UMI Number: 3102286 UML UMI Microform 3102286 Copyright 2003 by ProQuest Information and Learning Company
All rights reserved This microform edition is protected against
unauthorized copying under Title 17, United States Code
ProQuest Information and Learning Company
300 North Zeeb Road
P.O Box 1346
Trang 31996 and began graduate studies in Economics He received the Banco Central de Reserva del Peru Graduate Studies Fellowship between 1996 and 1998, and a Graduate
Studies Fellowship from the University of Rochester between 1998 and 2000 He pursued
Trang 4iti
Abstract
Three different aspects of Economics of Education are analyzed Chapter One focuses on the international eviderice on resources and education outcomes Chapter Two analyzes education preduction functions, controlling explicitly for teachers’
characteristics, and presents an evaluation of the Teach for America Program Chapter
Three covers the determinants of teachers’ salaries, assessing the impacts that varying
class sizes have on them, and how certification credentials are valued
in Chapter One, Economics of Education is faced from the international
perspective The data from the Third International and Science Study provides a way of comparing performance in different schooling systems The results from analyses of educational production functions within a range of developed and developing countries show general problems with the efficiency of resource usage similar to those found
previously in the United States These effects do not appear to be dictated by variations related to income level of the country or level of resources in the schools Neither do they appear to be determined by school policies that involve compensatory application of
resources The conventional view that school resources are relatively more important in
poor countries also fails to be supported
Trang 5positive effect of the TFA, among other results
In Chapter Three, the determinants of teachers’ salaries are assessed The
framework for the analysis is hedonic regression Instrumental variables are employed.to
correct for possible sources of biases Data for teachers and schools coming from the National Center for Education and Statistics is employed in the analysis The results show
Trang 6Introduction Chapter | Chapter 2 Chapter 3 Table of Contents "_— 1 Education production functions and the international evidence : 7 IHoc co ri ii atiidẢđdƠỔƠỔƠỔ 7 I€¡cii6o 1n - 8 1.3.Data and methodoÌOBV cuc HH no TH nh vn hy ky vu 16 1.3.1 Education production Đimcflons - 20 non c4 edd
1.4.1 Dimishing TetUTDS ch nh rên 26
1.4.2 Selection an compensafOry DỌÌICI€S cà 31 1.4.3 Family and sChOỌÌS HH HQ nh khien 38 I59o vn a.- 4 49
Assessing teacher effects with.an evaluation of the Teach for
America Program 0.0 cccccesceceeeeeceeeccececeeeeeteteneceeateseesenerses 52
2.1 Introduction cccceee cece eeeeeeeeeecedeeeeesseeneeaeeaneaeanstseen ss 52 2.2 Review of the literature 0.0.0 c cece eee ecsseceeeeeaeeseenenes 56 2.2.1 Experimental approach se se 357 2.2.2 Non-experimental evaluafion -.-.«-.cc se cvy 59 "SN '(( 59 2.2.4 Teacher charact€riSHCS cu nh, 61 2.2.5 Teacher certIfiCaLIOH ch chớp 62 2.2.6 Teach for America Program co sec se 64 2.3 Methodology con HH ng ng nà 65 “ "m›.` 5 3 72 ' ` 75 2.5.1 Fixed effects resulfs 8Ï 2.5.2 TPA evaluation " 83 2.5.3 Teacher cducation cv vn vs 87 2.5.4 Teacher fixed effects and direct measures of - f€aCheT QUẠÏẨY cu TS TH nh vn ve 88 2.5.5 Teacher quality and mobility patterns 89 "“"“® c.'ứA 93 Assessing the impact of class sizes and teacher requirements 18:0: >0/-. Ea 95 3.1 ÍntroduCHOn cuc HH VẤ HH ki Đo nh nề 95 3.2 — Previous research cu vn Kế HH 98 3.3 it 201020009 nHH‹ưd(4 103
3.3.1 Empirical SỈTAf€BV cv Thy 103
3.3.2 ° ieee ccececneceeceee sess ease sceeeeeeeaseeeneeseoes 107 3.4 Results 0.0 eee cece cece cece eset scene ne eee ees ceeneneeteneneeenens 111
Trang 8Table Table 1.1 Table 1.2 Table 1.3 Table 1.4 Table 1.5 Table 1.6 Table 1.7 Table 1.8 Table 1.9 Table 1.10 Table 1.11 Table 2.1 Table 2.2 Table 2.3 Table 2.4 Table 2.5 Table 3.1 Table 3.2 Table 3.3 Table 3.4 Table 3.5 vii List of Tables Title
Alternative estimates of the impact of resourcen on international
math and science performance across COUTTICS cuc seo [5
Distribution of estimated production function parameter across
Countries and AGe (2113 ng rr44Ầ 23
Distribution of Teacher qualification and Estimated Effect on
Student QufCOIm€S - HT nh Hi Tà ngà kg Tay 25
Distribution of estimated family background parameters across
countries and age STOPUS 0.0 cece cece ec ec eee eeeeeeneeeasaeeeseaeeninaees 30
Sign and statistical significance of alternative estimates of class
effects allowing for compensatory placement, age 9 cohort 36
Sign and statistical significance of alternative estimates of class
effects allowing for compensatory placement, age 13 cohort 37 Additions to explanatory power of school inputs, age 9 cohort 40 Addtition to explanatory power of school inputs, age 13 cohort 4I Estimated Class Size Effects Compared to Impact of
Disadvantaged Background, age 9 cohOoF co nen 44
Estimated Class Size Effects Compared to Impact of
Disadvantaged Background, age 13 cohorf cv ke se 45 Differences in family background slopes (age 13-age 9)
for pooled sampÌ@$ - TH HH K nh nh nà kh nà 48 Education production functions (Total Sample) 77 Education production functions (Teachers with less than 2 years
D0824 ii) 10 nh Ữ5Ả 80 Đistribution of Teacher Fixed efects con nen se 82 Average Teacher Fixed effects according to mobility patterns 90
\/13191)1A 81319 + —|-NaaadđdđdaaiidỎỐẮẦẢẮỒ 92
Estimated effect of on teacher salaries of selected variables 113 Estimated effect of class size on salaries by r€glOPS 116
Estimated effect of class size by metropolitan area
Trang 9Figure 1.5 Figure 1.6 Figure 2.1
Figure 2.2
Class size coefficient and Class Size (9 years oÌd) 32
Class size coefficient and Class Size (13 years oÌd) 33
Distributton ofteacher fixed effects (Total Sample) 85
Trang 11Appendix B _ Appendix Table Bl Appendix Table B2 Appendix Table B3 Appendix Table B4 Appendix C] Appendix Table C1 Appendix Table C2 Appendix C2 Appendix Table C4
Description Data from Houston School District CHSD) 133 Sample characteristics HSD oo cee eeeec eee ee eee ee ens ¬- 135
Education production functions: Adding last year score
on right-hand side (Total Sample) 136
Education production functions: Adding last year score
on right-hand side (Teachers with less than 2 years
of experience) ¬ ceed et eR eE ESA SEE ENR eS eel 37 Education production functions: Allowing non linearities in Class sizes (Total G003 mm 138
Description Data from SASS 2n ru uk 139 Description of sample eck cee eee eens teseauecebesecuneenes 141
Basic descriptive statisticts by region and So
Trang 13success and ultimately for the growth of national economies comes from its
relationship with individual productivity and earnings The central focus is on how systematic policy actions of governments affect the performance of students, Most of
the research attention has actually gone to the relevance of resources as a policy tool
Chapter One focuses on the resources allocated to schools in an internabonal
context Previous research for the United States, summarized by Hanushek (1986,
1997), shows that resources devoted to schools are not consistently related to student
outcomes However, the question arises is does this result change once we allow for
different institutional settings and different levels of resource allocation to schools
The literature on international evidence is sparce and, in addition, studies are hard to
Trang 15countries This analysis has been made possible by recent international testing and
data collection, which provides scores on common examinations across countries ‘Building upon the testing and surveys of the Third International Mathematics and Science Study (TIMSS), we consider specifically how families and schools contribute
te within country variations in student performance Among other school inputs, we
assess the impact of class sizes and teacher training on student outcomes Special
attention is devoted to the possible bias introduced by compensation policies within
schools We then go beyond the standard production function estimation to assess whether schooling systems in different countries work to narrow or widen
- performance differences, giving that students have similar opportunities
Trang 17previous literature Biases usually come from a number of unobservable
characteristics in students, teachers and schools
‘The uncertainty derived from the unobservable characteristics of students,
teachers and schools makes evaluation m education a difficult task Households choose te locate themselves in different sections of metropolitan areas frequently due
to a particular school or school district (Ticbout (1956)) Within schools we can expect that students will be sorted into classrooms and programs, based upon the
teacher quality and classroom characteristics Unobservable teacher characteristics include effort and innate ability, which are not directly observable by the researcher
or the policy maker, but that may have a direct impact on student outcomes
Nevertheless, it is important to find measures of teacher quality that will allow us to compare teachers, and make evaluations of the different teacher training programs or the characteristics of teachers
In this study we use a database which allows us to characterize such effects
Trang 19This data allows us to follow students, teachers and schools on a direct matching We have extensive information At the same time, the data allows estimation using | - teacher and school specific effects, controlling for possible sources of non-random
sorting At the same time, the data allows us to isolate specific teacher effects, construct their distribution, and compare the distribution between different sets of
teachers
A special emphasis is placed on the Teach For America (TFA) program TFA allows college graduates to teach without fulfilling the traditional requirements These teachers usually teach in inner city schools This program has been subject to
considerable criticism, given that TFA teachers have very little experience once they start teaching
The results show that resources measured as class size are not related to
student outcomes, despite the fact that the data for this study comes from an inner city school district The variance in teacher fixed effects, a measure of teacher quality
which comes directly from student outcomes, was big Direct measures of teacher
quality, such as teachers results on certification tests, showed a positive but small
Trang 21we focus on the impact of class sizes The analysis is based on the Theory of Wage Differentials Teacher salaries, at least partially, will represent the market equilibrium | value of inputs, such as class sizes
The effects of changes in class size have been the focus of recent intense policy discussions Almost all of ‘the attention has been related | to student
performance The related research has tried to quantify the effects on student
outcomes and to interpret them from a policy perspective While controversy over the magnitude of such effects and the costs of change continues, the range of differences
is narrowing, and the options are becoming clearer The existing research, however,
has generally neglected the overall effects of smaller class sizes on the teachers _ When considered, this literature tends to concentrate on teacher behaviors —
classroom management, time on task, and the like ~ but these are largely details of the
process by which achievement gains are realized Most importantiy, the analysis
ignores the interactions of class size with overall teacher satisfaction and its -
Trang 23feedback effects could significantly reduce the costs of lowering class sizes On the other hand, if the policies simply show a systematic but small quantitative relationship, the feedback through employment factors would not enter significantly | inte the policy debate
This research project focuses on the determinants of teacher salaries Roughly
speaking, we find that increasing the class size by one student increases teacher salary
up to around 1.0 percent This effect is found to be statistically sỉ gnificant in some but not all of the empirical specifications and for some but not all points on the salary schedule At the same time, other factors have a stronger influence on district salaries For example, we find that teachers facing a higher number of minorities
Trang 25The empirical literature on human capital has concentrated on the private returns to the quantity of schooling obtained by individuals Standard Mincer formulation shows how investment can be translated into observed differences across individuals (Mincer (1970, 1974)) The relationship between years of schooling and earnings has been studied deeply and in general provides evidence of the importance of schooling
However, this simple approach does not take into consideration the quality differences that appear in schooling The variable “years of schooling” does not
reflect the quality of schooling that an individual had during his life Test scores have
been introduced to allow for the measurement of quality differences in schooling, The
standard approach, studied by the literature, is that different quality of outputs in the
education is related to different inputs
Trang 27
Given that the quality of schooling seems to be important, it is necessary to investigate which factors affect student quality But at the same time, it is also
necessary to i nvestigate how these factors respond to specific country characteristics in this study we are able to match data fram different countries and investigate how
the student outcomes respond to changes in institutional settings and wealth
(measured by GDP) In addition, we present initial investigations on bow public education systems achieve their role of giving people equal opportunities (ie., reduce the inbound jnequality)
1.2 General Context
An important issue throughout the discussion of school quality has been the
relationship between outcome measures of quality (earnings, test scores, and the like)
and the resources devoted to schools This issue is twofold First, because direct
Trang 29resources as a policy tool On that score the US evidence has been reasonably clear
The resources devoted to schools are not closely or consistently related to student
si outcomes, While there has been some controversy over this analysis, only a minority of studies have found significant and positive relationships between resources and _ performance.“
The general structure of the production function estimation, designed to pinpoint causality has focused on a model such as:
Eq 1.1
0, = SF Py Sys A) + 2,
Where Oj, is the performance of student i at time f, Fy is a vector of family inputs —
- cumulative for student ? at time t, Py denotes cumulative peer inputs for student fat
time ¢, Sz denotes cumulative school inputs for student é at time #, A; denotes innate
ability of student i, and vy denotes a stochastic term
The proxy question ~i.e whether resources are an indicator of quality differences
Trang 31This stronger relationship couid simply reflect a positive relationship between
“resources and other factors which might arise if wealthier parents ‘on average both
contribute more directly to performance and put more resources into their schools in
the growth setting, there is no direct evidence of the proxy relationship Resources
tend to give incorrect signs and to be poor proxies (Hanushek and Kimko (2000))
Empirical work on quality in an international setting has, however, been even rarer than in the United States Few international data sets have information on outcomes
and resources, although- when available- there seems to be slightly stronger relationships between resources and outcomes in the production function setting of
equation 1.1 (Heyneman and Loxley (1983), Hanushek (1995), Vignoles et al (20060)) When this data is available, it has been difficult to summarize because the
data sets tend to be very specialized and to be very different across studies In
addition, little is known about the value of proxy relationship across countries
? Por discussions of the basic result from estimation of the effects of resources, see Hanushek (1986, 1997) For a discussion of the controversies, see Hedgeds, Laine and Greenwald (1994), Greenwald,
Hedges and Laine {1996} and Hanushek (1996) , ,
Trang 33Such comparative analysis has been largely precluded in the past, although some
work does exist Perhaps the largest and most influential study is Heyneman and
Loxley (1983) They analyze data from the second International Mathematics and Science Study along with other country specific tests Their primary conclusion is that resource vatiations appear to be more closely related to student performance in developing countries than in the developed ones
To put the resource issue into perspective, it is perhaps most useful to begin with
aggregate differences across countries The comparison of cognitive achievement
across countries capitalizes on seven voluntary international tests of student
achievement in mathematics and science that were conducted over the past three
decades Two were conducted by the International Association for the Evaluation of Educational Achievement (IEA) and five were administered by the International
Assessment of Educational Progress (LAEP)*, The IEA, since its establishment in 1959, has had a long and unique role in developing comparative education research for almost all aspects of primary and secondary education On the other hand, the TAEP, started in 1988, builds on the statistical techniques and procedures developed
Trang 35
The concentration on mathematics and science corresponds to the theoretical
‘emphasis on the importance of research and development (R&D) activities as the source of growth (e.g Romer (1990)) Able students with good understanding of
mathematics and science form a pool of future engineers and scientists At least for the United States, Bishop (1992) provides separate confirmation of the importance of mathematics in determining individual productivity and income Additionally, while
some test information exists for other subjects, it cannot be compared readily with the
mathematics and science scores and therefore is not used here
An overview of the testing results is best seen from Figure 1.1 This figure shows countries results on each of the math and science tests from the early 1960s through the Third Internationa! Mathematics and Science Study in 1995, In the figure, all of
the scores in each year are normalized to a world mean of 50 (see Hanushek and Kim (1995); Hanushek and Kimko (2060)) While a different array of countries has
participated in the test, some sense of the overall pattern can be seen from the figure
There is an aggregate tendency for East Asian countries to perform better and for developing countries to score worse Nonetheless, the performance of individual
Trang 37New Zealand Japan Singapore Hungary Korea Austratia : Japan Germany
Netherlands ees hong
: United Kingdom 4 China Korea Slovak ‘ne Austria - ang Ko! ỉ
Sweden Hong Kong _ Singapore Swikerland Czech Reg iblic aed Whee Netherlands : Taiwan treland
e , li am France iriand Hungary an , - Syatzertand MHGSYY |
2 UNHTEDSTATES Wares Kems "Sou UNITED STATES
© Netherlands ftaly
3 Belgium: France ' New Zealand
~~ Sen đen Russian Federat
® United Kingd teraci - BH
S eae Poland” Unisd KingdowPhĐs3 Kingdom đảng
8 50 “France New Zealand fet THUY Qgếm THAMÙI ANH
2 7] Beigium Canada a UNITED STATES Slovenia eng
& lialy lsrael Australia tretand ireland Rowpgnia © Thailand UNITED STATES
-~ _ Hungary Swed en : Cyprus
® nhà gis
5 care: Portugal UNITED STATES canada
Trang 39through 1990, as demonstrated by Hanushek and Kimko (2000) Specially, there is no
pattern to scores and resources, at least after controlling for differences in families
over time Table 1.1 reproduces the estimated resource effects on achievement for a
sample including ail country-years of test data that also had complete input data Of the three separate resource ‘measures ~expenditure per pupil, proportion of GDP
devoted to public education, and pupil-teacher ratio in primary schools, all three go in the wrong direction
There are good reasons to be cautious about these results, however, since the
simplified production estimates do not measure any organizational or structural differences in the school systems of the various countries These factors — if important and if correlated with resources- wil bias the estimated coefficients Since these
estimates might be biased because they can mask substantiai within-country variation,
we go on to consider variations in scores for individual countries Similar estimation
has been conducted: for the TIMSS international testing by Wobmann (2000) He
combines the microleve! TIMSS information with data about characteristics of the
overall system —centralization, private school options, unionization, and the like- and