mai nei paesi dove la comunità scientifica operante nel settore scolastico è debole come è il caso per esempio in Italia  [2].


Quale è il problema? Gli effetti perversi del prolungamento dell’obbligo scolastico.

Tutta una serie di fattori concorrono a rendere drammatica e in certi casi inaccettabile la dispersione scolastica. In primo luogo occorre menzionare la teoria del capitale umano secondo la quale esiste una stretta correlazione tra livello di istruzione di base della popolazione e crescita economica. Da almeno una cinquantina d’anni le organizzazioni internazionali battono il chiodo su questo tema e producono documenti e statistiche che mirano a dimostrare la pertinenza di questa affermazione come se fosse un assioma indiscutibile. Molte politiche scolastiche nazionali hanno ripreso questa teoria e impostato, nel corso di questi ultimi decenni, riforme scolastiche miranti a prolungare la scuola dell’obbligo. L’opinione pubblica, influenzata da questo dibattito, ha sostenuto questa tendenza con il risultato di un’estensione della scolarizzazione oltre la fine dell’obbligo scolastico. Ne è derivato un crescente rischio di disoccupazione e di altre forme di esclusione per i giovani con insufficiente livello d’istruzione il che ha permesso di denunciare, sul piano politico, le conseguenze negative della mancanza d’istruzione o dell’insufficienza della formazione professionale, come se questi dati fossero la prova della validità della teoria e non l’effetto di una concezione peculiare dell’architettura scolastica e della sua organizzazione. 



Negli ultimi anni si è assistito al prolungamento del periodo di transizione dalla scuola all’occupazione, periodo che è diventato più lungo e complesso rispetto al passato. I confini di età della transizione dall’istruzione secondaria a quella terziaria si sono dissolti. Il passaggio ora si verifica in larga misura nell’età che va dai 15 ai 24 anni. Ciò offre ai Paesi l’opportunità di esplorare nuove strutture organizzative per l’apprendimento sia all’interno che all’esterno del sistema scolastico, ma la stessa tendenza concorre prepotentemente, quando non è controbilanciata da provvedimenti appropriati, all’aumento del numero dei dropout, ossia dei giovani che abbandonano la formazione prematuramente senza conseguire nessun diploma, con il rischio di non trovare un’ attività professionale e di sperimentare una transizione complicata e prolungata verso la vita attiva.

L’obbligo scolastico termina nei Paesi dell’OCSE tra i 14 e i 18 anni, nella maggior parte dei casi a 15 o 16 anni. Fino al termine dell’obbligo scolastico, praticamente tutti i giovani sono iscritti a scuola, ma una volta portato a termine l’obbligo scolastico, i tassi d’iscrizione cominciano a calare. Il calo è più rapido in alcuni Paesi che in altri.

Anche in Inghilterra

Il problema non è solo americano, come lo dimostra un articolo pubblicato il 13 agosto 2009 dal servizio scuola della BBC. Il numero dei giovani che in Inghilterra non erano né disoccupati né in formazione alla fine di aprile del 2009 era di 935.000 . Questa proporzione passata al 16% nel primo quartile di quest’anno, era rimasta stabile, attorno al 14%, fin dal 2001. In Inghilterra, oggigiorno, quasi un giovane su cinque in grado di trovare lavoro è disoccupato. Anche in Inghilterra, il dibattito in Parlamento tra conservatori e laboristi sulla pertinenza e il significato di queste cifre è immediatamente scoppiato. In questo caso, abbiamo un’interpretazione del concetto di dropout e di dispersione scolastica diverso da quello in auge negli USA. Qui si calcolano tra i dropout i giovani che hanno concluso con successo la scolarità ma che non trovano lavoro e che sono allo sbando nonostante gli studi fatti. Si potrebbe dire che il problema in questo caso è di natura economica, ma non lo è solo in parte. E’ anche scolastico, infatti, perché il sistema scolastico statale non riesce ad agganciarsi in modo armonioso al sistema economico. Questa difficoltà è cronica ma diventa acuta in periodo di crisi economica.
 

 

L’allarmismo americano

Quando il presidente Obama ha alluso alla crescita della dispersione scolastica, faceva riferimento al periodo che si situa tra la fine dell’obbligo scolastico e l’inizio dell’istruzione terziaria [3]. In altri paesi, segnatamente in quelli dove le bocciature e le ripetenze sono una pratica corrente [4], la dispersione scolastica è elevata anche nell’insegnamento secondaria di primo grado, dove vivacchiano studenti bocciati più volte, che hanno già raggiunto l’età che li esautora dall’andare a scuola mentre sono ancora nella scuola media. Logicamente, da un certo punto di vista, ma non da quello della teoria del capitale umano, questi studenti, appena possono, se ne vanno dalla scuola in barba alle raccomandazioni degli educatori e degli operatori sociali e raggiungono le fila dei dropout.

Quale significato ha la dispersione scolastica?

Nella prospettiva dello sviluppo economico non è affatto indifferente conoscere il livello medio d’istruzione iniziale della popolazione perché si ritiene che esista una correlazione tra il livello d’istruzione e formazione e il benessere economico. Dal punto di vista educativo o meglio scolastico però questa informazione non dovrebbe servire soltanto, come è il caso finora, per valutare la qualità del sistema scolastico, ma è anche un indice che dovrebbe incitare a riflettere sulla pertinenza di un determinato modello d’istruzione. Ci si può infatti chiedere se la finalità esclusiva dell’istruzione scolastica debba essere la ricchezza delle nazioni oppure se questo obiettivo, che è fondamentale per lottare contro la povertà e la miseria, non possa essere conseguito mediante uno modello d’ istruzione scolastica statale di natura diversa, che non obblighi strati di giovani a macerare nelle scuole o a trovarsi sul lastrico dopo anni passati sui banchi di scuola, senza nulla in mano, senza avere nemmeno acquisito una capacità minima di comprensione dei testi scritti e della cultura scientifica dopo otto o nov e anni di scuola, per incompatibilità totale con il modello scolastico standard. I devianti sono irrimediabilmente espulsi dalla scuola statale. Forse un altro modello d’istruzione scolastica potrebbe essere più appropriato. Ecco perché il dibattito sulla misura della dispersione scolastica e sull’evoluzione del numero dei dropout non è affatto banale. Se questo numero stagna o cresce nonostante le riforme scolastiche, ciò significa che qualcosa non quadra, ossia che il sistema scolastico statale non è adeguato allo sviluppo di tutti i profili di studenti.

Come si calcolano i dropout?

Per chiarire queste questioni, il Thomas B. Fordham Institute  ha chiesto a Christine O. Wolfe, che si è occupata dii queste questioni in seno all’ amministrazione federale americana, di redigere un documento per spiegare le formule complesse utilizzate per calcolare il numero dei dropout e riassumere le discussioni tra esperti su questa materia. Il documento, intitolato "The Great Graduation-Rate Debate" è allegato a questo articolo. In 20 pagine, l’autore presenta le formule più comunemente usate, descrive le percentuali di dropout più in voga negli Stati Uniti, specifica come si sono conseguite e infine estrapola le considerazioni che si possono trarre da questi dati. L’autore prende pure posizione sulla pertinenza delle misure che sono scodellate dai vari enti e tratteggia le probabili tendenze di calcolo che prevarranno in futuro, amalgamando da un lato l’evoluzione delle tecniche di calcolo con le preoccupazioni di natura politica.

 

---

Les documents attachés à cet article :

Première grande recherche longitudinale sur les effets de l’éducation scolaire en Allemagne

Les parcours de formation de 60.000 allemands passés au crible

Norberto (12/02/2009)

"Panel sur l’éducation pour l’Allemagne" (National Educational Panel Study, NEPS)

Description :

Annonce du lancement en Allemagne de la première étude longitudinale sur vaste échelle financée par le Ministère fédérale de l’éducation et de la recherche sur l’évolution tout le long de l’existence des compétences acquises au sein du système d’enseignement.

Politique de la recherche

Recherche pédagogique : les parcours de formation de 60.000 allemands passés au crible

Le 3 février 2009, la Ministre fédérale de l’enseignement et de la recherche, Annette Schavan, a donné à Bamberg le coup d’envoi d’une étude à long terme en recherche pédagogique : pendant plusieurs années, 60.000 Allemands d’âges différents seront suivis pour comprendre le déroulement de leur parcours de formation.

Ce "Panel sur l’éducation pour l’Allemagne" (National Educational Panel Study, NEPS)a pour objectif de mesurer le développement des compétences dans le parcours de chacun des participants du panel et devrait fournir des réponses aux questions-clés de la politique de l’éducation. Ainsi, il sera pour la première fois possible d’observer comment des enfants de même origine et disposant des mêmes compétences se développent dans des institutions éducatives différentes, de déterminer quels facteurs provoquent des évolutions positives, à quels éléments l’échec est lié et comment des cas dits "à risque" peuvent être précocement détectés. "Cette étude sur le long terme va fournir à la recherche pédagogique des données de fond détaillées, qui vont nous aider dans l’application de nos approches pédagogiques", espère Annette Schavan. Il s’agit de l’un des plus grands programmes jamais menés en sciences sociales en Allemagne. D’ici 2014, les coûts du panel devraient s’élever à 60 millions d’euros. 150 chercheurs issus des meilleurs instituts allemands prennent part à cette étude.

Le panel est coordonné par un réseau de recherche dirigé par le Prof. Hans-Peter Blossfeld, sociologue à l’Université de Bamberg. 

Le Ministère fédéral de l’enseignement et de la recherche (BMBF) finance les travaux de recherche à hauteur de 7,5 millions d’euros en 2009. Les moyens mis à disposition de l’étude par le BMBF devraient continuellement croître d’ici 2013 pour atteindre finalement 16 millions d’euros. Le Land de Bavière et l’Université de Bamberg financent eux aussi le projet. L’étude suivra l’origine et les caractéristiques sociales des participants, de même que leurs capacités en lecture, en mathématiques et en sciences naturelles. Cette année, le suivi portera d’abord sur un premier échantillon de 13.000 adultes âgés de 23 à 64 ans. En 2010 suivront des enfants de quatre ans, des écoliers de niveau CM2/sixième et troisième/seconde, ainsi que des étudiants.

La particularité du panel est son ancrage sur le long terme, qui est comparable à un film documentaire. Les mêmes personnes seront interrogées et testées régulièrement sur de longues périodes, afin de comprendre comment les compétences s’épanouissent au cours de la vie, comment elles influencent les décisions au moment des transitions dans le parcours de formation et de quelle manière elles sont marquées par l’origine familiale et les institutions éducatives fréquentées. "C’est exactement là que se situe le progrès, par rapport, par exemple, aux études transversales internationales comme PISA  [1] et PIRLS [2]  qui, telles des photographies, ne peuvent fournir que des instantanés de la situation au moment de l’étude", commente Annette Schavan.

Pour en savoir plus, contacts :www.uni-bamberg.de/neps/

---

Taille des classes/Effetto dimensione della classe

Teachers’ Training, Class Size and Students’ Outcomes

Pascal Bressoux (1/09/2008)

Evidence from Third Grade Classes in France

Description :

Analisi delle correlazioni tra risultati degli allievi di terza elementare in Francia, tipo di formazione iniziale degli insegnanti e dimensione delle classi.

Sugli effetti della dimensione delle classi, ossia sulla dibattuta questione dei benefici delle classi piccole, si veda anche la relazione presentata da Peter Blatchford all’AERA Meeting 2008 :Do low attaining and younger students benefit most from small classes?. I risultati presentati da Blatchford collimano con le conclusioni di Bressoux e collaboratori.

********************************************************************

Il documento integrale della ricerca con gli allegati, corredato di tutti i grafici e di tutte le tavole, è in calce.

Le texte intégral avec les annexes, les graphiques et les tableaux relatifs est attaché à ce document.

All tables and exhibits are in the full-paper attached

********************************************************************

Teachers’ Training, Class Size and Students’ Outcomes: Evidence from Third Grade Classes in France

 [1]

Pascal Bressoux*,

Francis Kramarz**

and Corinne Prost***

* Université de Grenoble

** CREST, CEPR, IZA

*** CREST, EHESS Paris-Jourdan, Cornell University

Preliminary Version, December 2005

Introduction

The literature on the effects of class size on student learning is huge. Yet there is no consensus on the impact of class size and the debate is still impassioned [2] . Some economists, who do not believe much that smaller class size can improve students’ performance, or who find that it is a very costly policy, argue that other policies besides class size reduction, such as improving teacher quality, are more important.

Understanding the relationship between teachers’ characteristics and students’ achievement is obviously of prime importance in the analysis of the education system. Research on this topic has often focused on specific characteristics such as teachers’ diplomas, experience and salaries. Few studies have specified the impact of teacher in- service training in developed countries. Angrist and Lavy (2001) present an evaluation of the effect of in-service teacher training in Jerusalem schools. They find that the causal effect of the program on pupils’ test scores is significantly positive. The cost- effectiveness analysis suggests that teacher training may provide a less costly means of improving pupil achievement scores than reducing class size or adding school hours.

In France, most studies on teachers have looked at teaching practices, and little empirical work has examined the consequences of teachers’ training on students’ outcomes. Bressoux (1996) partly fills this gap. In order to study the effect of teachers’ training and experience on third-grade pupil achievement, he uses a specific survey on third grade students and teachers in 1991, with a quasi-experimental design. This data source includes three types of teachers: untrained novice teachers, trained novice teachers, experienced teachers. Bressoux finds that training improves students’ scores in mathematics. Experience seems also to have a positive impact on pupil achievement.

Importantly though, the experiment used in the above study is not randomized. The ideal situation would involve the random assignment of pupils to the different types of teachers. In fact, Bressoux (1996) shows that classes differ according to the status, experienced, trained novice or untrained novice, of the teacher. Hence, in the absence of random assignment, Bressoux estimates the impact of training using regressions controlling for numerous variables. The estimated effect is the causal one if no unobserved student or class characteristic is correlated with the teacher’s type and with the student’s test scores. Otherwise, estimates are potentially biased.

This paper uses the same data, but relies on a methodology that takes care of the non- randomized design. The idea comes from the specificity of experienced teachers. The fact that the allocation of classes is not random is virtually only due to experienced teachers, who can choose their schools, and who often choose advantaged zones. But, in principle and in the data, trained and untrained novice teachers are assigned to almost similar classes. So our paper uses the fact that, when excluding experienced teachers, we are faced with a quasi-randomized design. We check the robustness of this feature using different estimation methods, both conditional and unconditional on other observed variables.

The data used here are very rich. The unit of observation is the student, a very important element for this kind of analysis (see Summers and Wolfe, 1977). Multiple students’ characteristics are collected. Furthermore, all students within a third-grade class are included in the sample. This gives us an opportunity to control for class effects. In addition, teachers also provide a lot of information on their personal characteristics, their teaching practices, as well as characteristics of their classes and their schools. Moreover, students’ achievement is extremely precisely measured by detailed test scores at the beginning and at the end of the year.

A first aim of this paper is to check that Bressoux’s findings on training – better trained teachers induce higher students’ outcomes – are robust. To perform this task, we use more recent statistical methods, controlling for the endogenous allocation of classes. The estimation is made excluding experienced teachers, in order to estimate the causal effect of training of novice teachers on pupils. Particular attention is given to heterogeneous effects. A second goal is to see if some particular characteristics of the teachers, such as their university background (which was not used in Bressoux (1996)), have any impact on their students’ outcomes. This paper also examines other class characteristics, more particularly class size. Indeed, when excluding experienced teachers, it appears that class size is not correlated with pupils’ initial test scores. There is no sign of a relation between class sizes and class mean initial achievement or class socio-economic background. Thus, it seems that no selection bias in class size allocation is present when the sample is restricted to novice teachers. Consequently, we use similar methods to assess the effect of class size as were used to estimate the effect of training effect.

The findings on the training effect are very close to those found by Bressoux (1996): the training of novice teachers promotes students’ learning in mathematics. Yet it seems that within classes, less able students do not benefit from their teachers’ training. In addition, training allows teachers to improve significantly students’ reading scores only in high achieving classes.

We also find that teachers’ education background has a significant impact since untrained teachers who majored in sciences at university have the same effect as trained teachers. It seems that their past studies help them to compensate for the lack of training.

The estimated impact of class size implies that reducing class size has a positive and substantial effect on third-graders. These results are close to the findings of Piketty (2004) on the effect of the size of French second-grade classes. It appears that the effect is similar on students’ reading scores within the classes; it is larger for less able pupils in mathematics. Moreover, a smaller class size improves more students’ scores when they are in a less achieving class, which could be the consequence of higher frequencies of disruptions in this kind of classes, as described in Lazear (2001).

The paper is organized as follows. Following a description of the data in Section I, Section II describes the statistical model and the empirical tests. Section III reports the main estimation results and Section IV concludes.

I. Data and descriptive statistics

The data come from a survey conducted by the French Ministry of Education. They cover a sample of classes of third-graders (8 years old) and their teachers. The quasi- experimental design is due to a feature of the system of teachers’ training in France. This characteristic implies that some novice teachers start their job before any training.

In France, except for a subset of private schools, teachers are civil-servants recruited and paid by the State. After having passed a competitive examination, primary school teachers are trained in specific schools. At the beginning of the 1990’s, these schools were called ‘écoles normales’. France was, and still is, geographically divided into administrative ‘départements’ and there was an ‘école normale’ in each ‘département’. [3]

Novice teachers are recruited among students who have passed a competitive examination for entering an ‘école normale’. To take this examination, students have to have already passed an examination corresponding to two years in a university. The number of slots in the école normale’ is limited and determined each year at the central level, using forecasts for teachers’ positions. All applicants are ranked according to their grades in this examination. The students ranked first enter the ‘école normale’ and are trained during two years. Students who are ranked just after the last admitted candidate on this primary list are assigned and ranked within a waiting list.

In September, the number of vacant job slots is often greater than the one expected two years earlier. So students who have finished their training at the ‘école normale’ are assigned to some of these job slots, and, in October, some students in the waiting list are assigned to the vacant slots. Hence, these persons have to teach a class for an entire school year without receiving any training. They enter the ‘école normale’ the year after.

The survey was conducted in the school year 1991-1992. The sample included explicitly three categories of teachers: untrained novice teachers, trained novice teachers and experienced teachers. The sample covered third-grade students and their teachers in 12 ‘départements’. The teachers were teaching in third grade classes or in multi-grade classes including third graders. In the 12 ‘départements’ selected, all novice teachers were surveyed while experienced teachers were chosen randomly. Finally, the survey covered 4,001 students and 197 teachers. The numbers of teachers within each category were not perfectly balanced: there were 96 experienced teachers, 65 trained novice teachers and 36 untrained novice teachers (see table 1). [4]

The information about the students is comprehensive: parents’ occupations, sex, month of birth, nationality (French or not), number of siblings, number of years spent in a pre- elementary school, repeated classes (see the statistics in table 2). In addition, two sets of scores are available in the data. In France, there is national testing of all pupils just at the beginning of the third grade, both in reading and mathematics. The reading tests comprise grammar, vocabulary, spelling and reading comprehension per se. The mathematics tests comprise arithmetic, geometry and problem-solving. For this specific survey, covered pupils have also been tested at the end of the school year in both subjects, using a design similar to that prevailing in the entry tests. For each of the two subjects, initial and final scores are standardized (mean=100, standard error=15).

In addition, teachers had to answer a questionnaire on their personal characteristics, on their teaching practices and on the characteristics of their classes and their schools. The main variables used in the following are the field of specialization of the teacher during his/her studies at the university (sciences, unknown, other), the class size, the fact that the class is or not a combination class mixing students across grades, the category of the area of the school (rural, semi-rural, urban), and the priority status (see the statistics in table 3). The mean of the class sizes is 23.9 students per class, with a standard deviation of 4.1. By comparison with data on all elementary schools, Piketty (2004) finds that the average class size in the primary schools (first grade to fifth grade) is close to 23.3 in the school year 1991-1992.

Unfortunately, the scores are not available for all the students. This attrition comes from two reasons. First, some students were not in class when the tests were conducted. Second, for some classes, all the scores are missing. The scores of reading tests are not known for 974 students and the scores of math tests are not known for 778 students. The class size also is not known for all classes: for 8 classes, the class size is unknown and can not be approximated by the number of students of the sample, because these classes mix students of different grades.

Nevertheless, this attrition should not induce any bias: tables 4 and 5, compared to tables 2 and 3, show that the characteristics of the students whose scores and class size are known do not significantly differ from those of all the students. It seems that the absence of information on the scores or the class sizes have random origins.

The data do not come from an experimental design. In fact, the assignment of the different types of teachers to the classes is not randomized. Indeed, the system of job assignment depends on the teachers’ choices. When the choices of different teachers are the same, the final assignments depend on the years of experience and on a mark given by the administration, this mark being well correlated with the years of teaching experience. Hence, as they accumulate experience, teachers are able to choose the schools they want, and mostly go from disadvantaged schools to advantaged ones. On the contrary, novice teachers go to schools that have not been chosen by experienced teachers, or where there are free job slots because some experienced teachers retired or are absent for the year.

The data show that the aggregate characteristics of pupils vary with teachers’ types (see tables 6 and 7). Indeed, experienced teachers have on average better classes. In these classes, compared to those with novice teachers, initial scores are higher, the share of non-French pupils is lower, children have fewer siblings, fathers and mothers have more often a high occupation and students less often repeated the first grade. In addition, the class sizes are on average larger, and the schools are less often in a priority educational area.

The classes with trained novice teachers and those with untrained novice teachers are more alike. Nevertheless, trained novice teachers are more often in urban areas and in priority zones than untrained novice teachers.

There is a potential source of bias due to the fact that the trained novice teachers may have had better rankings at the entrance examination at the ‘école normale’ than the untrained novice teachers. If the examination measures the initial teaching abilities (a fact that should be proved), this bias could imply that the trained novice teachers are better able to teach than the untrained novice teachers. Fortunately (for us), the survey has been conducted during the school year 1991-1992, which is an atypical year, as can be seen in figure 1. Indeed, in 1991, the number of students selected for entry into the teacher training centers was very small. So the surveyed untrained novice teachers, who had taken the entrance examination in 1991, had very good rankings and would have been selected for entry had they competed for the examination during another year, and especially during the year 1989, when the surveyed trained novice teachers had passed their entrance examination. So the selection bias is likely to be weak.

II. Statistical method

The non-randomized assignment of the three types of teachers can also be observed through a regression of initial test scores on student and teacher characteristics. If the coefficients of the dummy variables for the types of the teacher are significant, it means that the assignment is non-randomized since the students have not been exposed to these teachers’ teaching yet.

The regression is estimated on all the students. It includes random class effects, in order to take account of the correlation between students within classes. Indeed, class variables may be not sufficient to control for these correlations. So, it is important to incorporate class effects: without them, the standard deviations could be underestimated. It would be the case with OLS estimation (see Moulton (1986)). However, Moulton stresses the problem of the precision of coefficient estimates, but he also shows that the coefficients may be different when the estimation incorporates random class effects without imposing the absence of correlation between these effects and the other covariates. Indeed, this kind of estimation results in substantial gain in efficiency. Throughout this paper, class effects are estimated through mixed models (see Robinson (1991)). These models allow a general specification of class effects, fixed effects being only a specific case of this specification. Identification of class effects uses more information than for “classic” fixed effects: it uses the variance of the class effects instead of only the mean, thanks to a more general prior distribution (see appendix A).

The results of the regression of initial test scores on teacher type are detailed in table 8 (full results in table 11). They confirm that experienced teachers teach in better classes. Table 8 reports that the correlation between student initial scores and the dummy variable for the teacher experience is significant, in reading and in mathematics as well. These two correlations remain significant, even when controlling for student characteristics. On the contrary, it seems that classes with untrained novice teachers and classes with trained novice teachers are not different in terms of initial achievement, since the correlations between initial scores and the dummy of the teacher training are not significant, with or without other controls. Thus it seems that there is no selection of trained teachers, so that the classes of such teachers appear similar to those of untrained teachers according to pupils’ initial achievement.

This idea is checked with the same regression of initial scores, but with the sub-sample of the pupils having novice teachers. The results are given in table 9 (full results in table 12). The coefficient of the training dummy is never significant.

Bressoux (1996) assumes that the selection bias of experienced teachers can be controlled with the observed variables, including initial test scores. The causal interpretation of the coefficients related to the type of teacher relies on the assumption that no selection bias comes from unobserved variables.

This paper takes care of the non-randomized design in order to assess the robustness of the teachers’ training effect found in Bressoux (1996). The idea is that trained and untrained novice teachers are randomly assigned to classes, at least according to our observed variables. Hence we have chosen to estimate the training effect on the sub- sample of novice teachers. In the spirit of matching classes to classes, either taught by trained or untrained novice teachers, we focus on pupils with novice teachers and discard those with experienced teachers. It means that we manage to have a sample of similar students, some have trained novice teachers and constitute the treatment group, and some have untrained novice teachers and constitute the control group. Thus we can expect that no bias perturbs the coefficients in the estimation and that the coefficient of the treatment estimates the causal effect. The idea is close to the one in Angrist and Lavy (2001). In this paper, they observe that pupils in the treatment group have initially lower score than pupils in the control group. As they would like pupils in the control group to be comparable to pupils in the treatment group, they match individual pupils on the basis of their initial test scores, by dividing test scores into quartiles and comparing treatment and control scores in each quartile. Here, we restrict the sample in order to have similar pupils in the treatment and control group. But, on the contrary to Angrist and Lavy, we keep a regression strategy, in order to control for the other covariates, and more specifically to control for class effects. We will see that these controls are important.

Thus, we will be able to estimate the effect of training on achievement using this specific sample. Nevertheless, we will have to keep in mind this restriction while interpreting the results: what we estimate is the effect of a trained novice teacher on pupils’ achievement, compared to the effect of an untrained novice teacher, and not compared to all other kinds of teachers.

In the meantime, we will keep this strategy to estimate the class size effect. Indeed, table 8 reports that the correlation between initial scores in reading and class size is positive and significant when all pupils are included in the regression. When adding other covariates, this correlation remains significant and positive, even if it is less significant. On the contrary, table 9 reports that even without any other control, class size is no more correlated with initial scores when the sample is reduced to the students with novice teachers.

Figures 2 and 3 present these results. These figures show the link between class sizes and class means of initial test scores in reading. The classes with experienced teachers are presented in figure 2 while figure 3 presents the classes with novice teachers. It is clear that all experience teachers teach high achieving classes whereas the scores of the classes with novice teachers are much more dispersed. Also, experienced teachers more often have larger classes. At last, the positive correlation between class size and scores can be seen in figure 2 with experienced teachers, even if it is not obvious, and it appears in figure 3 that there is no more correlation for those classes with novice teachers.

The idea that class size can be positively correlated with student achievement is well known: the education system is often organized in order to support less advantaged pupils by gathering them in small classes whereas more advantaged students are assigned to larger classes. Hence the differences in class sizes are often in relation to students’ socioeconomic background and scores. The selection bias in the relationship between test scores and class size can be generated within schools as well as between schools. This selection bias is one reason why causal effects of class size can be difficult to measure.

There are several reasons explaining why, in France, the selection could be weak for third grade classes. First, the system of assignment of teachers is centralized, and is not supposed to make any difference between schools in terms of resources. The only official exception is the policy of education priority areas (ZEP, ‘zones d’éducation prioritaires’). The ZEP policy is a program implemented in 1982, which gives more resources to disadvantaged schools (for a description and an assessment of this program on sixth and seventh graders, see Benabou, Kramarz, Prost (2003)). According to our data, the classes in the ZEP have on average 23.8 students per class, whereas the mean class size in the non priority zones is 25.25. [5]

The other case where it is known that some schools have smaller classes than the other schools is the one of rural schools: because of small enrollments, these schools have often small classes, even if they often organize combination classes by mixing students of different grades in a class. Yet the conclusion in terms of selection is not clear since, as we will comment on this later, pupils in rural schools have better achievement at the beginning of the third grade (but improve less during the year).

Nevertheless, there may be selection within schools. This selection is possible in large schools, when there are several third grade classes. Yet we will see that when the enrollment exceeds 30 students, it does not always entail a new third grade class, but sometimes some third graders are assigned to a class with students of other grades. To facilitate this assignment, the school may choose good pupils to go to this combination class, so that students who stay in the larger class are not necessarily the better ones.

The organization of a selection needs large schools. Since experienced teachers are much more often in urban areas, where schools are bigger, this may explain why the selection on initial scores can be observed for classes with experienced teachers and not for novice teachers. On the contrary, it seems that pupils with novice teachers are not assigned to classes with different sizes according to their abilities.

Finally, we will estimate the class effect on the sample excluding experienced teachers. As the correlation between class size and observed initial scores is significant on the whole sample, we suspect that there may also be a selection on unobserved variables, which could disturb the estimation of the causal effect of class size. On the contrary, the correlation between class size and initial scores is not significant with the sub-sample of pupils with novice teachers. Hence we assume that the “traditional” bias selection is expurgated. Finally, to check the robustness of our findings, we will also estimate the class size effect on all the students, using instrumental variables.

III. Results A. Global effects

The results of the estimation of the effects of teacher and class characteristics on pupil achievement are reported in table 10 (details in table 13). It is a regression of final scores on initial scores and student, teacher and class characteristics. The estimation includes class effects and is estimated on the sub-sample of students with novice teachers.

The data include a lot of information about the teachers and their teaching practices. They include in particular the diploma, the subject studied at university, the number of hours per week used for teaching reading or mathematics, the number of hours asked for homework per week, the practice of organizing the class in groups, and how these groups are chosen.

All these variables have been tested in the regression of final test scores on initial scores, student and class characteristics. When the variables on the number of hours per week used for teaching reading or mathematics, the number of hours asked for homework per week, the practice of organizing the class in groups, and how these groups are chosen are added, the coefficients on these variables are not significant. The small number of classes in the sample may prevent from identifying these effects that may be non linear. Therefore, these variables are not included in the final estimation.

The only teacher characteristic that is used is the subject studied at university. More precisely, dummy variables are included for teachers having majored in sciences at university (14% of novice teachers) and for teachers having majored in a discipline not reported in the survey (roughly 14% of novice teachers). The reference group therefore comprises those teachers who majored in humanities (often French or another language, sociology, psychology, history). Novice teachers are all endowed with similar diplomas since it is compulsory to have a diploma equivalent to two years university to enter an ‘école normale’. This was not the case in the past, and among experienced teachers, only a few went to university.

The regressions of final test scores include class characteristics. Some class characteristics can be calculated using the means of individual characteristics. We built class variables such as the share in the class of students with advantaged parents as measured by occupations, the share of girls, the share of non-French students and the share of students who repeated at least one grade. These variables are calculated for each student, excluding his/her own characteristics in the calculation of the means. None of these variables give significant coefficients. They are not included in the regressions presented in this paper. This confirms the difficulty in estimating peer effects without a clean experimental design.

On the contrary, means and standard deviations of initial test scores per class have significant correlations with final scores. For the regression of final scores in reading, the included variables are the class means and standard deviation of initial scores in reading. Likewise, the means and standard deviation included in the regression of scores in mathematics are calculated on initial scores in mathematics. These means are also calculated for each student, excluding his/her own characteristics. Table 13 reports the effects of class characteristics on final test scores, and means and standard deviations of initial test scores have negative impacts on pupil improvement, meaning that students have better results in a homogeneous class and when the average achievement is not too high.

The estimated impact of training is not significant on reading achievement but it is significant and large on mathematics achievement: students gain more than 3 points on their final scores when their teachers have been trained. This effect is substantial; it is more than one fifth of the standard deviation of final scores. These results are close to the findings in Bressoux (1996).6 They are also close to the raw differences of the means: as can be seen in table 6, students with untrained novice teachers have similar initial scores than students with trained novice teachers; yet, they improve much less during the year. The raw differences-in-differences estimator gives an effect of 2.5 in reading, and 2.1 in mathematics. Incorporating student and class characteristics show that the effect is larger in mathematics, since it is close to 3. [6] On the contrary, the effect is much weaker in reading, since it is close to 1.3 and is not significant. The estimation of the regression without class effects would have drawn to a significant effect equal to 2.6. Hence incorporating random class effects shows that the coefficient is weaker and non significant.

The teachers’ educational background has also a substantial impact. The finding is that teachers who majored in sciences improve their pupils’ mathematics achievements more than other teachers do. These teachers are either trained or untrained. The effect is not significantly different for these two kinds of teachers, and not significantly different from the training effect. Hence, even though the training effect is substantial in mathematics achievements, teachers who have not been trained, but who have studied mathematics or sciences when they were at university, compensate for this lack of training. Nevertheless, the sample is small and this result is weak.

The teachers whose fields of specialization are unknown seem to improve their students’ achievements, the effect being very significant in reading and slightly less significant in mathematics. This group of teachers comprises some individuals who did not report this information, potentially because of multiple fields of specialization, as well as some teachers who did not go to university. Indeed, very few people were entitled to take the examination for entering an ‘école normale’ without having studied at university; this was particularly the case of mothers of three or more children.

Nevertheless, it is worth noting that the repartition of teacher among classes according to their field of specialization is not random. The correlations between initial test scores and teacher dummy variable for unknown field of specialization are significant for the initial mathematic achievement (table 12). Hence, even if the regression of final score control for initial scores, estimates may be affected by selection bias since these teachers appear to be assigned to better classes.

Class size has also a significant impact on students’ outcomes. The impact is quite similar in reading and in mathematics. For test scores in reading and in mathematics, the estimated effect is -0.34. This impact is substantial: reducing the class size by 10 students increases the final test scores by 3.4 percentage points. This is nearly the same impact as the one obtained for teachers’ training in mathematics.

This impact seems to be robust to the problem of combination classes. Indeed, the regression is estimated with a sample including multiple-grade classes. In the case of a combination class, the class size is then the size of the entire class, and not the number of third-graders. Yet the dummy for multiple-grade classes is not significant [7] Likewise, results are similar when excluding these combination classes.

The coefficients of the other class characteristics are also of interest. Students in rural schools increase their achievements much less than the other students. However, as can be seen in table 12, their initial scores are higher. These results are consistent with Brizard (1995) and Thaurel-Richard (1995): pupils in rural schools have better scores at the beginning of their third-grade, but then they tend to improve less.

Students in ZEP schools improve less during the year than other students do. Nevertheless, they have lower scores even at the beginning of the school year. So it is difficult to conclude that it is a causal effect.

B. Heterogeneous effects

We then estimate heterogeneous training and class size effects. To estimate the effects on heterogeneous students, the methodology chosen is to split training and class size variables into different kinds of students, in the same regression. Indeed, it is important to keep controlling for class effects. This methodology allows us to estimate the effect on some kinds of students within the classes. In addition, we estimate the effects on heterogeneous classes. Thus we will be able to interpret more precisely some results on heterogeneous effects. Indeed, some papers find that the class size effect is larger for low achieving students. We will see that this can be explain by the fact that these low achieving students are more often in disadvantaged classes; but class size affects similarly all students within a class.

Tables 14 and 15 present the results for the regression of final test scores including heterogeneous effects. These effects are measured by breaking down the dummy variable for trained teachers or the class size variable according to the quartiles.

To estimate the heterogeneous effects on pupils, the quartiles are defined by students’ initial scores in reading and are measured within the classes. No significant training effects appear for scores in reading. For scores in mathematics, the training effect is substantial and significant for the more achieving students. On the contrary, it is not significant for the low achieving students. It seems that training helps teachers to improve students’ results in mathematics, except for the least able ones.

In order to identify different kinds of classes, the quartiles are estimated on the class means of initial scores in reading. The decomposition reveals significant effects of training on scores in reading for the highest achieving classes. In mathematics, the effect seems to be significant in more achieving classes and not in low achieving classes. Yet the Fisher tests accept the hypotheses that the coefficients for the third and the fourth quartiles are not significantly different from the coefficient for the first quartile. So training seems to help teachers to improve their teaching, except when they face a class where the mean achievement is low: training is no help for less advantaged classes.

Among students, no heterogeneity of the class size effect appears for scores in reading. So it seems that within the classes, class size affects similarly all students. Yet, the effect on mathematics scores decrease when the “quality” of the students increases.

Among classes, the class size effect appears much more substantial for less advantaged classes and decreases when the “quality” of the class increases. The clearer results are for the scores in mathematics; the Fisher tests accept the equality of the coefficients for reading tests.

This heterogeneity is confirmed by table 16, which reports the results of the estimation of class size effect for all schools and for ZEP schools alone, estimated in the same regression. The coefficients are significant only for ZEP schools and are very substantial:
 0.6 in reading and -1.1 in mathematics. This finding confirms recent results obtained by Piketty (2004) who also finds substantial impacts of class size in ZEP school, albeit marginally significant because of the small number of students in ZEP schools in his sample.

These results show that the students in ZEP schools and in disadvantaged classes in general are more sensitive to class size as a group than the other groups of students. It may come from problems of behavior in class, the probability of a troublemaker among students of a class being larger in these schools (see Lazear, 2001).

C. Instrumental variables for the class size effect

The effect of third-grade class size, as estimated in this paper, stands between -0.3 and - 0.4 percentage point of final test scores. Piketty (2004), on second-grade class size, finds an impact of -0.4 to -0.5 percentage point. He applies a methodology developed in Angrist and Lavy (1999). His method is based on the following specificity of French class openings: when second-grade enrollment goes beyond 30, another class is opened in most cases. Hence, the two new classes have an average size of 15 pupils. Piketty uses this discontinuity as an instrumental variable. He finds that a reduction in class size would induce a significant and substantial increase in mathematics and reading scores, and that the effect is larger for low achieving students.

In our data, we find similar specificities as those observed by Piketty (2004) (see figure 4). [8] There are often two classes when the number of third-graders in the school is greater than 30. Yet there are some classes gathering up to 34 pupils. And the link between class size and enrollment is complicated by the existence of combination classes. When the enrollment goes just beyond 30 students, the schools do not open another third-grade class, but instead, assign some third-graders to classes with students of other grades.

When we exclude combination classes, there is less diversity in class sizes. Figure 5 shows the link between the enrollment of third graders and the class sizes. Yet when there are two classes, these classes have often different sizes; it could then be a source of bias if the sizes are determined according to the socio-economic background or the achievement of the students.

To check the robustness of our class size effects estimated on the novice teachers, we use instrumental variables on the whole sample. The instrument is based upon the enrollment of third-graders in the school when we exclude combination classes. In order to work with all classes, the instrument is also based upon the numbers of third-graders and students who are in a class with third-graders (see figure 6). In all cases, the instrumental variable is the mean of the class sizes in the school: the sizes of the third-grade classes in the first case and those of all classes with third-graders in the second case. This instrument takes care of the selection bias which exists when schools organize classes so that small classes gather low achieving students and high achieving students are assigned to larger classes.

As can be seen on figures 5 and 6, the instrumental variable is very close to the actual class size. Indeed, in our data, we identify few schools with more than one third-grade class. And when there are two classes, the sizes of these two classes are not very different. Hence the findings are easy to foresee: the results estimated with the instrumental variable are very close to the OLS results.

The idea in Angrist and Lavy (1999) is to use the discontinuity of the class size resulting from the creation of several classes when the enrollment goes beyond some level, assuming that this discontinuity is exogenous. One way of using this discontinuity is to estimate the class effect only when the enrollment is close to the “breaking point”. We use this method by estimating our instrumented regression for school where the enrollment is close to 34 students, the “breaking point” according to our data. We have chosen to restrict the sample to enrollments between 29 and 40 or between 24 and 45. The coefficients are then much more substantial, even if they are not always significant (see table 19).

All these results confirm the size of the effect: class size effect is between -0.3 and -0.5 percentage point of the final test scores.

IV. Conclusions

Thanks to the use of other statistical methods, this paper confirms the finding of teachers’ training effect found in Bressoux (1996). The data used have a quasi-experimental design; the French system is such that some novice teachers teach before being trained. The effect of teachers’ training is substantial: final test scores in mathematics of students with a trained teacher are greater by 3 percentage points than the scores they would have had if their teachers had not been trained. The estimation of heterogeneous effects shows that the training effect on reading achievement is significant in high achieving classes.

The importance of teachers’ training is confirmed by the effect of teachers’ educational background. Teachers who majored in sciences at university improve their students’ outcomes in mathematics. This impact is the same for trained and untrained teachers. It means that for the untrained, past scientific studies compensate for the lack of training in mathematics.

The effect of class size is shown to be significant and negative: a smaller class size improves student achievement. The impact is evaluated between -0.3 and -0.5 percentage points. Hence, training teachers is equivalent to reducing class size by 10 students, in terms of final test scores in mathematics. It is worth noting that this equivalence is true on average. But the effects vary according to the characteristics of the classes. The effect of class size is even more beneficial in low achieving classes; these students would benefit most from a decrease in class size. The effect is particularly large for classes in priority education areas. On the contrary, it seems that this type of classes do not benefit from the training of their teachers.

References

Angrist Joshua D., Lavy Victor (1999), “Using Maimonides’ Rule to Estimate the Effect of Class Size on Scholastic Achievement”, Quarterly Journal of Economics, Vol. 114, No 2, pp. 533-574

Angrist Joshua D., Lavy Victor (2001), “Does Teacher Training Affect Pupil Learning? Evidence from Matched Comparisons in Jerusalem Public Schools”, Journal of Labor Economics, Vol. 19, No 2, pp. 343-369

Benabou Roland, Kramarz Francis, Prost Corinne (2003), “Zones d’éducation prioritaire : quels moyens pour quels résultats ? Une évaluation sur la période 1982-1992”, CREST Working Paper, No 38

Bressoux Pascal (1996), “The Effects of Teachers’ Training on Pupils’ Achievement: the Case of Elementary Schools in France”, School Effectiveness and School Improvement, Vol. 7, No 3, pp. 252-279

Brizard Agnes (1995), “Écoles rurales, écoles urbaines : performance des élèves en français et en mathématiques”, Éducation et Formations, Vol. 43, pp.105-111

Ehrenberg Ronald G., Brewer Dominic J., Gamoran Adam, Willms Douglas J. (2001), “Class size and student achievement”, Psychological Science in the Public Interest, Vol. 2, No 1, pp. 1-30

Hanushek Eric A. (1997), “Assessing the Effects of School Resources on Student Performance: an Update”, Educational Evaluation and Policy Analysis, Vol. 19, No 2, pp. 141-164

Hanushek Eric A., John F. Kain, Steven G. Rivkin (2005), “Teachers, Schools, and Academic Achievement”, Econometrica, forthcoming

Hoxby Caroline M. (2000), “The Effects of Class Size on Student Achievement: New Evidence from Population Variation”, Quarterly Journal of Economics, Vol. 115, No 4, pp. 1239-1285

Krueger Alan B. (1999), “Experimental Estimates of Educational Production Functions”, Quarterly Journal of Economics, Vol. 114, No 2, pp. 497-532

Krueger Alan B. (2000), “Economic Considerations and Class Size”, Economic Journal, Vol. 113, pp. 34-63

Lazear Edward P. (2001), “Educational Production”, Quarterly Journal of Economics, Vol. 116, No 3, pp. 777-803

Moulton Brent R. (1986), “Random Group Effects and the Precision of Regression Estimates”, Journal of Econometrics, Vol. 32, No 3, pp.385-397

Oeuvrard Françoise (1995), “Les performances en français et en mathematiques des écoles à classe unique”, Éducation et Formations, Vol. 43, pp.113-116

Piketty Thomas (2004), “L’impact de la taille des classes et de la ségrégation sociale sur la réussite scolaire dans les écoles françaises : une estimation à partir du panel primaire 1997”, mimeo

Robinson Geoff K. (1991), “That BLUP is a Good Thing: the Estimation of Random Effects”, Statistical Science, Vol. 6, Issue 1, pp. 15-32

Summers Anita, Wolfe Barbara (1977), “Do Schools Make a Difference?”, American Economic Review, Vol. 67, No 4, pp. 639-652

Thaurel-Richard Michèle (1995), “Les progrès des élèves au CE2 en milieu rural”, Éducation et Formations, Vol. 43, pp.117-123

Woessman Ludger, West Martin R (2002), “Class-Size Effects in School Systems Around the World: Evidence from Between-Grade Variation in TIMSS”, IZA Discussion Paper No 485

***************************************************************

All figures, tables and attachments are in the full-paper below

Teachers’ Training, Class Size and Students’ Outcomes

Complete paper with charts and tables

---

Studi longitudinali

Preschool : First Findings From the Third Follow-up of the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B)

Norberto (23/11/2007)

Effetti dell’educazione prescolastica /Effets de l’éducation préscolaire

Description :

The USA NCES (National Center for Education Statistics) published the third report of his longitudinal study on Early Childhood Education. As said in the article presenting the UK Millenium Study the longitudinal studies are very productive and provide a lot of information about education output. The USA study is connected to a political agenda promoting the development of early childhood education. The discussion in the States is very alive despite the decennial efforts of the militants of early childhood education like Prof. Lilian Katz to show the efficacy and utility of implementing adequate forms of early childhood education for an healthy development of children( ndr.)

Introduction by NCES

The Early Childhood Longitudinal Study, Birth Cohort (ECLS-B) is designed to provide detailed information on children’s development, health, and early learning experiences in the years leading up to entry into school. The ECLS-B is the first nationally representative study within the United States to directly assess children’s early mental and physical development, the quality of their early care and education settings, and the contributions of their fathers, as well as their mothers, in their lives.

The children participating in the ECLS-B are followed from birth through kindergarten entry. To date, information has been collected from children and their parents during three rounds of data collection, conducted when the children were about 9 months of age (2001), about 2 years of age (2003), and about preschool age (age 4, 2005). Their experiences are representative of the experiences of the approximately 4 million children born in the United States in 2001. This First Look report provides information on certain characteristics of this population of children when they were about age 4. The information in this report complements that presented in Children Born in 2001 : First Results from the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B) (Flanagan and West 2004) and Age 2 : Findings From the 2-Year-Old Follow-up of the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B) (Mulligan and Flanagan 2006).

The purpose of this First Look report is to introduce new ECLS-B survey data through the presentation of selected descriptive information. Readers are cautioned not to draw causal inferences based on the univariate and bivariate results presented. It is important to note that many of the variables examined in this report may be related to one another, and complex interactions and relationships among the variables have not been explored. The variables examined here are also just a few of the several thousand that can be examined in these data ; they were selected to demonstrate the range of information available from the study. These findings are examples of estimates that can be obtained from the data and are not designed to emphasize any particular issue. The release of this report is intended to encourage more in-depth analysis of the data using more sophisticated statistical methods.

The tables in this report present information collected during the preschool wave of the ECLS-B in the following areas : demographic characteristics of children and their families (table 1) ; children’s language, literacy, mathematics, color knowledge, and fine motor skills (tables 2 through 6) ; and children’s experiences in early care and education (table 7).

Performance on measures of children’s language, literacy, mathematics, color knowledge, and children’s fine motor skills is sensitive to the age at which the children were assessed. The preschool data collection of the ECLS-B was intended to assess children when the majority of the sample would be about 48 through 57 months of age. However, during the preschool round, children were assessed when they were as young as 44 months and as old as 65 months. Therefore, in this report, the first table on language, literacy, mathematics, color knowledge, and fine motor skills presents information by age at the time of assessment (table 2). For this table, age at assessment is divided into three categories : less than 48 months ; 48 through 57 months (roughly the target age for the assessment), and more than 57 months. Because age at assessment is not independent of certain child and family characteristics (certain groups of children may be older when assessed in a given wave)1, it is inappropriate to analyze the ECLS-B cognitive and fine motor information without addressing age at assessment (for more information on this issue please see the Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), Methodology Report for the Preschool Data Collection (2005–06), Volume I : Psychometrics [Najarian, Lennon, and Snow 2007]). Therefore, after a table presenting cognitive and fine motor data by overall age of assessment (table 2), a series of tables (tables 3 through 6) present information on the 75 percent of the children who were assessed in the target age range (48 through 57 months) by certain child and family characteristics. All comparisons made in the text were tested for statistical significance to ensure that the differences were larger than might be expected due to sampling variation. All differences reported are significant at the p<.05 level.

Appendix A provides technical documentation for the findings presented in this report, and general information about the study. Appendix B reports the standard errors for tables 1 through 7.

---

Les documents attachés à cet article :