Do you want your child to have a certified teacher

Economics of Education Review 30 (2011) 65–78Contents lists available at ScienceDirectEconomics of EducationReviewj o u r n a l h o m e p a g e :w w w.e l s e v i e r.c o m /l o c a t e /e c o n e d u r evDo you want your child to have a certified teacher?Christian Andersson a ,Per Johansson b ,∗,Nina Waldenström ca Swedish National Audit Office (SNAO),Swedenb Institute for Labour Market Policy Evaluation and Department of Economics Uppsala University,Sweden cMinistry of Employment,Swedena r t i c l e i n f o Article history:Received 3October 2008Received in revised form 30June 2010Accepted 2July 2010JEL classification:I21Keywords:Teacher certification Teacher qualityStudent achievement Instrumental variablea b s t r a c tThis study examines how the teaching staff composition with respect to certification affects student achievement in compulsory Swedish schools.We apply an instrumental variable to estimate the effect of the share of non-certified teachers on student achievement (measured by grade point average,GPA ).We find statistically significant negative effects on the GPA .The effect is stronger for students with highly educated parents.A one percentage point increase in the share of non-certified teachers is expected to decrease student’s GPA by,on average,1.8standard deviations per year.This is a substantial effect considering the large differences in the shares of non-certified teachers across schools and municipalities.© 2010 Elsevier Ltd. All rights reserved.1.IntroductionTeacher quality is considered to be one of the most important factors affecting student achievement,and is for this reason central in the shaping of school and education policies.Despite the consensus that teachers are important,there are different opinions among researchers and poli-cymakers about what teacher quality exactly means.This paper deals with a frequently discussed aspect of teacher quality:certification,i.e.,the evidence of having completed a formal teacher training.Although certification entails slightly different features in different countries,there is a widespread agreement that it creates a lower bound on teacher quality,ruling out most “poor quality teachers”in schools.Two arguments are often raised to support this view.First,a formal teacher education improves an individual’s teaching specific human capital.∗Corresponding author.Tel.:+46184717086.E-mail addresses:christian.andersson@riksrevisionen.se (C.Andersson),per.johansson@ifau.uu.se (P.Johansson),nina.waldenstrom@employment.ministry.se (N.Waldenström).Second,a teacher certification is often thought to be an effective screening device to achieve higher teacher qual-ity.The argument for this is that only those who have the necessary qualities and therefore are able to complete the education would invest in such an education.In the light of these arguments students exposed to teachers without a formal certification are supposed to obtain worse results than students whose teachers are certified.However,it has also been argued that certification requirements may dis-courage potentially effective teachers from entering the profession,and in that way actually lowering the quality of the teaching staff (see e.g.Angrist &Guryan,2004;Ballou &Podgursky,1998).Existing literature on the effect of teacher certification does not present a uniform picture of these effects.The effectiveness of teacher certification has been discussed by Goldhaber and Brewer (2000),Goldhaber and Brewer (2001)and Darling-Hammond,Berry,and Thoreson (2001),among others,and the limitations of much of the stud-ies on teacher certification are summarized in Wayne and Youngs (2003).In a recent study Kane,Rockoff,and Staiger (2008)find that teacher certification on average has at most small impacts on Student’s test performances.However,0272-7757/$–see front matter © 2010 Elsevier Ltd. All rights reserved.doi:10.1016/j.econedurev.2010.07.00366 C.Andersson et al./Economics of Education Review30 (2011) 65–78among teachers with the same certification status,they find large and persistent differences in teacher effective-ness.Also Clotfelter,Ladd,and Vigdor(2007)find positive and statistical significant effects of certification on student achievement.Also the literature dealing with the impact of teachers’initial education in terms of subject-matter knowledge and pedagogical preparation is quite limited.The evidence is mixed and studies like Monk and King(1994)find pos-itive as well as negative,usually insignificant,effects of teachers’subject-matter education on students’results. Although conducted at a high level of aggregation and thus without accounting for differences across grade levels and subject areas,existing studies have found a rela-tively strong and consistently positive effect of pedagogical course work on teacher effectiveness;see for example Ashton and Crocker(1987)and Evertson,Hawley,and Zlotnik(1985).Kukla-Acevedo(2009)argues that it is likely that the effect of experience varies with teachers’qualities and abilities andfinds that math teachers’undergradu-ate performance is predictive of students’math attainment during the teachers’fifth year in classroom.However,hold-ing a master’s degree in education has been shown not to improve teachers’skills and effectiveness;see Hanushek (1986)and Rivkin,Hanushek,and Kain(2005).Ehrenberg and Brewer(1994)provide somewhat ambiguous evidence on whether school and teacher characteristics,such that for example teacher-student ratio,expenditure per student, teachers’degree and experience,matter.Hanushek(1997, 2002,2003)present literature surveys showing that only9 percent of all estimates of teacher education on student achievement are positive and at the same time statisti-cally significant,while the corresponding percentages for teacher experience,teacher–student ratio and expenditure per student are29,14and27respectively.Given the con-tradictory results offered by previous studies it is hard to conclude that there exist any strong link between teacher certification and teacher education,on the one hand,and teacher effectiveness,on the other.A common problem with many of these studies is that they rest on specifications subject to biases from deter-minants of student achievement other than certification status(or other teacher characteristics).At any point in time there are several factors that affect student achieve-ment and complicate the estimation of the effect of teacher certification(or other teacher characteristics).Potential non-random sorting of families among schools,of students among classes and of teachers among schools are exam-ples of factors that may lead to biased estimation results when using retrospectively collected data from registers or surveys;for example Goldhaber,Choi,and Cramer(2007)find that most disadvantaged districts,schools,and stu-dents in North Carolina are least likely to have access to teachers certified by National Board for Professional Teaching Standards,these teachers being recognized as highly qualified.Another problem is that the definition of teacher certification varies across regions(e.g.states in the U.S.)and if regional variation in teacher certifica-tion is used to estimate the effects of certification then this most likely would lead to biased results.In this study we consider these methodological problems when estimating average effects of having a non-certified teacher on student achievement.The analysis is based on information on teachers active in Swedish public compulsory schools and on students completing ninth grade during the time period1997/98 until2003/04.The advantages with the present study are that:(i)there is no variation in certification within Sweden; (ii)we use the within school variation in the share of non-certified teachers which allows us to avoid any problems with,time constant,non-random sorting;(iii)we have an instrument that exogenously shifts the distribution of non-certified teachers over time for each school.This allows us to control for the fact that the within school variation in the share of non-certified teachers most likely depend on the quality of the teachers at a given time as well as for time varying sorting.The instrument used is the interaction between the unemployment rate among certified teach-ers and a temporary and targeted government grant,the Wärnersson Grant(WG).The grant was instituted in2000by Sweden’s Minis-ter of Primary Education,Ingegerd Wärnersson,to increase personnel density in schools.Fortunately for evaluation purposes,the grant frame was based solely on demo-graphical aspects that could not be influenced by schools or municipalities,at least not in the short run.More-over the receivers of the grant,i.e.,municipalities,were allowed to use these additional resources only for employ-ment of school personnel(i.e.teachers,student assistants, school nurses).Thus,the introduction of the WG implied an exogenous demand shock for teachers,certified as well as non-certified.However,the effect of the WG on the teaching staff composition depends on the availability of certified teachers that could be employed.On the negative side is that our identification strategy does not allow us to identify the effect of becoming certifi-cated on student achievement since the identified effects captures both any potential enhancement in teacher spe-cific human capital from the certification but also screening effects and effects from experience.1Thus,these results cannot,unfortunately,give a definitively answer to the much debated question of whether imposing tighter stan-dards for employment of non-certified teachers,and even the prohibition of employing such teachers on indefinite contracts,leads to better student achievement.Wefind that a one percentage point increase in the share of non-certified teachers is expected to decrease the aver-age GPA by around1.8standard deviations per year in upper level of compulsory schooling(i.e.grade7–9).This is a sub-stantial effect considering the large differences in shares of non-certified teachers across schools and municipalities (see Section3and Fig.1).The effect is stronger for students with highly educated parents.The remaining of the paper is structured as follows.In Section2we discuss the Swedish education system and1The data provide a teacher experience measure from1999in our data. Experience and certification are however both most likely endogenous and in order to separate the two effects we would need two instru-ments:one that influences mainly the experience distribution and one that mainly influences the certification distribution,why the two effects cannot be separately identified in our data unfortunately.C.Andersson et al./Economics of Education Review30 (2011) 65–7867Fig.1.The distribution of the share of non-certified teachers.All Swedish municipalities.describe in more detail the special government grant,the Wärnersson grant.In Section3we discuss the teachers’labor market.The model and our instrumental variables approach are discussed in Section4.The data and variable definitions are presented in Section5.The results are pre-sented in Section6and the paper is concluded in Section7.2.The Swedish compulsory school and theWärnersson grantThe Swedish compulsory school is9years long.During thefirst6years students are usually taught by one and the same teacher irrespective of subject,while during the last3 years they are taught by specialist teachers in each subject. Children normally start compulsory school at the age of 7although all6-years-olds are offered to participate in a 1-year long pre-school class.The school-preparing class is part of the public school system,but is voluntarily.The reforming of the Swedish educational system in the 1990s implied the introduction of school choice as well as freedom to start up new schools.Independent schools can be set up freely,although applications must be approved by the National Agency for Education after consultation with the local authorities.Introduced in1992,the“freedom of school choice reform”replaced the“neighborhood princi-ple”meaning that parents now freely can choose the school they prefer,instead of being allocated to a school dictated by their place of residence.However,in practice the choice is made subject to the availability of slots(Björklund,Clark, Edin,Fredriksson,&Kreuger,2005).At the compulsory level students do not have much freedom to choose courses and/or teachers within schools.Ability tracks may occur, but the course goals are the same for all students(Lpo94).In1994afive-step norm-referenced grading system was replaced by a three-step criterion-referenced system related to the curriculum goals.The new system implies also that no grades are awarded for subjects before8th and9th grade.The grades are,basically,used as a selec-tion instrument for higher education.The criteria for how teachers may grade students are formulated by the Swedish National Agency for Education.In the latest curriculum (Lpo,94)it is stated that grades should reflect skills and knowledge in the subject in accordance with the goals stated in the course syllabi.It is also clearly stated that the grades should not reflect attention in the classroom,dili-gence,ambition,home works and work during lesson,as long as it is not a prerequisite for attaining the goals.In order to be eligible for upper-secondary school stu-dents need to pass in all the core subjects(Swedish,English and Mathematics).National tests are also performed in these subjects during the spring semester in9th grade (at time of grading).The tests are corrected at the school level and graded in accordance with national grading rules. Teachers are supposed not to correct their own students’exams,but they are allowed to do so.22.1.The special government grant–the WärnerssongrantTo counteract a trend of decreasing resources to schooling the Swedish government instituted a targeted government grant,the WG,to be distributed to munic-ipalities from the academic year2001/02until2006/07. The aim of the grant was to increase the personnel den-sity in pre-schools,9-year compulsory schools,special schools,after-school recreation centers and upper sec-ondary schools,in order to give children and youths increased chances of reaching their educational goals(SFS, 2001:36).3In terms of number of employees,this repre-sents a school personnel increase by almost20percent; from77,000to92,000.In total17.5billion Swedish kro-nor,SEK(about2.87billion USD)4were set aside to be distributed to the municipalities.Only municipalities could apply for and receive the WG,but they were allowed to dis-tribute it freely among public and independent schools.The government commissioned the Swedish National Agency for Education to administer,follow up,and evaluate the WG.In2001/02and2003/04all except two municipali-ties applied for and received the grant,while in2002/03all municipalities applied for and received it.5A grant frame was calculated yearly for each munici-pality.The frame was based on the number of inhabitants between6and18years-of-age living in the municipal-ity during the calendar year preceding the grant year,i.e., the grant frame for2001/02was based on the number of children and youths aged6–18during2000.The deci-2For a detailed description of the rules for grading and for an empirical analysis of the difference in grading and the standardized test see Lindahl (2007).3The grant was hence targeted and temporary.The fact that it was supposed to be temporary may have affected the hiring of teachers.It may e.g.be easier to hire non-certified teacher teachers since they are more often hired on temporary contracts.This does not,however,destroy our identification strategy that is based on differences in the supply of certified teachers.In2005,the government changed some of the orig-inal conditions of the grant.However,coming into force after the end of our period of study and because these changes could not be antici-pated during the period of study,they do not affect our analysis.Further information about how the conditions changed in2005is available at http://www.skolverket.se.4Mean exchange rate in2001was1USD for10.30SEK.The exchange rate in June2008was6.10SEK for1USD.We use the exchange rate in June2008throughout the paper.5The two municipalities that did not receive the grant in2001/02were Österåker and Umeå.Nacka and Sundbyberg did not apply in2003/04.68 C.Andersson et al./Economics of Education Review30 (2011) 65–78sion whether or not to approve new grants in subsequent years was based on the principle that the municipality had to increased the school personnel density compared to2000/01(the index year).6The National Agency for Edu-cation might decide to stop further payments or reclaim already disbursed payments if a grant receiving municipal-ity did not succeed in carrying out the measures that the grant was aimed for.The National Agency for Education, however,has not reclaimed any of the distributed funds so far.Municipalities had no knowledge about the conse-quences of having a high school personnel density in 2000/01,until after the end of the2000/01.The fact that the grant was aimed at increasing the number of teach-ers in schools was mentioned both in the government bill that introduced the WG and in the decree that the govern-ment issued later.How the funding would be distributed was,however,not specified.The“index year comparison”was described for thefirst time in the instructions from the National Agency for Education in2001.Hence,there can-not have been any tactical reasoning in the hiring process of schools and municipalities in2000/01.Almost one bil-lion SEK was distributed to municipalities in2001/02and in 2002/03municipalities were allocated almost two billion SEK.In2003/04the amount was three billion SEK.By divid-ing the amount allocated to a municipality with the total number of students the amount per student in2001/02, 2002/03and2003/04was628SEK(about103USD),1258 SEK(206USD)and1897SEK(311USD),respectively.The grant could be used to employ school personnel in pre-school,compulsory and upper secondary schools,but the vast majority of money was spent at the upper level of compulsory schools(7–9th grade)(Skolverket,2006,2007). Under the assumption that the grant was used entirely to employ teachers at the upper level of compulsory schools the amount per student in2001/02,2002/03and2003/04 was approximately2945SEK(about483USD),5666SEK (929USD)and8082SEK(1325USD),respectively.7This can be seen as a monetary upper limit of the potential increase in the number of teacher directed to the cohort of students studied in this paper.In the academic year of2003/04an average school had189students and an upper level compulsory school had on average350students.The implication of this was, under the assumption of an evenly distributed grant across compulsory schools,that a typical school received around 58,776USD.8However,if the grant was instead evenly distributed across upper level compulsory schools only, a typical school received around463,721USD.9The cost of employing a teacher on a full-time basis was about 390,000SEK(63,900USD).The grant was thus large enough to employ one(funds evenly distributed in compulsory schools)to seven(all funds distributed to upper level com-6The academic year2000/01was the index year until2005.For theremaining period,i.e.2005/06and2006/07the index year was changed to2003/04.7The cohort sizes in grades7–9in2001,2002and2003were339,528, 352,975and371,173,respectively.8189students×1,897SEK=358,533SEK or58,776USD.9350students×8,082SEK=2828,700SEK or463,721USD.pulsory schools)new teachers in an average sized school in2003/04.10It is important to note that all teachers teach all grades at the upper level,i.e.non-certified teachers are not more common to teach7th graders compared to certified teach-ers.Hence,the students graduating in2004have been exposed to the WG throughout the whole upper secondary compulsory schooling while those graduating in2002were only exposed to the extra funding in2001/02.3.The teacher labor market11In Sweden a person may become a certified teacher,i.e., receives a teacher certification,by attending and complet-ing a teacher education program.12Another possibility is to receive a minor or major in the teaching subject and sup-plement it with a minimum of1.5years of preparation in pedagogy,didactics and teaching practice.This alternative entrance to the teaching profession makes it possible for people in other professions to switch to the teaching pro-fession and,thus,gives non-certified teachers a chance to become certified.One consequence of the decentralization of the Swedish educational system in the early1990s was that local authorities became the employer of most of the school per-sonnel and thus have the responsibility for the recruiting process.13However,although the municipalities and the head of the independent schools are the formal employ-ers of teachers,the decisions on recruiting,selecting and employing teachers are made at the school level.Teachers may be employed on three different types of contracts:indefinite term contracts,probationary period employment andfixed-term contracts.Individuals employed under an indefinite term contract may only be dismissed on one of the grounds mentioned in the Employment Security Act,such as for example lack of work.Probationary period employment contracts are mainly intended for newly cer-tified teachers.These contracts normally lead to indefinite term employments.Teachers who do not have a teach-ing degree are often employed on afixed-term contract (normally maximum1year).According to the Education Act,employers must only employ teachers“whose education meets the main teacher requirements associated with the position”.This usually means a Swedish teacher certification,a teacher certifica-tion from any EU or Nordic country or a corresponding education approved by the National Agency for Higher Education.However,if it is not possible tofind a certi-fied teacher,local authorities may employ individuals with other educations onfixed-term contracts.Individuals with-10Without adjustments for age and experience the earnings for non-certified teachers are on average30percent lower than the earnings of certified teachers(Andersson and Waldenström,2007).Thus,if the grant was used to employ only non-certified teachers the inflow of teachers could have been about30percent higher.11See Andersson and Waldenström(2007)for a thorough description of the Swedish teacher labor market.12Teacher education programs are provided at24of Sweden’s over40 universities and university colleges(MESS,2003).13Except for the municipalities also independent schools may employ school personnel.C.Andersson et al./Economics of Education Review30 (2011) 65–7869out teaching degrees may be employed on indefinite term contracts only“if there are no applicants with a teaching degree,special reasons exist and the applicant has equiva-lent qualifications for the teaching associated with the post and appears to be suited for the task”.In comparison with the previousfixed pay scheme the decentralization of the school system allowed for individu-ally negotiated components in teacher contracts.14The pay agreements specify a minimum wage level after1year’s employment,while the actual levels are determined by the school management.Relative to workers’wages,teacher wages have declined precipitously since the beginning of the1940s(Björklund et al.,2005).Since1945the decline in the relative wage amounts to almost50percent.The same phenomena have been seen in a number of countries,but the decline in Sweden has been particularly steep.Teachers in Sweden today are paid less than their Nordic and OECD counterparts(OECD,2002).Statistics from the Swedish National Agency for Educa-tion show that the teacher density in public compulsory schools,measured as full time equivalent teachers per 100students,decreased from7.9in1995to7.6in2000. This development was,however,interrupted in2001when the teacher density started to increase,reaching a level of8.1teachers per100students in2004(see Table A1 for the1998–2004development in our data).The num-ber of teachers increased between1996/97and2003/04 by around10,000individuals.The distribution of the share of non-certified teachers in public compulsory Swedish schools at the municipality level between1996/96and 2003/04can be seen in Fig.1.As Fig.1shows,the median share of non-certified teach-ers in public schools increased during most of the years included in the study.We can,however,see a slowdown of this development towards the end of the period.More precisely,the median share of non-certified teachers in compulsory schools increased from6.8percent in1996to 17.8percent in2003/04and remained relatively constant at this level during the last year of study.We can also notice that the dispersion across municipalities has increased over years.The gradual increase of the number/share of non-certified teachers over time may have several causes as for example,increased wage differences between certified and non-certified teachers,usually in combination with tight budget constraints,as well as reduced quality of the cer-tified teachers,stemming either from a reduced quality in the teacher education or a reduced supply of quali-fied individuals.The supply of certified teachers is most likely a function of the attractiveness of the teaching pro-fession which in turn depends on factors such as relative wages,alternative opportunities,merit-based incentives, working conditions,status,and education(certification) costs.The increased dispersion in teacher staff composition between Swedish municipalities may partly be explained by differences in the municipalities’economic and social situation,as well as other municipality and school spe-cific factors such as political governance and geographical 14MESS(2003).location.General economic distress and concomitant social problems within a municipality may scare away certified teachers who usually have other labor market options.A small budget allocated to teacher employment in combi-nation with relatively inexpensive non-certified teachers may lead to an intensified use of non-certified teachers.Andersson and Waldenström(2007)show that the earning difference between certified and not-certified teachers are small and also decreasing over time.Hence the increase in the employment of non-certified teacher is most likely not driven by municipalities’financial incen-tives.Instead,some of the cross sectional variation seems to be explained by a limited supply of certified teachers.When certified teachers are not available schools rely on employ-ing non-certified teachers in order tofill vacant positions.4.Model and econometric frameworkThe aim of the present paper is to estimate the effect of teacher staff quality,measured as the percentage of non-certified teachers at the school,on student achievement. The following simple framework defines the parameter we are estimating.The basic idea is that student achievement,y,depends on the(q)uality of the teacher,henceˇ>0in y=ˇ0+ˇq+Á. We have information on grades for a student at a(s)chool at time t.Thus,if quality was observed and the teachers were randomly assigned at the school level we could estimateˇfrom the following regression model:y ist=ˇ0+ˇq st+Áist,(1) where i indexes student.There are two types of teachers:certified,C,and non-certified,NC.Let q Cstand q NC st be the average quality among certified teachers and non-certified teachers at school s in year t,respectively.Average teacher quality can then be written as:q st=q C st+p NC st×(q NC st−q C st),(2) where p NC st is the share of non-certified teachers at the school.Hence if the average quality of certified teachers is higher than the average quality of non-certified teachers it follows directly from Eq.(2)that teacher quality is decreas-ing in the share of non-certified teachers.Substituting(2) into(1)we get:y ist=ˇ0+ˇp NC st×(q NC st−q C st)+ˇq C st+Áist(3) Letı(ı<0)be the average quality difference between the non-certified and certified teachers,that is E(q NC st|q C st)=ı+q C st,then q NC st−q C st=ı+ωst.Hence,ωst is the residual difference in quality between the non-certified and certi-fied teachers at time period t,where hence E(ωst|q C st)=0. We can now write Eq.(3)as:y ist=ˇ0+ˇ1p NC st+εist(4) whereˇ1=ıˇandεist=ˇq C st+ˇp NC stωst+Áist.Here,sinceıis assumed to be negative andˇis assumed to be positive,ˇ1is expected to be negative.Since p NC st is most likely endogenous(i.e.when q C st is low and/orωst is high p NC st is most likely high)an ordi-nary least squares(OLS)estimator ofˇ1would be biased.。