DOI: 10.1126/science.1215343, 1118 (2012);335 Science

et al.Gary T. HenryScience and Mathematics TeachersThe Effects of Experience and Attrition for Novice High-School

This copy is for your personal, non-commercial use only.

clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others

  here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): May 16, 2012 www.sciencemag.org (this information is current as of

The following resources related to this article are available online at

http://www.sciencemag.org/content/335/6072/1118.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

http://www.sciencemag.org/content/suppl/2012/02/29/335.6072.1118.DC1.html can be found at: Supporting Online Material

http://www.sciencemag.org/content/335/6072/1118.full.html#ref-list-1, 2 of which can be accessed free:cites 17 articlesThis article

http://www.sciencemag.org/cgi/collection/educationEducation

subject collections:This article appears in the following

registered trademark of AAAS. is aScience2012 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on

May

16,

201

2w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

distribution continuous, we computed an “achievementrank” by combining the number of the stage reachedand the number of successful manipulations (seesupporting online material).

28. G. Csibra, G. Gergely, in Processes of Change in Brainand Cognitive Development, Y. Munakata, M. Johnson,Eds. (Oxford Univ. Press, Oxford, 2006), pp. 249–274.

29. D. Wood, J. S. Bruner, G. Ross, J. Child Psychol.Psychiatry 17, 89 (1976).

30. R. L. Kendal, J. R. Kendal, W. Hoppitt, K. N. Laland,PLoS ONE 4, e6541 (2009).

31. A. Whiten et al., Curr. Biol. 17, 1038 (2007).32. M. Dindo, B. Thierry, A. Whiten, Proc. R. Soc. B 275, 187

(2008).

33. M. Tomasello, M. Carpenter, J. Call, T. Behne, H. Moll,Behav. Brain Sci. 28, 675 (2005).

Acknowledgments: Supported by the CULTAPTATION project(European Commission contract FP6-2004-NESTPATH-043434),a European Research Council Advanced Grant (EVOCULTURE,232823) (K.N.L.), and a Royal Society Dorothy HodgkinFellowship (R.L.K.). Chimpanzees at M. D. Anderson CancerCenter were supported by NIH Cooperative AgreementRR-15090. We thank S. Pavonetti, A. Whiten, C. Caldwell,C. Tennie, M. Tomasello, C. van Schaik, the CCBC/EARGjournal club (Durham), and members of the Laland lab foradvice and/or useful comments on earlier drafts of ourmanuscript, and J. Zampol for drawing Fig. 1. All research was

approved by the ethics committee at the University ofSt. Andrews and the relevant committees at the researchfacilities and follows the relevant legislation in theUK, USA, and France. Data are available from the authorsupon request.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/335/6072/1114/DC1Materials and MethodsFigs. S1 and S2Tables S1 to S6

14 September 2011; accepted 23 January 201210.1126/science.1213969

The Effects of Experience andAttrition for Novice High-SchoolScience and Mathematics TeachersGary T. Henry,1* C. Kevin Fortner,2 Kevin C. Bastian3

Because of the current high proportion of novice high-school teachers, many students’ masteryof science and mathematics depends on the effectiveness of early-career teachers. In thisstudy, which used value-added models to analyze high-school teachers’ effectiveness in raisingtest scores on 1.05 million end-of-course exams, we found that the effectiveness of high-schoolscience and mathematics teachers increased substantially with experience but exhibiteddiminishing rates of return by their fourth year; that teachers of algebra 1, algebra 2, biology,and physical science who continued to teach for at least 5 years were more effective as noviceteachers than those who left the profession earlier; and that novice teachers of physics, chemistry,physical science, geometry, and biology exhibited steeper growth in effectiveness than didnovice non–science, technology, engineering, and mathematics teachers.

In the past two decades, the teacher labormarket has dramatically changed in responseto more employment opportunities for wom-

en, increased demand for teachers, and policiesopening new pathways into the profession (1).For instance, the modal value of experience forU.S. teachers dropped from 15 years in 1987–1988 to 1 year in 2007–2008 (2). Additional-ly, turnover for beginning teachers is high: Afterjust 5 years, nearly 50% of all novice teachershave exited the profession (3). This churn of be-ginning teachers in and out of public schoolsresults in more students, particularly poor and/orethnic minority students, being taught by noviceteachers (4, 5), and that in turn leads to reducedstudent achievement (6–11). Investigating theconsequences of these teacher labor marketconditions for high-school students’ science andmathematics achievement—specifically, the ef-fects of experience and attrition among novice

teachers—will shed light on challenges facingeducation and career preparation in science andmathematics.

We quantified the growth in effectiveness ofhigh-school science and mathematics teachersand the effects of those teachers who exit publicschool classrooms. We analyzed effectivenessusing scores on standardized tests given to high-school students in three mathematics courses andfour science courses. We define teachers’ effec-tiveness in terms of the increases in their students’test scores, adjusted for the prior achievement ofthe individual students and for other student,classroom, and school covariates. Prior researchshows that the average effectiveness of noviceteachers increases during their first 3 years andflattens thereafter (12–14); and that after differ-ences in effectiveness that are attributable to ex-perience are removed, less effective teachers aremore likely to exit the profession (10, 15, 16).We extended this research to investigate teach-er effectiveness in specific high-school coursesand addressed three questions: (i) To what ex-tent do novice high-school science and math-ematics teachers become more effective withadditional experience? (ii) Are novice high-schoolscience and mathematics teachers who exit pub-lic schools more or less effective than those whostay? [There is little published research on whereteachers who exit public schools are subsequent-ly employed, but it suggests that leaving for

higher-paying jobs in the private sector is rel-atively rare (17).] (iii) Do the rates of change ineffectiveness for high-school science and math-ematics teachers vary by course?

Student outcomes are related to variationsin many school-related factors, including leader-ship; an orderly environment; high student expec-tations; a focus on student outcomes; a positiveschool culture; parental involvement; and, mostclosely, to teachers’ effectiveness (18, 19). Teach-ers are the most important school-related vari-able explaining variation in student achievement(12, 18, 20), and teacher experience positivelyaffects student performance (6–11). Most gainsin effectiveness occur in the first 3 years ofteachers’ careers, with minimal increases there-after (13).

It is likely that through teaching experience,trial and error, professional development, men-toring, and/or collaboration with fellow educa-tors, teachers learn rapidly during their first fewyears on the job. However, some of the averageincreases in effectiveness that have been attrib-uted to experience may be a statistical artifactcaused by the exit of less effective early-careerteachers, thereby overstating the year-to-year dif-ferences in the statistics related to experience.Recent research supports this second explana-tion, finding that exiting teachers are less ef-fective than comparable teachers who remain inthe profession (10, 15, 16). Here we disentanglethe effects of teacher development from differ-ential attrition among high-school teachers ofscience and mathematics courses.

To assess the effectiveness and attrition ofnovice science and mathematics teachers, we de-veloped a data set from North Carolina contain-ing end-of-course test scores for seven scienceand mathematics courses—algebra 1, algebra 2,geometry, biology, chemistry, physical science,and physics—and three other courses—English1, U.S. history, and civics/economics—which wegrouped together for the purposes of our analysisand label non-STEM (science, technology, engi-neering, and mathematics) courses. Our studysample included all teachers in tested subjectswith less than 5 years of experience employedin any regular North Carolina public high schoolfrom 2005–2006 through 2009–2010. The mostcrucial feature of the data set is that students andteachers were linked on the basis of actual class-room rosters, which allowed us to match approx-

1Department of Public Policy and Education Policy Initiative atCarolina, University of North Carolina at Chapel Hill, AbernethyHall, Campus Box 3435, Chapel Hill, NC 27599–3435, USA.2Department of Educational Policy Studies, College of Edu-cation, Georgia State University, 30 Pryor Street, Atlanta, GA30303–3083, USA. 3Department of Public Policy and Edu-cation Policy Initiative at Carolina, University of North Carolinaat Chapel Hill, Abernethy Hall, Campus Box 3279, Chapel Hill,NC 27599–3435, USA.

*To whom correspondence should be adddressed. E-mail:gthenry@unc.edu

2 MARCH 2012 VOL 335 SCIENCE www.sciencemag.org1118

REPORTS

on

May

16,

201

2w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

imately 93% of high-school teachers to studentsin their courses; to construct student, teacher, andclassroom covariates; and to account for stu-dents who had multiple teachers for any course.In total, the sample included 1.05 million testscores, 624,842 unique students, and 7961 uniqueteachers with less than 5 years of experience.

The outcome variable analyzed was the stu-dents’ score on a standardized exam: the NorthCarolina End-of-Course Tests. These exams aredesigned to test students’ knowledge and skillsbased on the North Carolina Standard Course ofStudy objectives for each course, which followthe frameworks of the National Assessment ofEducational Progress and recommendations fromnational standard-setting groups (21, 22). Theitems on these tests covered a range of cognitiveskills, from remembering to analyzing and eval-uating; included four potential response options;and were designed to cover a range of difficultyfrom easy (25%) to medium (50%) to hard (25%).Multiple versions of each test with differentitems reduce teachers’ ability to predict the items,while providing equivalent scores across differ-ent versions. For example, there were 12 versionsof the chemistry test, each of which included 92unique items, meaning that a total of 1104 itemswere in use during the study period (21). To re-move any year-to-year differences in test scoresand to estimate effects in standard deviation units(SDUs), we standardized all tests within courseand year by centering observations on the meanand dividing by the standard deviation.

To assess the development and attrition ofnovice science and mathematics teachers, weestimated two sets of subject-specific models, inwhich focal development variables were fourdichotomous experience indicators coded 1 forteachers with the designated experience level (1,

2, 3, or 4 years, respectively) and 0 for teacherswith other experience levels (comparisons wereto teachers with 0 prior years of experience).The two sets of models included different focalvariables to measure the impact of teacher attri-tion and separate its effect from teacher devel-opment. Based on certified salary files suppliedby the North Carolina Department of Public In-struction, the focal variable in model 1, labeled“leaves within 5 years,” was coded 1 for teach-ers who leave North Carolina public schools be-fore beginning a sixth year of service and 0 forall others; the focal variable in model 2, labeled“last year,” was coded 1 for teachers in a givenyear who will not be paid as teachers in the fol-lowing school year and 0 for all others. In mod-el 1, the coefficient on “leaves within 5 years”provided an estimate of the overall average ef-fectiveness of exiting teachers, while netting outthe effects of attrition [see eq. 1 in the support-ing online material (SOM)]. In model 2, the co-efficient on the “last year” indicator estimatedthe average effectiveness of novice teachers intheir final year, while removing the effect ofteachers who will leave at the end of the year(which is commonly known as the marginal rateof return to experience). In addition, model 2included interaction terms between experienceand “last year” to allow the average effective-ness of departing teachers to vary based on theyear of exit (see eq. 2 in the SOM).

To estimate differences in the rates of changeof teachers’ effectiveness by course, we combinedall courses in a third model and substituted a con-tinuous measure of experience and its squaredterm into the value-added models. In model 3,we interacted both experience measures withcourse indicator variables to compare slopes andrates of change by subject (see eq. 3 in the SOM).

Results from this model compare the average re-turns to experience for novice science and math-ematics teachers to those of novice teachers innon-STEM courses. All model specifications usevalue-added models with student, classroom, andschool covariates, including students’ prior testscores (Table 1).

We find (Table 2) that teachers of all fourscience subjects experience gains of at least 0.11SDU (“2nd Year Teachers”) between their firstand second year of teaching, as compared to lessergains of 0.06 to 0.09 SDU for all three mathe-matics courses and non-STEM courses. Second-year physics teachers post the largest gains of0.38, with chemistry, physical science, and biol-ogy teachers posting gains of 0.17, 0.16, and0.12 SDU, respectively. For all teachers, returnsto experience diminish rapidly. The effectivenessof novice algebra 2, geometry, biology, and chem-istry teachers peaks in their fourth year, whereasalgebra 1, physics, and non-STEM teachers con-tinue to increase in effectiveness through theirfifth and final year that is included in this dataset. In the middle panels of Table 2, the coeffi-cients on the experience variables quantify themarginal rates of return to experience, separat-ing out the effects of teachers who leave thenext year, which are either larger or smallerthan the coefficients on experience in model 1,depending on the effectiveness of teachers whoexited.

Comparing average effectiveness, novice teach-ers who will remain teaching beyond 5 years aremore effective in four of seven science and math-ematics courses than teachers who will leavebefore completing 5 years (Table 2). In courseswhere differences are significant, the average dif-ference in effectiveness for teachers who exit with-in 5 years ranges from –0.102 for physical science

Table 1. Covariates used in analyses.

Student Covariates Classroom Covariates School Covariates

www.sciencemag.org SCIENCE VOL 335 2 MARCH 2012 1119

REPORTS

on

May

16,

201

2w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

to –0.036 for biology when compared to teach-ers with the same level of experience who stay.Exiting geometry, chemistry, and physics teach-ers are neither more nor less effective than thosewho stay.

In the middle panels of Table 2, we examinethe effectiveness of exiting teachers in their finalyear of teaching (“last year”). Teachers in fourscience and mathematics courses—algebra 1, al-gebra 2, geometry, and physical science—and thenon-STEMcourseswho leave the professionwithin5 years are less effective in their last year on the job.

For non-STEM courses, novice teachers ex-hibit a positive slope and diminishing returns toexperience (Table 3). Novice algebra 1 and alge-bra 2 teachers exhibit patterns similar to those ofnon-STEM teachers. Novice biology and physicalscience teachers exhibit higher rates of growth,but also rates of return to experience that dampenmore quickly than those of non-STEM teachers.Finally, novice geometry, chemistry, and physicsteachers show steeper slopes but diminishing re-turns to experience similar to those of non-STEMteachers.

Figure 1 shows the average effectiveness ofteachers in each of their first 5 years in the pro-fession, by subject, to allow effectiveness trajec-tories to be directly compared. On average, noviceteachers’ trajectories plateau or fall by their fifthyear. Furthermore, the order of the lines indicatesthe consequences of the loss of experienced sci-ence and mathematics teachers when they arereplaced by first-year teachers. These conse-

quences, in terms of student achievement, aremost sizable for physics and chemistry, rep-resenting losses of approximately 0.25 to 0.4SDU, on average; then physical science, geom-etry, and algebra 2, representing losses between0.1 and 0.2 SDU, on average; and finally, al-

gebra 1, biology, and non-STEM, representinglosses of less than 0.1 SDU, on average.

Overall, novice teachers of high-school sci-ence and mathematics exhibit significant returnsto experience early in their careers that diminishafter 4 years of teaching. This indicates that (i)

Table 2. Average effectiveness of novice teachers who stay and who exit by course (focal variables only). All results are in comparison to first-yearteachers who stay in North Carolina public schools for at least a second year. *, **, and *** indicate significance at the P < 0.05, P < 0.01, and P < 0.001levels, respectively.

Focal Variables Algebra 1 Algebra 2 Geometry Biology Chemistry Physical Science Physics Non-STEM

Table 3. Returns to experience for novice high-school teachers by course. *, **, and *** indicatesignificance at the P < 0.05, P < 0.01, and P < 0.001 levels, respectively.

Main Effects Coefficient

Interaction Effects Coefficient

2 MARCH 2012 VOL 335 SCIENCE www.sciencemag.org1120

REPORTS

on

May

16,

201

2w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

beginning teachers have a tremendous capacityfor improving quickly, and (ii) students of be-ginning teachers will not achieve at the samelevels as students with more experienced teach-ers. Our results show that increased reliance onnovice teachers leads to lower average teachereffectiveness. In light of current teacher labor mar-ket conditions, key questions include the follow-ing: (i) How to get more effective science andmathematics teachers into high-school classrooms?(ii) Can science and mathematics teachers bebetter prepared through more preservice experi-ence in classrooms, increased focus on instruc-tional skills, and deeper knowledge of content bythe time they begin teaching?

Different high-school subjects show differ-ent impacts of teacher turnover. For courses withsteeper effectiveness growth curves—physics,chemistry, and geometry—the loss of these ex-perienced teachers has the greatest consequencesfor student performance. For courses with lesssteep growth curves—algebra 1, algebra 2, biol-

ogy, and physical science—the loss of more ex-perienced teachers has less severe consequences.But both cases call for recruiting more able, mo-tivated, and committed teachers. Could theseteachers be better screened by evaluating theiracademic performance, persistence, ability to en-gage audiences, and projected commitment toteaching, specifically teaching STEM coursesto high-school students? Would incentives, suchas higher salaries, assignment to fewer courses peryear, or paid opportunities for research with uni-versity faculty during summers or semester leaves,help retain more of the experienced teachers?

The current churn of the teacher labor marketis working against higher student achievementin STEM courses. Although most current educa-tion policies that affect teachers do not distinguishbetween teachers of different types of courses,our results strongly suggest that distinctions be-tween STEM and non-STEM teachers, and evenamong STEM teachers who teach different courses,may be warranted.

References and Notes1. M. Bacolod, Rev. Econ. Stat. 89, 737 (2007).2. R. Ingersoll, L. Merrill, Educ. Leadership 67, 14 (2010).3. R. Ingersoll, T. Smith, Educ. Leadership 60, 30 (2003).4. H. Lankford, S. Loeb, J. Wyckoff, Educ. Eval. Policy Anal.

24, 37 (2002).5. C. Clotfelter, H. Ladd, J. Vigdor, Econ. Educ. Rev. 24, 377

(2005).6. D. Boyd, P. Grossman, H. Lankford, S. Loeb, J. Wyckoff,

Educ. Finance Policy 1, 176 (2006).7. C. Clotfelter, H. Ladd, J. Vigdor, Econ. Educ. Rev. 26, 673

(2007).8. C. Clotfelter, H. Ladd, J. Vigdor, J. Hum. Resour. 45, 655

(2010).9. D. Goldhaber, J. Hum. Resour. 42, 765 (2007).

10. G. T. Henry, K. C. Bastian, C. K. Fortner, Educ. Res. 40,271 (2011).

11. S. Rivkin, E. Hanushek, J. Kain, Econometrica 73, 417(2005).

12. J. Rockoff, Am. Econ. Rev. 94, 247 (2004).13. D. Staiger, J. Rockoff, J. Econ. Perspect. 24, 97 (2010).14. M. Chingos, P. Peterson, Econ. Educ. Rev. 30, 449

(2011).15. D. Goldhaber, B. Gross, D. Player, J. Policy Anal. Manage.

30, 57 (2011).16. J. Krieg, Econ. Educ. Rev. 25, 13 (2006).17. B. Scafidi, D. Sjoquist, T. Stinebrickner, Adv. Econ. Anal.

Policy 6, 1 (2006).18. C. Teddlie, D. Reynolds, The International Handbook of

School Effectiveness Research (Falmer Press, London,2000).

19. D. Reynolds, P. Sammons, B. De Fraine, T. Townsend,J. Van Damme, Educational Effectiveness Research(EER): A State of the Art Review, paper presentedat the annual meeting of the International Congressfor School Effectiveness and Improvement,Cyprus, 2011.

20. J. S. Coleman et al., Equality of Educational Opportunity(U.S. Government Printing Office, Washington, DC,1966).

21. North Carolina Department of Public Instruction, NorthCarolina Mathematics Tests. Ed. 3: Technical Report(Raleigh, NC); downloaded 29 December 2011 fromwww.dpi.state.nc.us/docs/accountability/reports/mathtechmanualdrafted2.pdf (2008).

22. North Carolina Department of Public Instruction,North Carolina Science Tests: Technical Report(Raleigh, NC); downloaded 29 December 2011 fromwww.dpi.state.nc.us/docs/accountability/testing/reports/eocsciencetechmanual.pdf (2009).

Acknowledgments: We recognize ongoing financial supportfor this research from the Teacher Quality Research Initiativeprovided by the University of North Carolina GeneralAdministration.

Supporting Online Materialwww.sciencemag.org/cgi/content/full/335/6072/1118/DC1SOM TextTables S1 and S2

17 October 2011; accepted 30 January 201210.1126/science.1215343

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1 2 3 4 5

Teac

her

s' A

vera

ge

Valu

e-A

dd

ed T

est

Sco

re G

ain

s

(Sta

nd

ard

Dev

iati

on

Un

its)

Teaching Year

Teacher Effectiveness Growth by Subject

Physics

Chemistry

Physical Science

Biology

Geometry

Algebra 2

Algebra 1

Non-STEM

Fig. 1. Average effectiveness of novice high-school teachers.

www.sciencemag.org SCIENCE VOL 335 2 MARCH 2012 1121

REPORTS

on

May

16,

201

2w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from